FM / LICENSES

/opt/td-agent/LICENSES

Name	Size	Mode	Actions
cacerts-index.815ca599c9df.txt	16726	0644	edit dl rm
config_guess-config.guess	44016	0644	edit dl rm
config_guess-config.sub	36428	0644	edit dl rm
libedit-COPYING	1615	0644	edit dl rm
libffi-LICENSE	1132	0644	edit dl rm
libiconv-COPYING.LIB	25291	0644	edit dl rm
liblzma-COPYING	2774	0644	edit dl rm
libxml2-COPYING	1289	0644	edit dl rm
libxslt-COPYING	2968	0644	edit dl rm
libyaml-LICENSE	1058	0644	edit dl rm
ncurses-ncurses-license.html	147613	0644	edit dl rm
ncurses-ncurses.faq.html	307933	0644	edit dl rm
nokogiri-LICENSE.md	1198	0644	edit dl rm
openssl-LICENSE	6281	0644	edit dl rm
postgresql-COPYRIGHT	1192	0644	edit dl rm
ruby-BSDL	1282	0644	edit dl rm
ruby-COPYING	2502	0644	edit dl rm
ruby-LEGAL	29228	0644	edit dl rm
zlib-README	5187	0644	edit dl rm

Edit: /opt/td-agent/LICENSES/ncurses-ncurses.faq.html (307933B)

<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01//EN">

<title>NCURSES &mdash; Frequently Asked Questions (FAQ)</title>
 <link rel="author" href="mailto:dickey@invisible-island.net">
 <meta http-equiv="Content-Type" content=
 "text/html; charset=us-ascii">
 <meta name="keywords" content=
 "ncurses, curses, terminfo, termcap">
 <meta name="description" content=
 "This FAQ gives some background and discussion for frequently encountered problems with the ncurses library, the terminal database and applications.">
 <link rel="SHORTCUT ICON" href="/img/icons/vile.ico" type=
 "image/x-icon">
 <link rel="stylesheet" href="/css/simplestyle.css" type=
 "text/css">
 <link rel="stylesheet" href="/css/inline-code.css" type=
 "text/css">
 <style type="text/css">
 @import "/css/simplenavXX.css" all;
 </style>
 <meta name="viewport" content=
 "width=device-width, initial-scale=1">
</head>

<a href="/">http://invisible-island.net/</a><a href=
 "/ncurses/">ncurses/</a> 
 Thomas E. Dickey 
 $Date: 2020/06/20 13:48:36 $

<hr>

<a href="/ncurses/ncurses.faq.html">Here</a> is the latest
 version of this file.

<h1 class="no-header">NCURSES &mdash; Frequently Asked Questions
 (FAQ)</h1>

<li><a href="#what_is_it">What is NCURSES?</a></li>

<li><a href="#who_did_it">Who wrote NCURSES?</a></li>

<li class="nav-break">
 <a href="#who_owns_it">How can it be distributed?</a>

<ul>
 <li><a href="#who_claims_it">Is it Free Software or Open
 Source?</a></li>

<li><a href="#tack_copying">What about the tack
 program?</a></li>

<li><a href="#terminfo_copying">What about the terminfo
 database?</a></li>
 </ul>
 </li>

<ul>
 <li><a href="#growing_features">Growing features</a></li>

<li><a href="#uses_of_library">Types of library users</a></li>

<li><a href="#related_to_vi">Relationship with vi</a></li>

<li><a href="#not_library_users">Applications miscited as
 library users</a></li>
 </ul>
 </li>

<li class="nav-break"><a href="#what_platforms">What
 platforms does it run on?</a></li>

<li><a href="#latest_version">What is the latest version?</a></li>

<li><a href="#other_versions">What versions are "stable"?</a></li>

<li><a href="#related_programs">What other programs are
 there?</a></li>

<li><a href="#known_problems">Known/Frequent problems</a></li>

<li><a href="#report_bugs">How do I report bugs?</a></li>

<li><a href="#posix_compliant">POSIX Compliant?</a></li>

<li><a href="#y2k_compliant">Y2K Compliant?</a></li>

<li><a href="#terminology">Terminology</a></li>

<li><a href="#additional_reading">Additional Reading</a></li>

<li><a href="#copyright_text">Copyright</a></li>
 </ul>
 </div>

<h2 id="what_is_it-id"><a name="what_is_it" id="what_is_it">What
 is NCURSES?</a></h2>

Ncurses (new curses, pronounced "enn-curses") started
 as a freely distributable "clone" of System V Release 4.0 (SVr4)
 curses. It has outgrown the "clone" description, and now contains
 many features which are not in SVr4 curses. Curses is a pun on
 the term "cursor optimization". It is a library of functions that
 manage an application's display on character-cell terminals
 (e.g., VT100).

The name &ldquo;ncurses&rdquo; was first used as the name of
 the curses library in Pavel Curtis's pcurses, dated
 1982. It was apparently developed on a 4.3BSD system, at Cornell.
 Parts of pcurses are readily identifiable in ncurses, including
 the basics for the terminfo compiler (named <code>compile</code>
 in that package):

<ul>
 <li>the header filename <code>ncurses.h</code></li>

<li>
 the <code>Caps</code> file, used to define the terminfo
 capabilities
 </li>

<li>
 awk scripts <code>MKcaptab.awk</code>,
 <code>MKnames.awk</code>
 </li>

<li>
 the library modules used for the terminfo compiler.
 </li>
 </ul>

While the name of the library itself comes from
 &ldquo;pcurses&rdquo;, it had a different connotation to Curtis.
 His makefile built <code>libncurses.a</code> (normal) and
 <code>libdcurses.a</code> (debugging). In the documentation,
 Curtis referred to his package as &ldquo;the new curses&rdquo;
 like this, in <code>changes.ms</code>:

<blockquote class="code-block">
 <pre>
 New Features in Curses and Terminfo

Pavel Curtis

1.  Introduction

This  document  describes  new  features that are being
     added to the Berkeley curses subroutine  package.   It  also
     describes  the  new  terminfo  database,  which replaces the
     Berkeley termcap database.  The emphasis is on the new  fea&#8208;
     tures.

2.  New Features in Curses
     ...
     2.7.  Mini&#8208;Curses*

&#9472;&#9472;&#9472;&#9472;&#9472;&#9472;&#9472;&#9472;&#9472;&#9472;&#9472;
     *This feature is not supported in the current test
     release.  It will be implemented in  the  official
     distribution.

&#8208;6&#8208;

The new curses is bigger than the old one, and has to
 copy from the current window to an internal screen image for
 every call to refresh(). If the programmer is only inter&#8208;
 ested in screen output optimization, and does not want the
 windowing or input functions, an interface to the lower
 level routines is available. This will make the program
 somewhat smaller and faster. The interface is a subset of
 full curses, so that conversion between the levels is not
 necessary to switch from mini&#8208;curses to full curses.
</pre>
 </blockquote>

Rather than being at the start a replacement for AT&amp;T
 curses, initially it was an extension of BSD curses. Later,
 supplanting AT&amp;T curses became more important, with different
 developers.

Besides ncurses, parts of pcurses still survive in
 recognizable form in Solaris, etc., because the <a href=
 "man/tic.1m.html"><code>tic</code></a> program from pcurses was
 used by AT&amp;T (see related <a href=
 "tctest.html#history_terminfo">discussion</a>). Some
 misinformation regarding pcurses was provided in a
 <a href=
 "https://groups.google.com/d/msg/comp.terminals/qAF3c68AgkA/me4A_X3AYagJ">
 posting</a> to comp.terminals long ago:

<blockquote>
 <pre class="code-block">
Mark took "terminfo" to AT&amp;T with him, and it was adopted for UNIX System
V Release 2 as a replacement for "termcap" (which was temporarily still
supported in SVR2). UCB, however, stuck with termcap for quite a while
(through 4.3BSD at least), merely bringing the termcap manual entry up to
date with the SVR2 version of terminfo insofar as possible (I was the
editor for the 4.3BSD termcap manual entry). Pavel Curtis provided his
own public-domain implementation of terminfo/curses, but I don't think it
really caught on.
</pre>
 </blockquote>

<h2 id="who_did_it-id"><a name="who_did_it" id="who_did_it">Who
 wrote NCURSES?</a></h2>

<ul>
 <li>
 Zeyd Ben-Halim started it from a previous package
 <code>pcurses</code>, written by Pavel Curtis. The first
 widely-used version (<a href=
 "/ncurses/ncurses-license.html#ncurses_1_8_1">1.8.1</a> in
 November 1993) was not much larger than pcurses. 
 
 It added freely-available examples such as
 <code>worm</code>, 
 
 and a crude configure mechanism.
 </li>

<li>
 Eric Raymond continued development, apparently starting
 <a href="/ncurses/ncurses-license.html#ncurses_1_8_5">in late
 1993 or early 1994</a>.
 </li>

<li>
 Juergen Pfeifer wrote most of the form and menu libraries.
 Eric added those libraries to ncurses in <a href=
 "ncurses-license.html#ncurses_1_9_4">August 1995</a>, after I
 had started working with Eric and Zeyd in March 1995, before
 <a href="/ncurses/ncurses-license.html#ncurses_1_9">ncurses
 1.9</a> was released. 
 
 </li>

<li>
 I've done most of the configuration work, do most of the
 ongoing development and testing.
 </li>

<li>
 numerous people (more than I know about) contributed
 fixes.

For details, see the <a href="/ncurses/NEWS.html">NEWS</a>
 file in the source distribution. It accurately reflects
 contributions since 1.9.9e.
 </li>
 </ul>

In preparing copyright transfer in 1997, I identified <a href=
 "ncurses-license.html#metrics19970613">more than 20
 contributors</a> based on my software archives.

These individuals were <a href=
 "ncurses-license.html#final19970919">cited in the agreement</a>
 as having contributed more than 20 lines of code each:

<blockquote>
 <pre class="code-block">
Heinz-Ado Arnolds, Jeremy Buhler, Andrey Chernov, J.T. Conklin,
Ulrich Drepper, Juergen Ehling, Werner Fleck, Per Foreby, Gerhard
Fuernkranz, Anatoly Ivasyuk, Andrew Kuchling, H.J. Lu, Alexander
V. Lukyanov, David MacKenzie, Rick Marshall, Hellmuth Michaelis,
Tim Mooney, Philippe De Muyter, Eric Newton, Andreas Schwab,
Jesse Thilo, Warren Tucker, Peter Wemm.
</pre>
 </blockquote>

In addition to Florian La Roche, who agreed to act as
 maintainer, these were the principal authors of ncurses, for
 assigning copyright to the Free Software Foundation:

<ul>
 <li>Zeyd M. Ben-Halim</li>

<li>Thomas E. Dickey</li>

<li>Juergen Pfeifer</li>

<li>Eric S. Raymond</li>
 </ul>

Pavel Curtis' work is in the public domain, hence not needed
 for copyright assignment. (The <code>README</code> file in the
 ncurses distribution also identifies the authors).

Florian acted as maintainer for about a year. I continued to
 do the bulk of development, and prepared the 5.0 release. After
 Florian left unexpectedly (before 5.0), I <a href=
 "ncurses-license.html#February_1998">resumed</a> my pre-4.2 role
 as the project maintainer.

Occasionally someone says that it is &ldquo;written
 by the GNU Project&rdquo;. That is incorrect:

<ul>
 <li>
 The developers and contributors are clearly identified in
 the <a href="NEWS.html">NEWS</a> file.
 </li>

<li>
 The <a href=
 "https://github.com/ThomasDickey/ncurses-snapshots/blob/master/AUTHORS">
 AUTHORS</a> file in the sources identifies the primary
 developers.
 </li>

<li>
 A complete list of contributors runs to several pages (see
 <a href=
 "/personal/changelogs.html#examples">summary</a>).
 </li>

<li>
 The Free Software Foundation holds a copyright on the
 redistributable work to reduce disputes. It exists as a
 501(c)(3) non-profit organization, i.e., a <a href=
 "https://www.irs.gov/charities-non-profits/charitable-organizations">
 charity</a>. As such, it is prohibited from
 involvement in politics.

The GNU Project, on the other hand, is an informal
 organization which ostensibly promotes the work of the Free
 Software Foundation.
 </li>

<li>None of the major contributors to ncurses appears
 to be involved with either organization.</li>
 </ul>

See the <a href="ncurses-license.html">ncurses
 license</a> page for more information.

<h2 id="who_owns_it-id"><a name="who_owns_it" id=
 "who_owns_it">How can it be distributed?</a></h2>

The major ncurses developers (exclusive of Pavel Curtis, who
 put his work in the public domain several years before) early in
 1998 assigned their copyright to the Free Software Foundation,
 which promised to use the following distribution terms for at
 least five years.

<blockquote class="code-block">
 Permission is hereby granted, free of charge, to any person
 obtaining a copy of this software and associated documentation
 files (the "Software"), to deal in the Software without
 restriction, including without limitation the rights to use,
 copy, modify, merge, publish, distribute, distribute with
 modifications, sublicense, and/or sell copies of the Software,
 and to permit persons to whom the Software is furnished to do
 so, subject to the following conditions:

The above copyright notice and this permission notice shall
 be included in all copies or substantial portions of the
 Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY
 KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE
 WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE
 AND NONINFRINGEMENT. IN NO EVENT SHALL THE ABOVE COPYRIGHT
 HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
 WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
 OTHER DEALINGS IN THE SOFTWARE.

Except as contained in this notice, the name(s) of the above
 copyright holders shall not be used in advertising or otherwise
 to promote the sale, use or other dealings in this Software
 without prior written authorization.
 </blockquote>

<h2 id="who_claims_it-id"><a name="who_claims_it" id=
 "who_claims_it">Is it Free Software or Open Source?</a></h2>

Ncurses is free software. It is not `open source'.

That term applies to a mixture of proprietary software and
 quasi-free software, and is being promoted currently by several
 people for a variety of reasons: some as a compromise (in the
 pejorative sense) between free software and proprietary, and
 others to take credit for brokering the release of some
 proprietary software under less stringent conditions.

By relabeling free software (and revising the order of causes
 and events), the supporters of `open software' are doing the
 development community a disservice.

Surprisingly, some people cite ncurses as an example of GPL or
 LGPL. The copyright notice (which is the above-quoted license)
 appears 577 places in the 5.1 sources, including all of the
 header files. Presumably therefore, these people have not
 actually looked at ncurses.

Adding to the confusion are misleading comments such as
 <a href="http://cygwin.com/faq/faq.html#faq.what.free">this</a>:

<blockquote class="code-block">
 In particular, if you intend to port a proprietary
 (non-GPL'd) application using Cygwin, you will need the
 proprietary-use license for the Cygwin library. This is
 available for purchase; please contact sales@cygnus.com for
 more information. All other questions should be sent to the
 project mailing list cygwin@sources.redhat.com.
 </blockquote>

By omitting some of the facts, this paragraph states in terms
 of ncurses, for example, that I cannot work on ncurses on cygwin
 without buying a license for Cygwin. The same applies to the
 developers of about half of the contributed software for Cygwin,
 since not all are GPL'd. 
 
 There is a better attempt at explaining Cygwin licensing
 <a href="http://cygwin.com/licensing.html">here</a>, but the
 other page does not use it:

<blockquote class="code-block">
 This means that you can port an Open Source(tm) application
 to cygwin, and distribute that executable as if it didn't
 include a copy of libcygwin.a/cygwin1.dll linked into it. Note
 that this does not apply to the cygwin DLL itself. If you
 distribute a (possibly modified) version of the DLL you must
 adhere to the terms of the GPL, i.e. you must provide sources
 for the cygwin DLL.
 </blockquote>

Probably more people read the FAQ (and are misled) than read
 the licensing page.

This type of license, by the way, is often referred to as
 "MIT-style", referring to the MIT X distribution terms. Before
 <a href="/ncurses/ncurses-license.html#February_1998">assigning
 copyright</a> to the FSF (Free Software Foundation), <a href=
 "ncurses-license.html#patch_960907">substantial portions</a> of
 ncurses were copyrighted in this style. The main restriction that
 affects most people is that the copyright notice must be kept on
 copies&mdash;or portions of the copies. That is not done <a href=
 "http://www.mkssoftware.com/docs/cur_index.asp">in this online
 reference</a>, which documents an older version of ncurses. The
 translation from manpage to html retains the content, but removes
 the copyright notice, which one may observe is not permitted.
 Compare with <a href="/ncurses/man/ncurses.3x.html">this</a>
 (copyright notices are retained in the online content, as you can
 see in the source-view of the page).

For what it's worth, the <a href=
 "ncurses-license.html#final19970919">agreement</a> which we
 (original ncurses developers) made with the Free Software
 Foundation reads in part:

<blockquote class="code-block">
 The Foundation promises that all distribution of the
 Package, or of any work "based on the Package", that takes
 place under the control of the Foundation or its agents or
 assignees, shall be on terms that explicitly and perpetually
 permit anyone possessing a copy of the work to which the terms
 apply, and possessing accurate notice of these terms, to
 redistribute copies of the work to anyone on the same terms.
 These terms shall not restrict which members of the public
 copies may be distributed to. These terms shall not require a
 member of the public to pay any royalty to the Foundation or to
 anyone else for any permitted use of the work they apply to, or
 to communicate with the Foundation or its agents in any way
 either when redistribution is performed or on any other
 occasion.
 </blockquote>

As is well known, that precludes relicensing to the GPL in any
 version, since it would place restrictions on which programs may
 link to the libraries. That would deprive a substantial fraction
 of the current user base of the use of subsequent versions of the
 software. No such restriction exists in the ncurses license.

I have never considered it a possibility (see the preceding
 section). It would make the package unusable for most of its
 current user base, because GPL is a more-restrictive license than
 MIT-X11 or any of the similar BSD licenses.

However, since the FSF holds a copyright to the releases
 published on its website, it is not impossible that someone might
 publish a version of ncurses relicensed under the GPL. In that
 case, I would continue development based on the previous version,
 using the existing license&mdash;the one to which I initially
 agreed. Because I do almost all of the development, and provide
 the development website, doing that would have little effect on
 subsequent releases.

The original agreement stated that changes which I made to the
 source would be copyright by the Free Software Foundation. That
 clause expired after five years (in 2003). It does require
 written notice (for instance today is June 25, 2007), so in the
 event of serious disagreement with the FSF, this webpage
 satisfies that. It is worth noting that all changes that I have
 made since the most recent release would be in that event
 copyright by me.

Moving forward to February 2020, relicensing is a moot issue.
 However, some issues with FSF (such as <a href=
 "#misinfo-man7">this</a>) could best be done by taking ownership
 of the copyrights. In preparing for releasing ncurses 6.2, I sent
 mail to FSF advising them that ncurses 6.2 (and all changes since
 ncurses 6.1) will be copyright by me (same license, of
 course).

<h2 id="tack_copying-id"><a name="tack_copying" id=
 "tack_copying">What about the tack program?</a></h2>

That is not part of ncurses. As a convenience (to reuse
 library functions that are part of <code>tic</code> and
 <code>infocmp</code>), it was distributed with ncurses since
 before 5.0 (patch date <a href="NEWS.html#t990417">990417</a>).
 Beginning with release <a href=
 "/ncurses/tack/CHANGES.html#t20170726">1.08</a> in July 2017, it
 is able to work with other implementations of curses (although
 editing a terminal description&mdash;done
 rarely&mdash;requires ncurses).

However, <code>tack</code> is licensed differently: the GNU
 General Public License, version 2 (GPLv2).

This confused some packagers, who then labeled ncurses as
 GPL. Most packagers correct the designation when
 requested. Some do not. To avoid this confusion, it was removed
 from the ncurses distribution in 2007, <a href=
 "NEWS.html#t20070203">shortly after 5.6 release</a>.

Unlike ncurses, FSF does not have a release-page on its
 website for tack, and no one has suggested that it was written by
 the GNU Project. Except for a few (small changes), this was
 written by Daniel Weaver and me (see <a href=
 "https://github.com/ThomasDickey/tack-snapshots/blob/master/AUTHORS">
 AUTHORS</a> file in the source). Its homepage is
 <a href="/ncurses/tack.html">here</a>.

<h2 id="terminfo_copying-id"><a name="terminfo_copying" id=
 "terminfo_copying">What about the terminfo database?</a></h2>

<h3><a name="terminfo_license" id="terminfo_license">Copyright
 and Licensing</a></h3>

The terminfo database is a special case. Ncurses provides a
 different version of the <code>terminfo.src</code> file
 originally collected by Eric Raymond. The ncurses file is not
 maintained by Eric Raymond, since the agreement which transferred
 control to FSF states:

<blockquote class="code-block">
 We hereby agree that if we have or acquire, or any one of us
 has or acquires, hereafter any patent or interface copyright or
 other intellectual property interest dominating the Program (or
 use of the same), such dominating interest will not be used to
 undermine the effect of this assignment,
 </blockquote>

Changes made to this file are (unsurprisingly) copyrighted via
 the <a href=
 "http://www.wipo.int/treaties/en/ip/berne/trtdocs_wo001.html">Berne
 convention</a>. No explicit "Copyright &copy;" is required. It
 only requires that the author be identified (and this is done in
 the history comments at the end of the file):

<blockquote class="code-block">
 (1) In order that the author of a literary or artistic work
 protected by this Convention shall, in the absence of proof to
 the contrary, be regarded as such, and consequently be entitled
 to institute infringement proceedings in the countries of the
 Union, it shall be sufficient for his name to appear on the
 work in the usual manner. This paragraph shall be applicable
 even if this name is a pseudonym, where the pseudonym adopted
 by the author leaves no doubt as to his identity.
 </blockquote>

and noting the first comment in the file states that it is
 maintained as part of ncurses, this applies:

<blockquote class="code-block">
 (3) In the case of anonymous and pseudonymous works, other
 than those referred to in paragraph (1) above, the publisher
 whose name appears on the work shall, in the absence of proof
 to the contrary, be deemed to represent the author, and in this
 capacity he shall be entitled to protect and enforce the
 author's rights. The provisions of this paragraph shall cease
 to apply when the author reveals his identity and establishes
 his claim to authorship of the work.
 </blockquote>

Raymond wrote a disclaimer in the <code>terminfo.src</code>
 file which disagrees with some of this. However, Raymond's
 disclaimer was based on more than one misconception. There is an
 interesting story about that in <a href=
 "/personal/copyrights.html#removing_notes">another place</a>.

<h3><a name="terminfo_attribution" id=
 "terminfo_attribution">Attribution</a></h3>

Occasionally someone in a newsgroup posts a terminfo which has
 been exported using <code>infocmp</code>, saying that it is
 theirs (and even written by them). Sometimes the claim is true,
 though more often the data is identical to that from ncurses or a
 package which includes ncurses. The latter case is
 interesting.

For instance <a href="http://www.openqm.com/">OpenQM</a> used
 terminfo entries which were obtained from ncurses using
 <code>infocmp</code>. In discussion, one aspect glossed over was
 that some of the content was copied not from ncurses itself but
 from a packager's patch which merged a <a href=
 "/xterm/xterm.html">xterm</a> terminfo which I wrote. Ultimately,
 the responses from that discussion boiled down to saying that
 they found it and it's theirs. (The maintainer did agree to add a
 comment noting the origin of the "public domain" entries).

<h3><a name="terminfo_others" id="terminfo_others">Other
 Versions</a></h3>

Eric Raymond's website has an old version misleadingly
 numbered "11.0". It actually is much older than ncurses's
 terminfo (whose major version I have left as "10").

The content of "11.0" is derived from <a href=
 "announce-5.0.html">ncurses 5.0</a>. It makes more than one
 change, but most are cosmetic (e.g., reordering the entries
 within the file, adding about 300 lines of comments&mdash;in an
 18,655 line file&mdash;to make the reordering look nicer). None
 of the added comments are useful.

It also modifies the changelog from ncurses to make it appear
 that people who reported problems to me were the ones who did the
 subsequent investigation and patches. He had done the same thing
 prior to ncurses 4.2, for several months in 1997 and 1998,
 copying changes from ncurses development, and then later
 "revising" the change history. (I later restored that portion of
 the changelog).

The last version (11.0.1) copies one of my fixes from ncurses,
 (prompted by a bug-report by Tijs Michels on the rxvt-workers
 mailing list in January 2000, in response to ESR's announcement
 of "11.0.0"). In the mailing list, I pointed out the reason for
 my change. My original change comment in 1997 said

#&nbsp;10.1.16&nbsp;(Sat&nbsp;Dec&nbsp;13&nbsp;19:41:59&nbsp;EST&nbsp;1997)&nbsp;

#&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;*&nbsp;remove&nbsp;hpa/vpa&nbsp;from&nbsp;rxvt,&nbsp;which&nbsp;implements&nbsp;them&nbsp;incorrectly.&nbsp;

#&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;*&nbsp;add&nbsp;sgr0&nbsp;for&nbsp;rxvt.&nbsp;

#&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;*&nbsp;remove&nbsp;bogus&nbsp;smacs/rmacs&nbsp;from&nbsp;EMX&nbsp;descriptions.

</blockquote>

GNU termcap 1.3.1 distributes "11.0.1", generated using
 ncurses' <a href="/ncurses/man/tic.1m.html">tic</a> program.
 Disregarding the issue that the file is old and faulty, the
 reason for generating the file is that ESR's website does not
 provide a version which resolves all but the last
 "<code>tc=</code>" inclusions.

The answer depends on why you are asking. Some of the reasons
 for asking include

<ul>
 <li>
 does it have some given feature.
 </li>

<li>
 will it fit in some limited-size embedded application,
 </li>
 </ul>

<h3><a name="growing_features" id="growing_features">Growth of
 the feature set</a></h3>

As noted, ncurses began as a clone of SVr4 curses (up til
 around 1995). Later, The Open Group began a specification of
 curses (XSI Curses), which is now known as X/Open Curses. That
 provided for "wide-characters", such as Unicode, although at that
 point Unicode was not the obvious answer.

There were a few functions (such as <code>attr_get</code>)
 which were added into the ncurses library in anticipation of XSI
 Curses. However, I chose to implement the wide-character support
 using a different library name, "ncursesw". Doing that allowed me
 to maintain compatibility with applications that used the
 existing "ncurses" library.

Given that, the "ncurses" library would be comparable to a
 SVr4 implementation (such as Solaris, IRIX64 or HPUX), while the
 "ncursesw" library would be comparable to one of the XPG4
 implementations (Tru64 being the notable implementation). The
 ncurses (and ncursesw) library provide some extended functions
 not found in SVr4/XPG4. A few implementations (<a href=
 "#other-pdcurses">PDCurses</a> and <a href="#other-netbsd">NetBSD
 curses</a>) provide some of these extensions.

Later, I added functions to support simple threaded
 applications, and Juergen Pfeifer extended that. Again, to
 maintain compatibility, this is normally built as a a new library
 name, "ncursest" or "ncurseswt". First noted in mid-2011, at this
 time (mid-2020), there is no other implementation of these
 features.

The ncurses and ncursesw libraries are reasonably
 source-compatible. That is, an application written for "ncurses"
 will build with "ncursesw". But it will behave differently in
 response to your locale settings. (Some distributors, who do not
 care about the differences, have chosen to merge the names
 together as "ncurses").

A few applications require changes to use "ncursest", since
 internal details of the <code>WINDOW</code> object are not
 directly visible in the latter. However, the "ncurses" library
 has macros and functions which address this area.

Every implementation of curses uses both macros and functions
 to provide their features. ncurses follows the XPG4 convention
 where all macros (except for those such as <code>getyx</code>
 which must be implemented solely as macros) are also
 implemented as functions.

Here are counts comparing ncurses 5.9 with other
 implementations:

<blockquote>
 <table border="1" summary=
 "Counts of functions and macros for different curses implementations">
 <tr>
 <th align="left">Implementation</th>

<th align="left">Macros</th>

<th align="left">Public symbols</th>
 </tr>

<tr>
 <td>ncurseswt (sp-funcs)</td>

<tr>
 <td>ncursest (sp-funcs)</td>

<tr>
 <td>ncurseswt</td>

<tr>
 <td>ncursest</td>

<tr>
 <td>ncursesw</td>

<tr>
 <td>ncurses</td>

<tr>
 <td>Solaris 10 XPG4</td>

<tr>
 <td>Solaris 10 SVr4</td>

<tr>
 <td>PDCurses 3.4</td>

<tr>
 <td>NetBSD 5.1</td>

In each case, internal functions are not counted.

The Unix implementations include undocumented features for
 compatibility with older curses implementations that are not
 provided by ncurses.

Both <a href="#other-pdcurses">PDCurses</a> and <a href=
 "#other-netbsd">NetBSD curses</a> contain functions not counted
 here because they are not relevant to a comparison with SVr4/XPG4
 curses. For example, PDCurses in X11 contains functions for
 initializing the X window. On the other hand, relevant extensions
 such as PDCurses' version of <code>wresize</code> are
 counted.

<h3><a name="uses_of_library" id="uses_of_library">Types of
 library users</a></h3>

PDCurses on the other hand does not include the functions used
 to obtain terminfo information. That does not prevent it from
 being a curses implementation. X/Open Curses' documentation
 treats those separately, allowing for the possibility of curses
 implementations without terminfo (or termcap either for that
 matter).

That raises another issue: what types of interfaces do curses
 libraries (and ncurses in particular) support?

ncurses includes the conventional curses interfaces, with
 extensions (new functions) in each interface. The interfaces
 are:

<ul>
 <li><a href="/ncurses/man/form.3x.html">form</a>, <a href=
 "/ncurses/man/menu.3x.html">menu</a>: upper-level libraries
 provided in a few systems such as SCO and Solaris.</li>

<li><a href="/ncurses/man/panel.3x.html">panel</a>: an
 upper-level library which should have been standarized as part
 of the curses library. 
 Again (outside of ncurses), it is provided by SCO and
 Solaris.</li>

<li><a href="/ncurses/man/ncurses.3x.html">curses</a>: the
 higher-level interface used by all curses
 applications.</li>

<li><a href="/ncurses/man/curs_terminfo.3x.html">terminfo</a>:
 a lower-level interface used by some applications.</li>

<li><a href="/ncurses/man/curs_termcap.3x.html">termcap</a>: a
 lower-level interface used by some applications.</li>
 </ul>

Curiously, NetBSD provides variants of the form and menu
 libraries, but lacks the panel library. Conversely, PDCurses
 provides a panel library but lacks the form and menu libraries.
 Because NetBSD uses opaque structures, it is difficult
 to write portable applications using their form and menu
 libraries. The <a href=
 "ncurses-examples.html">ncurses-examples</a> have some
 workarounds for this to allow them to compile with NetBSD curses.
 One might have better luck with PDCurses' panel library, which
 has the same ancestor as ncurses' panel library.

The lower-level interfaces rely upon the application (rather
 than the library) to decide how to put characters on the display.
 That is referred to as "optimization", e.g., in the
 documentation:

<blockquote class="code-block">
 The ncurses library routines give the user a
 terminal-independent method of updating character screens with
 reasonable optimization.
 </blockquote>

Most applications made the transition to using the curses
 interface long ago. There are some applications which use one of
 the lower-level interfaces. In some cases there are technical
 reasons for this. Inertia is more often the reason.

Here are some examples of well-known applications which use
 curses:

<ul>
 <li><a href="https://lynx.invisible-island.net/">lynx</a> 
 A curses application from the beginning, lynx can be built with
 any version of curses, including PDCurses and VMS curses.</li>

Keith Bostic introduced nvi in 4.4BSD, revised it for
 4.4BSDLite in 1994. At that point, it was using curses
 (preferably System5 curses). Lacking that, it had a bundled
 copy of 4.4BSD curses, which sufficed. Bostic's interest in
 ncurses stemmed from his desire to eliminate the bundled 4.4BSD
 curses, and he was helpful in many ways, both technical (with
 bug reports) and <a href=
 "/ncurses/ncurses-license.html#January_1997">otherwise</a>.</li>

<li><a href="http://www.tin.org">tin</a> 
 Originally in K&amp;R C using termcap, I <a href=
 "/tin/tin.html#history">modified</a> it to use curses. Most of
 the subsequent work to handle wide-characters was done by
 others.</li>
 </ul>

Widely-used termcap applications include these:

<ul>
 <li><a href=
 "http://www.greenwoodsoftware.com/less/">less</a> 
 Several <code>less</code> features are undocumented. One of
 these features is the use of environment variables to override
 termcap capabilities. At startup, when <code>less</code>
 retrieves each termcap capability, it first checks for an
 environment variable (named by prefixing the capability name
 with &ldquo;<code>LESS_TERMCAP_</code>&rdquo;) and uses the
 environment variable's value if it exists. That appeared in
 <code>less</code> between versions 321 and 330
 (July&ndash;October 1996).</li>

<li><a href=
 "http://www.gnu.org/software/screen/">screen</a> 
 The <code>screen</code> program has had more impact on ncurses
 development than the other applications because there were
 features which <code>screen</code>'s developers wanted for
 compatibility with &ldquo;real&rdquo; (runtime adjustments of
 <code>sgr0</code> to imitate termcap's corresponding
 <code>me</code> in <a href=
 "/ncurses/NEWS.html#t20010908">2001</a> and using
 padding in <a href=
 "man/curs_util.3x.html#h3-delay_output"><code>delay_output</code></a>
 rather than time-delays in <a href=
 "/ncurses/NEWS.html#t990102">1999</a>). The latter was marked
 as a problem in screen's source code; it took several years for
 that to be addressed (see <a href=
 "/gnu-patches/index.html#related">related discussion</a>).</li>

<li><a href="http://w3m.sourceforge.net">w3m</a> 
 While <code>w3m</code> is a termcap application, it includes
 source code that makes it pretend to use curses. For instance,
 it includes a module <a href=
 "https://github.com/shinh/w3m/blob/master/terms.c">terms.c</a>
 whose header comment is An original curses library for
 EUC-kanji by Akinori ITO, December 1989. This module does
 not do optimization as a curses library would do, but
 uses a simpler approach (which works well with multibyte
 characters). What complicates the program is that some of its
 symbols duplicate those in the curses library with which it is
 linked (see <a href="#w3m-vs-gpm">related discussion</a>).</li>

<li><a href="http://www.vim.org">vim</a> 
 Vim's developers provide their own termcap definitions, just in
 case there is nothing good enough on the system, and provide
 features for overriding the system's termcap capabilities even
 when using the system definition as a starting point. Because
 of this, vim has had little impact on ncurses development; the
 change-log does not mention vim. On the other hand, unlike
 <code>screen</code>, the source-code for vim still contains
 <a href=
 "https://github.com/vim/vim/blob/61d7c0d52ca40ab8488c36e619d1e46503affd0b/src/term.c#L2601">
 a comment</a> about a bug in ncurses which was addressed in
 <a href="/ncurses/NEWS.html#t971220">late 1997</a>, before
 ncurses 4.2 was released.</li>
 </ul>

Applications which use only terminfo are less
 well-known. For example, there are

<ul>
 <li><a href=
 "/ncurses/ncurses-examples.html">ncurses-examples</a> (e.g.,
 <a href=
 "https://github.com/ThomasDickey/ncurses-snapshots/blob/master/test/demo_terminfo.c">
 demo_terminfo</a>, <a href=
 "https://github.com/ThomasDickey/ncurses-snapshots/blob/master/test/list_keys.c">
 list_keys</a> and <a href=
 "https://github.com/ThomasDickey/ncurses-snapshots/blob/master/test/test_sgr.c">
 test_sgr</a>)</li>

Because the terminfo and termcap programming interfaces are
 similar, it is trivial to make an application build with terminfo
 if it was originally written to use termcap. These are done using
 a table for the capabilities, taking advantage of the fact that
 all of the conventional termcap capabilities are associated with
 terminfo capabilities. Here are a few examples:

<li><a href=
 "https://github.com/ThomasDickey/xterm-snapshots/blob/master/xtermcap.c">
 xterm</a></li>
 </ul>

Going the other way (starting with a terminfo application and
 making it work with termcap) is not necessarily simple:

<ul>
 <li>
 The terminfo interface provides a data structure <a name=
 "TERMTYPE-struct" id=
 "TERMTYPE-struct"><code>TERMTYPE</code></a> holding all of
 the terminal's information which eliminates the need for
 calls to the tigetstr, tigetnum and tigetflag functions. If
 an application (such as tack) is written to use
 that, it is difficult to make it use termcap.
 </li>

<li>
 Additionally, since <a href=
 "#extended-terminfo-1999">1999</a>, ncurses provides <a href=
 "/ncurses/terminfo.src.html#toc-_N_C_U_R_S_E_S__U_S_E_R-_D_E_F_I_N_A_B_L_E__C_A_P_A_B_I_L_I_T_I_E_S">
 extended capabilities</a> in its terminal database
 which are not necessarily visible to termcap applications.
 XTerm, in its <a href=
 "/xterm/manpage/xterm.html#Application-Resources:tcapFunctionKeys">
 <code>tcapFunctionKeys</code></a> feature uses these
 capabilities; gnome-terminal in its analogous termcap
 interface (removed in 2014) could not.

The <a href="man/user_caps.5.html">user_caps(5)</a> manual
 page goes into greater detail.
 </li>
 </ul>

Because of these differences, it is best to speak of
 termcap-applications, terminfo-applications and
 curses-applications. Curses-applications may use terminfo
 functions, but the reverse is not true. Given the terminfo
 interface, there is no reason for a curses-application to use the
 termcap interface.

An easy way to distinguish between the three types of
 applications is to see what library calls they make:

<ul>
 <li>curses &ndash; <a href=
 "/ncurses/man/curs_initscr.3x.html">initscr</a> or newterm</li>

<li>terminfo &ndash; <a href=
 "/ncurses/man/curs_terminfo.3x.html">setupterm</a></li>

<li>termcap &ndash; <a href=
 "/ncurses/man/curs_termcap.3x.html">tgetent</a></li>
 </ul>

Even this categorization is simplified:

<ul>
 <li>
 Some curses applications (such as nvi) switch
 between full-screen and normal modes of operation, e.g., to
 support the &ldquo;open mode&rdquo; of vi. To
 accomplish that, nvi uses lower-level calls at
 times, e.g., to erase characters.
 </li>

<li>
 nvi (and <a href=
 "/dialog/dialog.html">dialog</a>) treat <a href=
 "/xterm/xterm.html">xterm</a>'s alternate screen feature
 specially, using low-level workarounds to disable the feature
 as they switch to/from full-screen mode.
 </li>

<li>
 nvi (and <a href="/vile/vile.html">vile</a>,
 using its curses driver) do not use the curses library's
 input function <a href=
 "/ncurses/man/curs_getch.3x.html">wgetch</a>. That is because
 vi treats the escape character specially.
 Consequently, the curses decoding of function- and
 special-keys is not usable by vi-like programs.
 </li>
 </ul>

Nothwithstanding their use of low-level calls, these are
 curses applications because they use the library to update the
 display. Non-curses (termcap and terminfo) programs use only the
 terminal database. Applications which do their own display
 optimization (such as vim) are not curses
 applications.

If someone simply categorizes any application which is linked
 to (or uses in some sense) the ncurses library as an ncurses
 application, that rapidly leads to absurd conclusions, as
 for example in Debian <a href=
 "https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=265631">#265631</a>,
 where a developer insisted that a bug (apparently in a different
 library) was a bug in ncurses because the application in question
 used the <a href="#using_gpm_lib">gpm library</a>. In other
 cases, e.g., lists made by nondevelopers based on package
 dependencies, this sort of misinformation is commonplace.

<h3><a name="related_to_vi" id="related_to_vi">Relationship with
 vi</a></h3>

Not all implementations of vi are curses
 applications. In fact, considering it carefully, most are
 not.

One likely objection to that statement is the generally
 accepted notion that Ken Arnold &ldquo;took&rdquo; functions from
 Bill Joy's vi to make his library. For instance, in the
 preface to <a href="#book_goodheart">UNIX Curses
 Explained</a>, Goodheart said (in 1991):

<blockquote class="code-block">
 It all started in the late 1970s when Bill Joy, in writing
 his editor ex (probably more famous by the name
 vi nowadays), wrote a set of routines which read a
 terminal capability database. The database, then named
 termcap, generally described how to manipulate
 individual terminals and what they where capable of. The
 routines he created, which accessed the termcap
 database, implemented optimal cursor movement.

Kenneth C.R.C. Arnold took these routines almost without
 changes and derived from them what is known today as the
 curses package.
 </blockquote>

People tend to refer to the second paragraph without noticing
 the first one. The "where" versus "were"
 typographical error in the first paragraph is an error by the
 publisher. The last sentence of the first paragraph is an error
 by the author: termcap does not, never did do that.

Optimal cursor movement (cursor optimization) in ex
 was done in the remainder of the application. The termcap
 functions were readily reusable; the remainder was not. Arnold
 added to Joy's original idea by making the library
 reusable.

<a href="http://www.tuhs.org/">The Unix Heritage Society</a>
 has copies of 3BSD, 4.2BSD, 4.3BSD (multiple). <a name=
 "contrast-vi-curses" id="contrast-vi-curses">Using those, one can
 make useful comparisons</a> :

<ul>
 <li><a href="/ncurses/42BSD-ex_vis.h.html">ex_vis.h</a> file
 from ex in the 4.2BSD distribution</li>

<li><a href="/ncurses/42BSD-curses.h.html">curses.h</a> from
 curses library in the same distribution.</li>
 </ul>

Joy's header file is littered with separate variables.
 Arnold's header file organizes those into structures, notably
 <code>WINDOW</code>. The <code>vtube</code> array in the former
 corresponds to the bulk of the storage used by
 <code>WINDOW</code>. In the same way, Arnold recognized useful
 aspects of terminal manipulation scattered through ex,
 and organized those into a library which improved on the original
 design. Aside from <code>beep</code>, there is no straightforward
 mapping between the functions in the two programs.

Traditional vi, however, does not use curses. With
 the release of 4.2BSD, it lost its private copies of the termcap
 functions, and, in the main branch of development, at AT&amp;T,
 became a terminfo application. With 4.3BSD, the termcap-based
 vi (and derived code such as <a href=
 "http://heirloom.sourceforge.net/vi.html">"heirloom vi"</a>) was
 no longer the main line of development. Bostic's nvi was
 a re-implementation (based on elvis), done specifically
 to disentangle the BSD project from AT&amp;T source code.

Switching to terminfo was simple. Switching back would be
 hard, since the AT&amp;T vi uses the <a href=
 "#TERMTYPE-struct"><code>TERMTYPE</code></a> structure for
 accessing the terminal capabilities. AIX, HPUX, IRIX64, Solaris
 all use the AT&amp;T-derived vi.

<h3><a name="not_library_users" id=
 "not_library_users">Applications miscited as library users</a></h3>

Here is a short list of applications commonly confused as
 users of curses, terminfo or termcap. The reason for the
 confusion is that they use the <code>TERM</code> environment
 variable, and some give a little attention to the terminal
 database. But the essential approach taken by their developers is
 to hard-code their assumptions about the terminals.

<ul>
 <li id="slang">
 The &ldquo;S-Lang&rdquo; (more commonly just <a href=
 "ncurses-slang.html">slang</a>) library may
 use the termcap file, or its own (subset of ncurses)
 interface to the terminfo database. However, it overrides the
 terminfo/termcap data with its author's assumptions which can
 differ radically from your actual terminal capabilities.
 </li>

<li>
 ELinks (and its ancestors links, links2), generically
 referred to as (e)links(2) <a href=
 "http://elinks.or.cz/documentation/html/manual.html-chunked/ch08.html">
 assumes</a> everything is some variation of Linux console,
 xterm or <a href="#vt100_color">color-vt100</a>. The <a href=
 "http://www.tldp.org/HOWTO/Text-Terminal-HOWTO-17.html">Text
 Terminal HOWTO</a> provides some misinformation on this
 topic

<blockquote>
 But links2 and elinks will work with
 any terminal by classifying the type as dumb like the first
 terminals.
 </blockquote>

Rather, (e)links(2) simply uses VT100 escape sequences if
 it notices that <code>TERM</code> is set to
 dumb.
 </li>

<li>
 URWid (see for example <a href=
 "http://stackoverflow.com/questions/2374614/are-there-any-tree-libraries-widgets-for-ncurses/42051716#42051716">
 are there any tree libraries/widgets for
 (n)curses</a>). According to its <a href=
 "http://urwid.org/manual/displaymodules.html">documentation</a>:

<blockquote>
 If you don&rsquo;t specify a display
 module, the default main loop will use <a href=
 "http://urwid.org/reference/display_modules.html#urwid.raw_display.Screen">
 raw_display.Screen</a> by default.
 </blockquote>

and goes on to mention that when using (n)curses, there is
 no support for 88- or 256-colors or mouse support. The point
 which URWid's developers are making is that the curses module
 is there solely to provide justification for the hard-coded
 &ldquo;raw display&rdquo; module.
 </li>

<li>
 Linux <a href=
 "http://man7.org/linux/man-pages/man1/setterm.1.html">setterm</a>,
 which knows many of the Linux escape sequences documents
 documented in <a href=
 "http://man7.org/linux/man-pages/man4/console_codes.4.html">console_codes</a>.
 However, of its 40 command-line options, 60% &mdash; all of
 those which are frequently used &mdash; duplicate <a href=
 "man/tput.1.html"><code>tput</code></a>, <a href=
 "man/tset.1.html#h3-reset---reinitialization"><code>reset</code></a>,
 or <a href="man/tabs.1.html"><code>tabs</code></a>. The
 others are rarely used, Linux-specific.
 </li>

<li>
 <a href=
 "http://man7.org/linux/man-pages/man1/ls.1.html">GNU ls</a>
 and <a href=
 "http://man7.org/linux/man-pages/man1/dircolors.1.html">dircolors</a>.
 The latter program has a list of <code>TERM</code> values
 which it assumes all support color, using <a href=
 "#bce_mismatches"><code>bce</code> (back color erase)</a>,
 including &ldquo;vt100&rdquo; (which <a href=
 "#vt100_color">never supported color</a>).

For other comments, see

<ul>
 <li><a href=
 "http://unix.stackexchange.com/questions/316878/solarized-colorscheme-in-fbterm/316912#316912">
 Solarized colorscheme in fbterm?</a></li>

<li><a href=
 "http://stackoverflow.com/questions/40574819/how-to-remove-dir-background-in-ls-color-output/40575734#40575734">
 How to remove dir background in `ls -color`
 output</a></li>
 </ul>
 </li>
 </ul>

<h2 id="what_platforms-id"><a name="what_platforms" id=
 "what_platforms">What platforms does it run on?</a></h2>

It should build and work on any POSIX platform. It also works
 on some platforms that are non-POSIX. However, it
 requires these POSIX features (available on all
 supported platforms):

<ul>
 <li>
 a standard C compiler (c89 or later)
 </li>

<li>
 a POSIX shell (which requires a workaround for Solaris'
 obsolete shell&mdash;see the <a href=
 "announce-6.0.html#h4-config-major">6.0 release
 notes</a>).
 </li>
 </ul>

Current development is focused on the wide-curses
 configuration (ncursesw). These platforms are known to work:

<ul>
 <li>BSD variants:

<ul>
 <li>FreeBSD 8-12</li>

<li>OpenBSD 6</li>

<li>NetBSD 7-8</li>
 </ul>
 </li>

<li>Linux-based systems

<ul>
 <li>Debian 6-10</li>

<li>Fedora 28-31, CentOS 6-8</li>

The normal (8-bit character) configuration is known to work on
 the same platforms. I've also built these in preparation for the
 current release:

<li>Solaris 11, 10, 9, 8</li>

<li>Windows 10, 8, 7, using Cygwin, MinGW and MSYS2.</li>
 </ul>

Previous releases listed other platforms (or versions) which
 are no longer available for development and testing, e.g.,
 IRIX64, SCO OpenServer, SunOS 4, Tru64. Most of those probably
 work as well, but keep in mind that because they are <a href=
 "/personal/bug-reports.html#dead_systems">defunct</a>, only minor
 changes would be considered in bug reports.

That is a summary of the recent work which
 I have done. Others package ncurses and provide it either as the
 system curses library, or as an add-on. Some people assume that
 ncurses is a &ldquo;Linux&rdquo; program. That is incorrect.
 Since half of my users are on non-Linux systems, here are some
 notes:

<ul>
 <li>
 <a href=
 "https://svnweb.freebsd.org/base/head/lib/ncurses/">FreeBSD</a>
 and <a href=
 "https://cvsweb.openbsd.org/src/lib/libcurses/">OpenBSD</a>
 use ncurses as the system curses library.

FreeBSD's system curses library limits its database to the
 hashed database <a href=
 "https://svnweb.freebsd.org/base/head/lib/ncurses/ncurses/termcap.c?revision=326025&amp;view=markup">
 termcap reader</a> from <a href=
 "https://github.com/ThomasDickey/ncurses-snapshots/blob/master/ncurses/tinfo/read_termcap.c">
 ncurses</a>. The ncurses utility programs are not used with
 FreeBSD's system curses (including <a href=
 "https://svnweb.freebsd.org/base/head/usr.bin/tabs/">tabs</a>
 and <a href=
 "https://svnweb.freebsd.org/base/head/usr.bin/tput/">tput</a>&mdash;
 but see discussion in the <a href=
 "man/tput.1.html#h2-HISTORY">manual page</a>). The add-on
 <a href=
 "https://svnweb.freebsd.org/ports/head/devel/ncurses/">ncurses
 port</a> provides a terminfo database, along with the
 ncurses utility programs. Originally that used a
 conventional directory-tree configuration, but changed in
 <a href=
 "https://svnweb.freebsd.org/ports/head/devel/ncurses/Makefile?r1=327793&amp;r2=337857">
 2013</a> to use ncurses' hashed-database feature.

OpenBSD, on the other hand, used to use its own terminfo
 reader for a hashed database. That was <a href=
 "https://cvsweb.openbsd.org/cgi-bin/cvsweb/src/lib/libcurses/tinfo/read_entry.c.diff?r1=1.16&amp;r2=1.17">
 discarded</a> late in <a href=
 "https://cvsweb.openbsd.org/src/lib/libcurses/tinfo/read_entry.c?rev=1.17&amp;content-type=text/x-cvsweb-markup">
 2015</a>.

Both of those systems (FreeBSD in 2020, OpenBSD until
 2015) use source-only entries in their respective hashed
 system terminal databases. Since <a href=
 "/ncurses/announce-5.6.html">2006</a>, ncurses
 supports an alternate configuration, storing compiled
 (binary) entries in its hashed database, to improve runtime
 performance.
 </li>

<li>
 <a href=
 "http://cvsweb.netbsd.org/bsdweb.cgi/src/share/terminfo/?only_with_tag=MAIN">
 NetBSD</a> and Solaris 11 use ncurses' terminal database,
 with their different implementations of curses.
 </li>

<li>
 As of January 2020, <a href=
 "https://opensource.apple.com/source/ncurses/ncurses-53.200.3/">
 MacOS</a> and <a href=
 "https://cvsweb.openbsd.org/src/lib/libcurses/">OpenBSD</a>
 provide <a href="/ncurses/announce-5.7.html">ncurses 5.7</a>
 (from 2008) for their system curses library.

Due to its age, ncurses 5.7 may not work as expected. For
 instance, the command <a href=
 "https://unix.stackexchange.com/questions/552845/tput-ed-is-empty/556123#556123">
 <code>tput&nbsp;ed</code></a> was seen to not work
 with MacOS.
 </li>

<li>
 Add-on packages (or &ldquo;ports&rdquo;) for recent
 releases are available for all of the BSDs, except for
 OpenBSD:

<ul>
 <li><a href=
 "https://svnweb.freebsd.org/ports/head/devel/ncurses/">FreeBSD
 ports</a></li>

<li><a href=
 "https://github.com/macports/macports-ports/blob/master/devel/ncurses/Portfile">
 MacPorts</a> and <a href=
 "https://formulae.brew.sh/formula/ncurses">brew</a></li>

<li><a href="http://pkgsrc.se/devel/ncurses">pkgsrc</a>,
 for NetBSD and others.</li>
 </ul>
 </li>

<li>
 Solaris has a <a href=
 "https://github.com/oracle/solaris-userland/tree/master/components/ncurses">
 component</a> for ncurses, which is a dependency of
 several other components, such as less,
 python and vim.
 </li>
 </ul>

<h2 id="latest_version-id"><a name="latest_version" id=
 "latest_version">What is the latest version?</a></h2>

The current version is 6.2 20200212, available from the
 ncurses homepage:

<blockquote>
 Ftp: <a href=
 "ftp://ftp.invisible-island.net/ncurses/ncurses-6.2.tar.gz">ncurses.tar.gz</a>
 (gzip'd tar)
 </blockquote>

as well as the GNU site:

<blockquote>
 Ftp: <a href=
 "ftp://ftp.gnu.org/pub/gnu/ncurses/">/pub/gnu/ncurses
 directory</a>
 </blockquote>

I also maintain patches toward the next release. See the
 <a href="/ncurses/NEWS.gz">NEWS</a> file for changes.

Ftp: <a href=
 "ftp://ftp.invisible-island.net/ncurses/6.2/">ftp://invisible-island.net/ncurses/6.2/</a>

<h2 id="other_versions-id"><a name="other_versions" id=
 "other_versions">Official releases</a></h2>

There have been a number of releases of ncurses. Some are
 available on CDROM (beginning with 1.9.4), and are archived on
 various ftp servers. If you are downloading, however, the older
 versions are of limited interest except for software testers.

<ul>
 <li>
 <a href=
 "ftp://ftp.gnu.org/gnu/ncurses/ncurses-6.2.tar.gz">6.2</a>
 <a href="/ncurses/announce-6.2.html">(12 February 2020)</a>.
 This corrects two problems in tic which interfered with
 extending the terminal database. It also improves checking
 for mis-typed user-defined capabilities in the terminal
 database.
 </li>

<li>
 <a href=
 "ftp://ftp.gnu.org/gnu/ncurses/ncurses-6.1.tar.gz">6.1</a>
 <a href="/ncurses/announce-6.1.html">(27 January 2018)</a>.
 This improves integration among the command-line utilities
 (tput, tset, clear). It also extends the range of numeric
 capabilities.
 </li>

<li>
 <a href=
 "ftp://ftp.gnu.org/gnu/ncurses/ncurses-6.0.tar.gz">6.0</a>
 <a href="/ncurses/announce-6.0.html">(8 August 2015)</a>.
 This updates the ABI to support 256-colors, as well as
 improving termcap compatibility, and the MinGW port.
 </li>

<li>
 <a href=
 "ftp://ftp.gnu.org/gnu/ncurses/ncurses-5.9.tar.gz">5.9</a>
 <a href="/ncurses/announce-5.9.html">(4 April 2011)</a>. This
 fixes a regression in <code>newwin</code>, and improves
 configurability/portability of the Ada95 binding when built
 as a separate tree. <a href="/ncurses/announce-5.8.html">(26
 February 2011)</a>. This extends support for threaded
 applications by providing a new API which eliminates the need
 for a global screen-pointer. It also introduces a port to
 Windows using MinGW, as noted in the <a href=
 "/ncurses/announce-5.8.html">5.8 announcement</a>.
 </li>

<li>
 <a href=
 "ftp://ftp.gnu.org/gnu/ncurses/ncurses-5.7.tar.gz">5.7</a>
 <a href="/ncurses/announce-5.7.html">(2 November 2008)</a>.
 This provides rudimentary support for threaded applications.
 It also distributes <code>tack</code> separately.
 </li>

<li>
 <a href=
 "ftp://ftp.gnu.org/gnu/ncurses/ncurses-5.6.tar.gz">5.6</a>
 <a href="/ncurses/announce-5.6.html">(17 December 2006)</a>.
 This supports hashed database for the terminal descriptions.
 It also improves support for magic cookies.
 </li>

<li>
 <a href=
 "ftp://ftp.gnu.org/gnu/ncurses/ncurses-5.5.tar.gz">5.5</a>
 <a href="/ncurses/announce-5.5.html">(10 October 2005)</a>.
 This improves support for wide- and multibyte characters,
 modifies the form- and menu-libraries to work with multibyte
 characters. It also improves support for cross-compiling.
 </li>

<li>
 <a href=
 "ftp://ftp.gnu.org/gnu/ncurses/ncurses-5.4.tar.gz">5.4</a>
 <a href="/ncurses/announce-5.4.html">(8 February 2004)</a>.
 This improves support for wide- and multibyte characters,
 implements the remaining parts of the X/Open Curses
 interface. It also improves support for termcap.
 </li>

<li>
 <a href=
 "ftp://ftp.gnu.org/gnu/ncurses/ncurses-5.3.tar.gz">5.3</a>
 <a href="/ncurses/announce-5.3.html">(12 October 2002)</a>.
 This provides support for wide- and multibyte characters,
 implements most of the corresponding X/Open Curses interface.
 It also improves support for termcap.
 </li>

<li>
 <a href=
 "ftp://ftp.gnu.org/gnu/ncurses/ncurses-5.2.tar.gz">5.2</a>
 <a href="/ncurses/announce-5.2.html">(21 October 2000)</a>.
 This provides better support for termcap, fixes for the
 manpage and shared library configurations, and improved
 checks for corrupt terminfo database.
 </li>

<li>
 <a href=
 "ftp://ftp.gnu.org/gnu/ncurses/ncurses-5.1.tar.gz">5.1</a>
 <a href="/ncurses/announce-5.1.html">(8 July 2000)</a>. This
 provides better support for termcap as well as implementing
 new extensions for colors.
 </li>

<li>
 <a href=
 "ftp://ftp.gnu.org/gnu/ncurses/ncurses-5.0.tar.gz">5.0</a>
 <a href="/ncurses/announce-5.0.html">(23 October 1999)</a>.
 This aligns interfaces to match version 2 of the X/Open
 Curses Single Unix specification from 1997. It also provides
 better termcap support, and termcap-like features such as
 user-defined terminal capabilities.
 </li>

<li>
 <a href=
 "ftp://ftp.gnu.org/gnu/ncurses/ncurses-4.2.tar.gz">4.2</a>
 <a href="/ncurses/announce-4.2.html">(2 March 1998)</a>. This
 implements C++ bindings for the library, as well as
 extensions to support runtime definition of keycodes.
 </li>

<li>
 4.1 <a href="/ncurses/announce-4.1.html">(16 May
 1997)</a>. This release features support for GPM (mouse on
 Linux console), and adds an extension for default colors.
 </li>

<li>
 4.0 <a href="/ncurses/announce-4.0.html">(24 December
 1996)</a>. A change in version numbering addresses a problem
 exposed by David Engel's <code>ld.so.1.8.5</code> program for
 Linux.
 </li>

<li>
 1.9.9g <a href=
 "/ncurses/ncurses-license.html#patch_961130">(1 December
 1996)</a>.
 </li>

<li>
 <a href=
 "/ncurses/ncurses-license.html#ncurses_1_9_9e">1.9.9e</a> is
 broken. A last-minute/untested change causes forms and menus
 to not refresh. Foreground/background colors are combined
 incorrectly, working properly only on a black background.
 </li>

<li>
 1.9.4 is the oldest version that you should consider
 installing. It came with Slackware 3.0; earlier releases had
 a number of problems, including incompatible terminal
 descriptions.
 </li>
 </ul>

<h2 id="related_programs-id"><a name="related_programs" id=
 "related_programs">What other programs are there?</a></h2>

The ncurses distribution only includes programs that must be
 maintained with it, since they rely on internal details of the
 library. Here are pointers to more generic curses programs, which
 can be built with ncurses or similar implementations:

<ul>
 <li><a href="/cdk/cdk.html">cdk</a> Curses Development Kit (a
 library of widgets)</li>

<li><a href="/dialog/dialog.html">dialog</a> Script-driven
 curses widgets</li>
 </ul>

<hr>

<h2 id="known_problems-id"><a name="known_problems" id=
 "known_problems">Known/Frequent problems</a></h2>

<ul>
 <li>
 <a href="#build_and_install">Building and Installing
 NCURSES</a>

<ul>
 <li><a href="#running_tests">How do I run the
 test-programs?</a></li>

<li><a href="#applying_patches">How do I apply patches?</a></li>

<li><a href="#where_patches">Where are the patches?</a></li>

<li><a href="#big_terminfo">The terminfo database is
 big&mdash;do I need all of that?</a></li>

<li><a href="#no_terminfo">Do I really need a terminfo
 database?</a></li>

<li><a href="#which_terminfo">Which terminfo database do I
 need?</a></li>

<li><a href="#compatible_tic">Is ncurses terminfo
 compatible with my system?</a></li>

<li><a href="#compatible_lib">Is the ncurses library
 compatible with my system?</a></li>

<li><a href="#remote_shells">Problems with remote
 shells</a></li>

<li><a href="#cross_compiles">Problems with
 Cross-Compiling</a></li>
 </ul>
 </li>

<li>
 <a href="#problems_configuring">Configuring NCURSES</a>

<ul>
 <li><a href="#config_small">Do I need all of those programs
 and libraries?</a></li>

<li><a href="#config_cplus">Do I need the C++ binding?</a></li>

<li><a href="#config_linux">Why does make
 menuconfig fail?</a></li>

<li><a href="#config_gcc296">Problem building with gcc
 2.96.x</a></li>

<li><a href="#config_gcc3x">Problem building C++ interface
 with gcc 3.x</a></li>

<li><a href="#config_leaks">Testing for Memory Leaks</a></li>
 </ul>
 </li>

<li>
 <a href="#problems_customizing">Customizing NCURSES</a>

<ul>
 <li><a href="#home_terminfo">How do I set up a private
 terminfo database?</a></li>

<li><a href="#unknown_term">My terminal is not
 recognized</a></li>

<li><a href="#extended_term">Why use terminfo instead of
 extensible termcap?</a></li>

<li><a href="#other_extended_term">What implementations of
 terminfo are extended?</a></li>
 </ul>
 </li>

<li>
 <a href="#problems_performance">Making NCURSES Fast</a>

<ul>
 <li><a href="#hash_scrolling">Why does my program scroll
 slowly?</a></li>

<li><a href="#no_padding">Why does VT100 refresh
 slowly?</a></li>
 </ul>
 </li>

<li>
 <a href="#problems_coloring">Making Color Work</a>

<ul>
 <li><a href="#vt100_color">How do I get color with
 VT100?</a></li>

<li><a href="#no_color">My terminal doesn't recognize
 color</a></li>

<li><a href="#bce_mismatches">My terminal shows some
 uncolored spaces</a></li>

<li><a href="#no_xterm_color">My xterm program doesn't
 recognize color</a></li>

<li><a href="#no_colorterm">Why doesn't ncurses use
 <code>$COLORTERM</code>?</a></li>

<li><a href="#xterm_color">Why not just use TERM set to
 "xterm-color"?</a></li>

<li><a href="#xterm_generic">Why not just use TERM set to
 "xterm"?</a></li>

<li><a href="#xterm_256color">Why not make "xterm" equated
 to "xterm-256color"?</a></li>

<li><a href="#xterm_16MegaColors">Why only 16 (or 256)
 colors?</a></li>

<li><a href="#white_black">Ncurses resets my colors to
 white/black</a></li>

<li><a href="#interchanged_colors">Why are red/blue
 interchanged?</a></li>
 </ul>
 </li>

<li>
 <a href="#problems_display">Other Display Problems</a>

<ul>
 <li><a href="#no_line_drawing">Line-drawing characters come
 out as x's and q's</a></li>

<li><a href="#bad_line_drawing">Line-drawing characters
 come out as Latin-1 characters</a></li>

<li><a href="#utf8_line_drawing">Line-drawing characters do
 not appear</a></li>

<li><a href="#emacs_cvvis">Why does my cursor blink in
 Emacs?</a></li>

<li><a href="#multithread">Why does (fill in the blank)
 happen when I use two threads?</a></li>
 </ul>
 </li>

<li>
 <a href="#problems_keyboard">Keyboard Problems</a>

<ul>
 <li><a href="#cursor_appmode">My cursor keys do not
 work</a></li>

<li><a href="#modified_keys">How can I use shift- or
 control-modifiers?</a></li>

<li><a href="#visible_keys">How can I see what my keyboard
 sends?</a></li>
 </ul>
 </li>

<li>
 <a href="#problems_hardware">Using Hardware (Real)
 Terminals</a>

<ul>
 <li><a href="#reset_logout">Why does reset log me out?</a></li>

<li><a href="#need_padding">Why do I get trash on the
 screen?</a></li>
 </ul>
 </li>

<li>
 <a href="#problems_programs">Interaction with Other
 Programs</a>

<ul>
 <li><a href="#handle_piping">Redirecting I/O to/from a
 Curses application</a></li>

<li><a href="#readline_library">What about readline?</a></li>

<li><a href="#setvbuf_once">Problems with Output
 Buffering</a></li>
 </ul>
 </li>

<li>
 <a href="#problems_xwindows">Mice and Windows</a>

<ul>
 <li><a href="#xterm_paste">I can't cut/paste in xterm</a></li>

<li><a href="#handle_resize">Handling SIGWINCH (resize
 events)</a></li>

<li><a href="#using_gpm_lib">Linking with GPM (Linux
 console mouse library)</a></li>
 </ul>
 </li>

<li>
 <a href="#problems_packages">Problems with Packages</a>

<ul>
 <li><a href="#packages_buggy">Comments on packages</a></li>

<li><a href="#version4_or_not">Ncurses 4/4.2/5.0</a></li>

<li><a href="#colorfgbg_bug">rxvt's <code>$COLORFGBG</code>
 variable</a></li>

<li><a href="#no_urxvt">Where is rxvt-unicode?</a></li>
 </ul>
 </li>
 </ul>

<h3 id="build_and_install-id"><a name="build_and_install" id=
 "build_and_install">Building and Installing NCURSES</a></h3>

<h4 id="running_tests-id"><a name="running_tests" id=
 "running_tests">How do I run the test-programs?</a></h4>

You must first install the terminfo data (i.e., "make
 install.data").

On many systems (those that have a SVr4 curses installed) you
 can run the test programs using the vendor's terminfo database
 (e.g., Solaris, IRIX, HP-UX) by setting the TERMINFO variable to
 point to that instead.

<h4 id="applying_patches-id"><a name="applying_patches" id=
 "applying_patches">How do I apply patches?</a></h4>

See also <a href="#patch_summary">How are patches
 organized?</a>

I name development patches after the base release, with the
 patch date originally (in 1996) in the form yymmdd and
 starting with the year 2000 yyyymmdd.

<ul>
 <li>
 you need patch 2.1 (originally, I believe, in the X
 distribution, but also from GNU now, e.g., ftp.gnu.org).
 Version 2.5 offers some advantages (handles lines wider than
 1024 characters), but introduces new bugs (its name
 resolution algorithm gets more easily confused by duplicate
 names in the tree). The interim versions (2.2, 2.3, 2.4) are
 unacceptable.
 </li>

<li>
 you also need gzip 1.2.4, since both the base release and
 the patches are gzip'd.
 </li>

<li>
 Assuming you have all of the patches and the 4.2 tar.gz
 file in the same directory and are running in Bourne
 shell:

zcat&nbsp;ncurses-4.2.tar.gz&nbsp;|tar&nbsp;xf&nbsp;- 
 cd&nbsp;ncurses-4.2 
 for&nbsp;n&nbsp;in&nbsp;../ncurses-4.2-*.gz&nbsp;;&nbsp;do&nbsp;zcat&nbsp;$n&nbsp;|&nbsp;patch&nbsp;-p1&nbsp;;&nbsp;done

</blockquote>
 </li>
 </ul>

<h4 id="where_patches-id"><a name="where_patches" id=
 "where_patches">Where are the patches?</a></h4>

Development patches (and a rollup patch updated periodically)
 are available for download: <a href=
 "ftp://ftp.invisible-island.net/ncurses/6.2/">patches to 6.2</a>
 Look for a file named something like
 "<code>patch-6.2-yyyymmdd.sh.gz</code>" The
 "yyyymmdd" part corresponds to the patch-date.

After providing a rollup patch, I remove the development
 patches to reduce clutter. Beginning in May 2013, I modified the
 process to provide all of the development patches since the
 release in "dev-patches.zip".

There are also a few rollup patches between releases:

Download:

<ul>
 <li><a href=
 "ftp://ftp.invisible-island.net/ncurses/patches/">patches</a></li>

<li><a href=
 "ftp://ftp.invisible-island.net/ncurses/patches/patch-4.2-5.0.sh">
 patch-4.2-5.0.sh</a></li>

<li><a href=
 "ftp://ftp.invisible-island.net/ncurses/patches/patch-5.0-5.1.sh">
 patch-5.0-5.1.sh</a></li>

<li><a href=
 "ftp://ftp.invisible-island.net/ncurses/patches/patch-5.1-5.2.sh">
 patch-5.1-5.2.sh</a></li>

<li><a href=
 "ftp://ftp.invisible-island.net/ncurses/patches/patch-5.2-5.3.sh.gz">
 patch-5.2-5.3.sh.gz</a></li>

<li><a href=
 "ftp://ftp.invisible-island.net/ncurses/patches/patch-5.3-5.4.sh.gz">
 patch-5.3-5.4.sh.gz</a></li>

<li><a href=
 "ftp://ftp.invisible-island.net/ncurses/patches/patch-5.4-5.5.sh">
 patch-5.4-5.5.sh</a></li>

<li><a href=
 "ftp://ftp.invisible-island.net/ncurses/patches/patch-5.5-5.6.sh.gz">
 patch-5.5-5.6.sh.gz</a></li>

<li><a href=
 "ftp://ftp.invisible-island.net/ncurses/patches/patch-5.6-5.7.sh.gz">
 patch-5.6-5.7.sh.gz</a></li>

<li><a href=
 "ftp://ftp.invisible-island.net/ncurses/patches/patch-5.7-5.8.sh.gz">
 patch-5.7-5.8.sh.gz</a></li>

<li><a href=
 "ftp://ftp.invisible-island.net/ncurses/patches/patch-5.8-5.9.sh.gz">
 patch-5.8-5.9.sh.gz</a></li>

<li><a href=
 "ftp://ftp.invisible-island.net/ncurses/patches/patch-5.9-6.0.sh.gz">
 patch-5.9-6.0.sh.gz</a></li>

<li><a href=
 "ftp://ftp.invisible-island.net/ncurses/patches/patch-6.0-6.1.sh.gz">
 patch-6.0-6.1.sh.gz</a></li>

<li><a href=
 "ftp://ftp.invisible-island.net/ncurses/patches/patch-6.1-6.2.sh.gz">
 patch-6.1-6.2.sh.gz</a></li>
 </ul>

The rollup patches include all patches through the cited
 version. You must apply them to the base release, e.g.,

zcat&nbsp;ncurses-4.1.tar.gz&nbsp;|tar&nbsp;xf&nbsp;- 
 cd&nbsp;ncurses-4.1 
 sh&nbsp;../patch-4.1-4.2.sh 
 </blockquote>

I have tested the rollup patches with patch 2.1 and 2.5,
 adjusting for their respective limitations.

<h4 id="big_terminfo-id"><a name="big_terminfo" id=
 "big_terminfo">The terminfo database is big&mdash;do I need all
 of that?</a></h4>

Not at all. You can load a subset of the terminfo database. I
 use a variant of this script to load the terminal descriptions
 that I need on my machine:

#!/bin/sh 
 #&nbsp;uses&nbsp;the&nbsp;-e&nbsp;switch&nbsp;on&nbsp;tic&nbsp;to&nbsp;selectively&nbsp;load&nbsp;descriptions&nbsp;that&nbsp;are

#&nbsp;interesting&nbsp;for&nbsp;my&nbsp;testing.

if&nbsp;test&nbsp;-f&nbsp;terminfo.src&nbsp;;&nbsp;then

&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;TMP=/tmp/load$$ 
 &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;trap&nbsp;"rm&nbsp;-f&nbsp;$TMP"&nbsp;0&nbsp;1&nbsp;2&nbsp;5&nbsp;15 
 &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;tic&nbsp;-s&nbsp;-1&nbsp;-I&nbsp;-e'

ansi,&nbsp;ansi-m,&nbsp;color_xterm,&nbsp;ecma+color, 
 klone+acs,&nbsp;klone+color,&nbsp;klone+sgr,&nbsp;klone+sgr-dumb,

linux,&nbsp;pcansi-m,&nbsp;rxvt,&nbsp;vt52, 
 vt100,&nbsp;vt102,&nbsp;vt220,&nbsp;xterm'&nbsp;terminfo.src&nbsp;&gt;$TMP

&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;tic&nbsp;-s&nbsp;$TMP

else 
 &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;echo&nbsp;'oops' 
 &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;exit&nbsp;1 
 fi 
 
 </blockquote>

<h4 id="no_terminfo-id"><a name="no_terminfo" id="no_terminfo">Do
 I really need a terminfo database?</a></h4>

You could compile-in fallback definitions for the most
 commonly used terminal types. To do this, you must already have
 <code>infocmp</code> installed (note that ncurses 5.0 infocmp's
 support for fallback descriptions is done differently from
 ncurses 4.2).

But fallback definitions are really only useful in embedded
 applications, where no external files are wanted.

<h4 id="which_terminfo-id"><a name="which_terminfo" id=
 "which_terminfo">Which terminfo database do I need?</a></h4>

The most reliable terminfo database is that distributed with
 ncurses 5.0, or via followup development patches. The original
 process of incorporating terminal descriptions from various
 sources corrects some errors in the originals, but introduces
 others (either translation errors, or misconceptions). Besides
 working to resolve these, from time to time we incorporate new
 sources.

As noted in 29 February 2004, the terminfo database at
 <a href="http://www.catb.org/~esr/terminfo/">http://www.catb.org/~esr/terminfo/</a>
 does not appear to be actively maintained. Since the release of
 ncurses 5.0 in late 1999, there are numerous fixes which are not
 in that database.

As of January 2012, this is still true. A check with

infocmp&nbsp;-x&nbsp;-F&nbsp;termtypes.master&nbsp;terminfo.src

</blockquote>

reports that there are 1807 differences for the entries which
 the two have in common, as well as 325 entries not in the
 "master" terminfo database.

Here are links to current versions:

<ul>
 <li><a href=
 "/datafiles/current/terminfo.src.gz">terminfo.src.gz</a> 
 </li>

<li><a href=
 "/datafiles/current/termcap.src.gz">termcap.src.gz</a></li>
 </ul>

If you choose to not install the ncurses terminfo database, we
 have found that the SVr4 systems (Solaris, IRIX&nbsp;6,
 OpenServer and HP-UX&nbsp;10) work well enough for many purposes.
 Other systems either are not binary compatible, are incomplete,
 or contain more errors than either of the choices mentioned. Some
 that are not binary compatible can be accommodated by configuring
 ncurses to use the native terminfo format. These include AIX,
 HP-UX&nbsp;11, Tru64 and U/Win.

<h4 id="compatible_tic-id"><a name="compatible_tic" id=
 "compatible_tic">Is ncurses terminfo compatible with my
 system?</a></h4>

Usually.

Terminfo is compiled into binary form, with booleans, numbers
 and strings in arrays. As long as the array items line up, and
 the headers (that tell how long the arrays are) are compatible,
 ncurses and your vendor's system can each use the same terminfo
 database. Older sytems (e.g,. those based on SVr3) implement a
 subset of the SVr4 terminfo. For example, HP-UX 9 is
 &ldquo;compatible&rdquo; up to the entries that would describe
 graphic (box) characters. There it diverges.

Other systems (e.g., Digital Unix 4.x and the older AIX 3.2.5)
 use different formats and are not compatible.

But those are (very) old systems. If you are not running a
 computer museum, you may not even recognize the names of these
 systems.

Even for very old systems, however, terminfo source
 is compatible and can be compiled using the appropriate tool
 (usually tic). Some terminfo descriptions may produce
 warnings (e.g., the memu/meml capabilities in the standard xterm
 distribution), but the tools compile what they recognize and warn
 about the rest.

The ncurses tic program recognizes a wider range of
 input than other terminfo compilers, including extensions
 coordinated with infocmp that make it easier to use:

<ul>
 <li>
 compile descriptions that use the long names that
 &ldquo;infocmp&nbsp;-L&rdquo; produces.
 </li>

<li>
 the &ldquo;infocmp&nbsp;-f&rdquo; option formats complex
 expressions into indented format. These are usable with the
 standard SVr4 tic program, though not with ncurses 4.0 and
 below due to a bug.
 </li>
 </ul>

More important, ncurses tic allows new terminal
 capabilities to be defined, by the
 <code>-x</code> option. Rather than warning
 about unrecognized capabilities and (otherwise) ignoring them,
 ncurses tic will add those to the end of the compiled
 terminfo file. Those additions (referred to as user-defined
 capabilities) are hidden from the older SVr4 tic
 programs because they see only the part described by the file's
 header.

The older SVr4 tic, of course, is unchanging.
 Development (and maintenance) stopped long ago. But ncurses
 development is still (as of 2020) ongoing. Occasionally some bug
 is found which affects the terminal database:

<ul>
 <li><a href="announce-5.7.html#bug-fixes">ncurses 5.7</a> fixed
 a case where tic could have an infinite loop if it
 encountered a cancelled user-defined capability.</li>

<li><a href="announce-6.2.html#h3-bug-fixes">ncurses 6.2</a>
 fixed a case where forward-references to a user-defined
 capability, e.g., via a &ldquo;use=&rdquo; clause, were
 mishandled.</li>
 </ul>

In each case, the tool was improved, but older versions cannot
 handle the newer terminfo source. If you must use an older
 version of ncurses, use the terminfo source which was distributed
 with that version of ncurses.

If you are using ncurses terminfo with another implementation
 of tic, the advice is different:

<ul>
 <li id="ncurses61-numbers"><a href=
 "announce-6.1.html#h4-config-major">ncurses 6.1</a> introduced
 support for large number capabilities, e.g., for more than
 32767 color pairs. Other implementations generally treated
 out-of-range values as zero. This was addressed two years later
 in NetBSD's tic program, which may be in NetBSD 10. As
 of June 2020, the current release is <a href=
 "ncurses-netbsd.html#netbsd9-performance">NetBSD 9</a>.</li>

<li>NetBSD's tic program is known to evaluate <a href=
 "ncurses-netbsd.html#netbsd8-cancels">cancelled</a>
 capabilities differently from SVr4 and ncurses.</li>
 </ul>

<h4 id="compatible_lib-id"><a name="compatible_lib" id=
 "compatible_lib">Is the ncurses library compatible with my
 system?</a></h4>

Yes/no. 
 Source compatible yes, binary compatible no.

You cannot simply replace your existing curses or termcap
 library with ncurses unless you are prepared to recompile
 applications that use the curses or termcap library (e.g.,
 <code>vi</code>, <code>telnet</code>).

For systems using the Linux kernel, that is feasible. But not
 Solaris or other proprietary systems. For those, I recommend
 configuring with the <code>--disable-overwrite</code> option.
 This directs the configure script to install the library so you
 would link it as <code>-lncurses</code>, not adding a symbolic
 link to make it link as <code>-lcurses</code>.

The <code>--disable-overwrite</code> option also installs the
 header files such as <code>curses.h</code> in a subdirectory,
 e.g., <code>/usr/local/include/ncurses/curses.h</code>, thereby
 avoiding a (mis)feature of recent versions of <code>gcc</code>
 which look first in <code>/usr/local/include</code> for header
 files. Since the vendor's compilers do not do this, a common
 problem results: compiling with
 <code>/usr/include/curses.h</code> and linking with
 <code>/usr/local/lib/libcurses.a</code>.

Starting with ncurses 5.3, the behavior for this option
 changed. If you do not install into <code>/usr</code>, the
 configure script will assume you do not wish to overwrite the
 existing version of curses. Configure scripts which do not check
 for ncurses headers in both locations are incorrect anyway. They
 can be accommodated by setting the <code>$CPPFLAGS</code>
 environment variable, e.g.,

setenv&nbsp;CPPFLAGS&nbsp;"-I/usr/local/include/ncurses" 
 
 </blockquote>

<h4><a name="remote_shells" id="remote_shells">Problems with
 remote shells</a></h4>

Some users notice that remote machines do not have the same
 set of terminal descriptions as their local machine. Running a
 remote shell on those machines can be a problem because typical
 <code>ssh</code> configurations export the
 <code>TERM</code> variable to the remote
 machine.

Developers who work on multiple machines should be used to
 this problem. It is not new: the various operating systems (Unix,
 BSDs, MacOS and those using the ncurses terminal database) have
 different features, different versions.

One interesting quirk is using the <code>screen</code>
 program. The manual page notes:

<blockquote class="code-block">
 When screen tries to figure out a terminal
 name for itself, it first looks for an entry named
 "screen.&lt;term&gt;", where &lt;term&gt; is
 the contents of your $TERM variable. If no such entry
 exists, screen tries "screen" (or
 "screen-w" if the terminal is wide (132 cols or
 more)). If even this entry cannot be found, "vt100" is
 used as a substitute.
 </blockquote>

Although <code>screen</code>'s mapping to different terminals
 is less than perfect, it is possible to improve the mapping by
 providing customized "screen.&lt;term&gt;" entries. The
 ncurses terminal database has done that since <a href=
 "terminfo.src.html#t20010331">early 2001</a>.

As along as the local and remote terminal databases are
 complete and up-to-date, things go well. When they are not, users
 encounter problems such as in

<blockquote>
 <a href=
 "https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=854414">Debian
 #854414</a> &mdash; screen: after sshing, some commands give
 error "Error opening terminal: screen.xterm-256color."
 
 </blockquote>

The discussion is rather long, with as they say, something for
 everyone. The resolution (adding documentation) does not really
 solve the problem, since the <code>screen</code> manual page has
 been available for some time.

Just to keep things interesting, ncurses 6.1 provides a new
 way to solve the problem. The <code>infocmp</code> program can
 print a terminal description in a a single line, which when
 assigned to the <code>TERMINFO</code> variable lets the library
 read the terminal description directly from the variable.
 Old-time developers who are knowledgeable about termcap might see
 a similarity, but that changes with a closer look:

<ul>
 <li>the terminfo information can be written to a single line
 (eliminating one of the problems with shell variable
 assignments), and</li>

<li>the terminfo description is encoded either in base-64 or
 hexadecimal (allowing the binary data to be assigned
 safely).</li>
 </ul>

This works with ncurses 6.1:

<blockquote>
 <pre class="code-block">
TERMINFO=$(infocmp -0qQ1) ssh remote
</pre>
 </blockquote>

and if the <code>TERMINFO</code> variable has been added to
 the list of environment variables in the <code>ssh</code> and
 <code>sshd</code>, then ncurses 6.1 applications on the remote
 system will read the terminfo description from the environment
 variables. Applying the changed <code>TERMINFO</code> to the
 single <code>ssh</code> process is preferable to
 blindly setting it in your shell's initialization scripts.

That would not work with systems that do not use ncurses 6.1,
 but unless the <code>TERMINFO</code> variable is accepted in
 their respective <code>sshd</code> configurations, they will not
 be confused by this extension.

For those curious about other libraries, slang will ignore the
 &ldquo;broken&rdquo; <code>$TERMINFO</code> and rely upon its
 <a href="ncurses-slang.html#env_TERMINFO_DIRS">compiled-in
 terminfo list</a>. Other problems can of course occur.

<h4 id="cross_compiles-id"><a name="cross_compiles" id=
 "cross_compiles">Problems with Cross-Compiling</a></h4>

I have done occasional cross-compiles of ncurses since 2003,
 using DJGPP as a target, and recently (writing this in 2010) some
 MinGW builds. For example

TARGET=mingw32&nbsp; 
 configure&nbsp;\ 
 &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;--with-build-cc=gcc&nbsp;\

&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;--host=$TARGET&nbsp;\

&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;--target=$TARGET

make 
 </blockquote>

Cross-compiles of ncurses require compiling utilities which
 are then executed at build time. The utilities are wrappers
 around ncurses library functions, which means that some special
 #define's may be needed to compile them. In particular (until May
 2010) cross-compiling the wide-character library required adding
 something like this:

--with-build-cppflags=-D_XOPEN_SOURCE 
 
 </blockquote>

Ncurses 5.7 introduced a separate problem. The terminal
 database distributed with it uses a feature that causes older
 versions of <a href="/ncurses/man/tic.1m.html">tic</a> to hang.
 If you are cross-compiling, the database is installed using your
 system's copy of tic. You can work around the problem by either
 updating ncurses (5.7 was released in 2008), or substituting an
 older version of <code>misc/terminfo.src</code> in the build
 tree.

<h3 id="problems_configuring-id"><a name="problems_configuring"
 id="problems_configuring">Configuring NCURSES</a></h3>

<h4 id="config_small-id"><a name="config_small" id=
 "config_small">Do I need all of those programs and libraries?</a></h4>

Well, I use them...

Wrong answer. 
 You may need only the ncurses library, or even just the terminfo
 database. The top-level Makefile in the build tree is designed to
 let you install various combinations according to your
 requirements. But there are a few constraints:

<ul>
 <li>
 You can install only the terminfo database by typing "make
 install.data". But you must have a copy of <code>tic</code>
 built. If you chose to build with shared libraries, you
 should first install the libraries and programs, e.g., with
 "make install.libs install.progs". Otherwise the shared
 libraries will not be found properly for the terminfo
 compiler to run.
 </li>

<li>
 The version numbers attached to the shared libraries do
 mean something. Do not rename them (or link other versions to
 them). 
 Version 5 is not compatible with version 4. 
 Version 4 is not compatible with version 3. 
 We have made corrections at each release to match the X/Open
 Curses interface specification. I have seen as many people
 complaining about mysterious problems with ncurses as I have
 seen people advising others to save time/space by linking
 them.
 </li>
 </ul>

<h4 id="config_cplus-id"><a name="config_cplus" id=
 "config_cplus">Do I need the C++ binding?</a></h4>

You may not need the C++ binding for ncurses. However, you
 should configure ncurses with C++ if it is available on your
 system. Otherwise, you will not be able to compile ncurses
 applications with the C++ compiler.

This point is not clear in the INSTALLATION instructions,
 having been thought too obvious to dwell upon. However, some
 distributors have "customized" ncurses, omitting the C++ binding
 to save space (or the time to issue separate "make install"
 commands for the components which they really need).

The problem is this:

<ul>
 <li>
 Both ncurses and C++ declare the bool type. It is
 part of each of their respective standards.
 </li>

<li>
 There is no standard that says what the size of
 bool is.
 </li>

<li>
 The ncurses interface uses bool.
 </li>
 </ul>

The ncurses configure script determines the actual size used
 for bool by the C++ compiler on your system. If you
 suppress this configuration check, the default size for
 bool is not guaranteed to work with your compiler.

With 5.0, the configure script provides two options
 (<code>--without-cxx</code> and
 <code>--without-cxx-binding</code>). Use the former to suppress
 the configure checks for the C++ compiler, e.g., when there is no
 working C++ compiler on your system. Use the latter to omit the
 C++ binding, if you must.

<h4 id="config_linux-id"><a name="config_linux" id=
 "config_linux">Why does make menuconfig fail?</a></h4>

I can only guess (people having trouble in this area generally
 do not answer email ;-)

The Linux "make menuconfig" attempts to build a customized
 dialog program called <code>lxdialog</code>. This uses ncurses,
 which of course is why you are reading this question/answer.

<ul>
 <li>
 The development libraries or header files for ncurses may
 not be installed.
 </li>

<li>
 The libraries may be installed in an unusual place. But in
 any case, /usr/src/linux/scripts/Makefile is simple
 to read, to see where <code>make</code> looks.
 </li>

<li>
 You may have attempted to upgrade your C library, and did
 not complete the job by upgrading libraries that depend upon
 it.
 </li>
 </ul>

<h4 id="config_gcc296-id"><a name="config_gcc296" id=
 "config_gcc296">Problem building with gcc 2.96.x</a></h4>

The C preprocessor in gcc 2.96 is broken. In particular, the C
 preprocessor has multiple bugs including

<ul>
 <li>
 does not handle whitespace correctly (breaks ncurses).
 </li>

<li>
 cannot compile its own preprocessor output (breaks
 lynx).
 </li>
 </ul>

These were apparently fixed in gcc 3.0.4 (do not waste time
 with gcc 3.0).

<h4 id="config_gcc3x-id"><a name="config_gcc3x" id=
 "config_gcc3x">Problem building C++ interface with gcc 3.x</a></h4>

Since releasing ncurses 5.2, this has been the most frequent
 reason for referring people to the <a href=
 "#where_patches">rollup patch</a>. The problem is that the C++
 bindings to C functions such as <code>vsscanf()</code> were
 altered (more than once, apparently as an afterthought) in
 preparing the gcc releases. Since the gcc changelog does not cite
 these changes except obliquely, and they are undocumented, it is
 possible that they may change again.

<h4 id="config_leaks-id"><a name="config_leaks" id=
 "config_leaks">Testing for Memory Leaks</a></h4>

Perhaps you used a tool such as <code>dmalloc</code> or
 <code>valgrind</code> to check for memory leaks. It will normally
 report a lot of memory still in use. That is normal.

The ncurses configure script has an option,
 <code>--disable-leaks</code>, which you can use to continue the
 analysis. It tells ncurses to free memory if possible. However,
 most of the in-use memory is &ldquo;permanent&rdquo; (normally
 not freed).

Any implementation of curses must not free the memory
 associated with a screen, since (even after calling
 <code>endwin()</code>), it must be available for use in the next
 call to <code>refresh()</code>. There are also chunks of memory
 held for performance reasons. That makes it hard to analyze
 curses applications for memory leaks. To work around this, build
 a debugging version of the ncurses library which frees those
 chunks which it can, and provides the <a href=
 "/ncurses/man/curs_memleaks.3x.html"><code>exit_curses()</code></a>
 function to free the remainder on exit. The ncurses utility and
 test programs use this feature, e.g., via the
 <code>ExitProgram()</code> macro.

<h3 id="problems_customizing-id"><a name="problems_customizing"
 id="problems_customizing">Customizing NCURSES</a></h3>

<h4 id="home_terminfo-id"><a name="home_terminfo" id=
 "home_terminfo">How do I set up a private terminfo database?</a></h4>

If you must maintain your own terminfo database, SVr4 curses
 and ncurses both use the $TERMINFO variable to override the
 standard location of the terminfo database. Ncurses also provides
 two related extensions: the $HOME/.terminfo directory and the
 $TERMINFO_DIRS search path.

Though ncurses tests $TERMINFO first, otherwise it reads from
 $HOME/.terminfo before the standard location, and writes to that
 location after failing in other places. This design is a
 compromise which is made more complicated if you have configured
 ncurses with the --enable-termcap and --enable-getcap-cache
 options. Unless you are prepared to deal with the hidden
 conflicts, you should simply remove the $HOME/.terminfo
 directory.

If you do not wish to use $HOME/.terminfo (and are not able to
 replace ncurses on your system), ncurses 4.2 and later work
 properly if you replace that directory with a file so it cannot
 write terminfo entries which would conflict with the standard
 location.

The <a href="/ncurses/man/toe.1m.html">toe</a> program can
 show a side-by-side comparison of terminal databases, making it
 simple to see conflicting entries from your private terminal
 database.

As noted, ncurses also provides the $TERMINFO_DIRS extension.
 Unlike $HOME/.terminfo, this allows you to specify one or more
 locations for the terminal database. When reading a terminal
 description, ncurses chooses the first one found. Using this
 feature, you can tell ncurses to look in your own database(s) as
 well as one or more system-defined locations.

The usual reason for creating a private terminal database is
 to work around inability to change the system's terminal
 database. Setting $TERMINFO suffices for most users, and happens
 to work with other implementations than ncurses. However, for
 non-ncurses implementations, it limits the user's environment to
 just the terminal database indicated by the $TERMINFO
 variable.

As noted in "<a href="#which_terminfo">Which terminfo database
 do I need?</a>", there are a few older systems whose compiled
 terminfo files differ slightly from the SVr4 layout used by
 ncurses. While ncurses can be compiled to match the system's
 format (I do this), it seems to be a less-used feature. Packagers
 prefer the simpler approach of letting ncurses use its database
 and leaving the system applications alone. There are then these
 cases for constructing private databases in an NFS-mounted home
 directory, shared across different operating system types:

<ul>
 <li>
 the system's terminfo format is SVr4-compatible.

<ul>
 <li>
 Use the $TERMINFO variable, e.g., to have all
 applications (ncurses and otherwise) use the same
 terminal descriptions.
 </li>

<li>
 Use the $TERMINFO_DIRS variable to make ncurses
 applications see a distinct set of terminal descriptions
 from the system, e.g., for "xterm".
 </li>
 </ul>
 </li>

<li>
 the system's terminfo format is not SVr4-compatible (AIX
 for instance).

<ul>
 <li>
 if ncurses is compiled to match the system's terminfo
 format

<ul>
 <li>
 as before, you can set either $TERMINFO or
 $TERMINFO_DIRS according to whether you want to use
 the same definitions for all applications (ncurses
 and system).
 </li>

<li>
 choose a different location for the database,
 e.g., $HOME/.terminfo-aix to avoid conflict with
 other platforms.
 </li>
 </ul>
 </li>

<li>
 if ncurses is not compiled to match the
 system's terminfo format, then it is likely that ncurses
 and the system curses/terminfo libraries cannot read the
 other's compiled terminfo files.

<ul>
 <li>
 You can set $TERMINFO_DIRS to tell ncurses to look
 at its own terminal database; the system libraries
 will not look there.
 </li>

<li>
 Do not set $TERMINFO, since this will cause both
 to look in the same place.
 </li>
 </ul>
 </li>
 </ul>
 </li>
 </ul>

A good place to start when customizing a shell initialization
 script for your private terminal databases is of course
 <code>uname</code>, since it is provided on the systems which are
 relevant to this topic, and its output can tell which system type
 is used. Alternatively (if your environment happens to have an
 inconsistent mixture of ncurses packages), the hostname is
 another place to get useful parameters for the initialization
 script.

As a rule, these settings should be done in the shell's
 login script rather than the one which is run on each
 shell initialization.

<h4 id="unknown_term-id"><a name="unknown_term" id=
 "unknown_term">My terminal is not recognized</a></h4>

Usually this happens because you have not installed the
 <code>terminfo</code> database, or it is not in the proper
 location. If you do not, and (if ncurses happens to be configured
 to provide termcap support using the "--enable-getcap-cache"
 configure option) the application is unable to locate the
 terminfo database, the ncurses library will attempt to recover by
 reading <code>/etc/termcap</code>, translating it into a <a href=
 "#home_terminfo">private terminfo database</a>, i.e., a
 directory:

$HOME/.terminfo 
 
 </blockquote>

This directory can be a nuisance, because the termcap file
 often does not contain complete or consistent terminal
 descriptions. Remove it and correct the problem (i.e., install
 the <code>terminfo</code> database). Better yet, do not enable
 the feature (it has been disabled by default since <a href=
 "NEWS.html#t961214">late 1996</a>).

You (or the person who configured ncurses) may have installed
 terminfo in the wrong, or an obsolete location:

<ul>
 <li>
 On Linux-based systems, as well as BSD variants such as
 FreeBSD and NetBSD the preferred location for terminfo is
 <code>/usr/share/terminfo</code>, with a symbolic link from
 <code>/usr/lib/terminfo</code> for compatibility with older
 applications. Prior to 1.9.9g, the <code>configure</code>
 script defaulted to a /usr/local prefix rather than /usr,
 necessitating an explicit

configure&nbsp;--prefix=/usr 
 </blockquote>
 </li>

<li>
 On other systems, the prefix defaults to /usr/local. You
 may wish to change this. Some users install ncurses replacing
 other versions of curses altogether. I don't do this. If you
 do, read the INSTALL file.
 </li>

<li>
 Some packagers have gone further, configuring ncurses for
 a given terminfo location (which is not the system default),
 and then treating ncurses's terminfo database as "optional".
 You can always recover in this case by setting the
 <code>TERMINFO</code> environment variable to point to the
 real terminfo database.
 </li>

<li>
 Finally, prior to 1.9.9g, the actual terminfo directory
 resided in <code>/usr/lib/terminfo</code>. It was moved to
 <code>/usr/share/terminfo</code> to conform to file-system
 standards. The installation script attempts to determine if
 you have a real directory at the old location, and will not
 delete that (because some applications, such as
 <code>mc</code> and <code>screen</code> install special
 terminal descriptions that you would not want to accidentally
 delete. The best solution in this case is to move the old
 directory to the new location and install the new terminfo
 data, merging into the old data.
 </li>
 </ul>

<h4 id="extended_term-id"><a name="extended_term" id=
 "extended_term">Why use terminfo instead of extensible
 termcap?</a></h4>

Several reasons:

<ul>
 <li>
 there is no standard for termcap. That is a drawback to
 developing applications using termcap. Amusingly enough, many
 of the termcap capabilities in common use now are defined in
 terms of terminfo, much like the <code>inch</code> is defined
 in terms of a <code>centimeter</code>.
 </li>

<li>
 termcap libraries are not as fast, and have a long history
 of susceptibility to buffer overflows. See the discussion of
 <a href="/ncurses/tctest.html">tctest</a> for example.
 </li>

<li>
 terminfo supports conditional expressions.
 </li>

<li>
 ncurses terminfo is extensible anyway. It has been since
 <a href="#extended-terminfo-1999">1999</a> (release 5.0). If
 you give the <code>-x</code> option to <code>tic</code>, it
 compiles additional information which may be in the terminal
 description, using the syntax to determine types. If
 <code>-x</code> is omitted, <code>tic</code> warns about the
 unrecognized information.
 </li>
 </ul>

As an example of how this extensibility feature can be used,
 consider the case of <a href="http://www.openqm.com/">OpenQM</a>,
 whose developers copied more than 1000 lines of ncurses library
 code into their application to support some extended terminal
 capability features. Here are two patches showing how that can be
 improved by removing the nonstandard features:

<ul>
 <li><a href=
 "ftp://ftp.invisible-island.net/ncurses/patches/qmsrc_2-4-4-060701a.patch.gz">
 Make the syntax standard</a></li>

<li><a href=
 "ftp://ftp.invisible-island.net/ncurses/patches/qmsrc_2-4-4-060701b.patch.gz">
 Remove the unnecessary code</a></li>
 </ul>

<h3><a name="other_extended_term" id="other_extended_term">What
 implementations of terminfo are extended?</a></h3>

This could be a much longer answer, but the FAQ is long. A
 really short, but unsatisfying answer might be just
 &ldquo;ncurses and libraries written to work with its database or
 to imitate it.&rdquo; Here is a compromise:

<ul>
 <li>Unix System V (mid/late-1980s) provided terminfo using a
 predefined set of terminal capabilities, which used 16-bit
 integers.</li>

<li>If you wanted to change any of that, you had to modify the
 curses library and programs. AT&amp;T did that, adding new
 capabilities, in its series of releases culminating in SVr4.
 Some vendors did that as well.</li>

<li>Some marketing people might call that
 &ldquo;extended&rdquo; but programmers would not.</li>

<li>X/Open Curses (early-1990s) documented SVr4's terminfo
 (removing some details). 
 Vendors probably continued to add capabilities.</li>

<li>In <a href="#extended-terminfo-1999">1999</a>, ncurses 5.0
 provided a way to add user-defined capabilities (i.e., no
 program changes were needed).</li>

<li>The ncurses terminal database could be used by existing
 Unix applications.</li>

<li>The user-defined capabilities would not be noticed by
 applications using other implementations of curses.</li>
 </ul>

Of course, termcap, unlike terminfo, is interpreted and users
 could always just add capabilities as needed (although the
 absence of a standard termcap was a drawback). That initial
 implementation addressed most of the supposed advantage of
 termcap over terminfo, except:

<ul>
 <li>A number in termcap was not limited to 16-bits, while
 ncurses continued to use 16-bits for compatibility with other
 implementations.</li>

<li>The ncurses implementation allowed a capability name to be
 only one type (boolean or number or
 string).</li>
 </ul>

A decade later, a few implementations were based on ncurses'
 extended terminfo:

<ul>
 <li><a href="https://github.com/mauke/unibilium">unibilium</a>
 provided applications with a different terminfo reader
 for ncurses' extended terminfo.</li>

<li><a href=
 "http://cvsweb.netbsd.org/bsdweb.cgi/src/lib/libterminfo/?only_with_tag=MAIN">
 NetBSD</a> abandoned termcap in favor of terminfo, and
 developed their own terminfo library and programs. The NetBSD
 implementation includes the same limitations relative to
 termcap as ncurses' initial terminfo extensions.</li>
 </ul>

But there were few applications which used the extended
 information aside from <a href="/xterm/xterm.html">xterm</a> and
 <a href="https://github.com/tmux/tmux">tmux</a>.

So far, that addresses only ncurses' initial implementation.
 But ncurses is still under development.

Extended numbers (32-bits) were provided by ncurses 6.1 early
 in 2018:

<ul>
 <li>There were no credible extant uses of termcap with large
 numbers, anyway, but extended numbers in ncurses were useful
 for a few who wanted to experiment with direct colors.</li>

<li>The required format change for extended numbers briefly
 broke unibilium (as well as slang, which does use
 terminfo).</li>

<li>NetBSD was not affected by that improvement. However
 (writing this more than a year later, in April 2019) some newer
 terminal descriptions would not work with NetBSD curses, if its
 terminal database were up-to-date (it was last updated in
 February 2017). Writing in mid-2020, future releases of
 <a href="#ncurses61-numbers">NetBSD</a> may provide this
 feature.</li>
 </ul>

While developing the extended numbers feature for ncurses 6.1,
 the other limitation (one type per capability name) arose in an
 end-user application:

<ul>
 <li>Knowledgeable terminal developers provide terminfo (or
 termcap) descriptions for their program. 
 Other developers simply copy/paste something that looks
 close.</li>

<li>In this case, someone read the manual page for tmux where
 it describes <a href=
 "http://man7.org/linux/man-pages/man1/tmux.1.html#TERMINFO_EXTENSIONS">
 terminfo extensions</a> which it supports. and did not
 understand that the corresponding terminal description has to
 fill in some string-values. The <code>Se</code> and
 <code>Ss</code> capabilities had no values. The change was
 accepted anyway, late in <a href=
 "https://git.suckless.org/st/commit/06f8cf8ca87a81db15816658c40b2afcd1ad5332.html">
 2016</a>.</li>

<li>Keeping in mind that copy/paste mentality, changes to
 ncurses' terminal database are more closely scrutinized (but
 see <a href=
 "https://github.com/ThomasDickey/ncurses-snapshots/commit/cccd5739099af0656f6cee0b8fb15ccd8f8e0de4#diff-842764d6827ee1f7e8605999bb22e2d6R5837">
 update</a>, which fixed other problems but not this one). In
 this case, the error was overlooked because there was no
 routine testcase for tmux's extensions.</li>

<li>A capability without a value is a boolean value, even if
 tmux would like to use it as a string.</li>

<li>By itself, that would only cause tmux to not find the
 capability. The problem was that some descriptions used the
 capability correctly, and the types differed between
 descriptions while <code>tic</code> compiled the database.</li>

<li>For example, <a href=
 "terminfo.src.html#toc-_S_I_M_P_L_E_T_E_R_M">this one</a> is
 built up from several other descriptions with the
 <code>use=</code> clause. When <code>tic</code> builds the
 result, it must merge those, accounting for differences in the
 lists of extended capabilities.</li>

<li>The initial implementation did not handle that properly;
 <code>tic</code> chose the wrong extended capability in
 building the merged terminal description. Since even a small
 change affects many lines of text, the damage was overlooked at
 first. Subsequent fixes <a href="terminfo.src.html#t20171230">a
 year later</a> overlooked the problem.</li>

<li>The mismatched string value was a problem because it told
 ncurses to expect a parameter when it was used, while
 the application did not expect to provide that (see <a href=
 "https://github.com/tmux/tmux/issues/1264">tmux
 #1264</a>).</li>

<li>Eventually it was reported early in <a href=
 "http://lists.gnu.org/archive/html/bug-ncurses/2018-02/msg00034.html">
 2018</a>, and the terminfo description fixed.</li>
 </ul>

Fixing the terminal description is easy. Improving the tool
 takes more time. Development toward ncurses 6.2 made these
 changes:

<ul>
 <li>adding checks for the conventional extended
 capabilities, to warn when a capability would use an unexpected
 number of parameters, and disallow that setting (<a href=
 "NEWS.html#index-t20190317">20190317</a>).</li>

<li>improving the terminfo-merging to handle differing types
 for extended capability names (<a href=
 "NEWS.html#t20190427">20190427</a>).</li>
 </ul>

Although the problem was fixed in ncurses, users continue to
 recommend using the corrupt terminfo file (for example, <a href=
 "https://github.com/tmux/tmux/issues/1593">tmux #1593</a>). The
 first of those changes to ncurses addresses that source of
 error.

<h3 id="problems_performance-id"><a name="problems_performance"
 id="problems_performance">Making NCURSES Fast</a></h3>

<h4 id="hash_scrolling-id"><a name="hash_scrolling" id=
 "hash_scrolling">Why does my program scroll slowly?</a></h4>

Besides <a href="#need_padding">padding</a> (i.e., time delay)
 information, which may be slowing your application down on a
 terminal emulator, ncurses provides two versions of scrolling
 optimization. The newer/improved version was incompletely tested
 at the time of release of ncurses 4.2, so it was marked
 experimental in the configure script.

The newer scrolling (hashmap) algorithm does not work properly
 in older versions of ncurses. Starting with ncurses 4.2, however,
 we recommend enabling this logic when configuring, using the
 --enable-hashmap option. It is configured by default in ncurses
 5.0

<h4 id="no_padding-id"><a name="no_padding" id="no_padding">Why
 does VT100 refresh slowly?</a></h4>

Some terminal descriptions contain <a href=
 "#need_padding">padding</a> (i.e., time delay) information.
 Ncurses uses this information to slow down the rate at which
 characters are sent to the terminal.

However, the vt100 terminal description, which is one of the
 most widely used (or misused) contains padding information for a
 real DEC VT100. It is not suitable as a replacement for the xterm
 terminal description. (Xterm requires no padding).

If you must use the vt100 terminal
 description, you may consider setting the NCURSES_NO_PADDING
 environment variable which is implemented in current versions of
 ncurses (since late 1998). That directs ncurses to ignore
 nonmandatory time delays in the terminal description.

<h3 id="problems_coloring-id"><a name="problems_coloring" id=
 "problems_coloring">Making Color Work</a></h3>

<h4 id="vt100_color-id"><a name="vt100_color" id=
 "vt100_color">How do I get color with VT100?</a></h4>

Sorry. Real vt100's do not do color (ANSI or otherwise).
 Likewise, vt220, vt340 do not support ANSI colors. You may be
 running a terminal emulator which does and do not like this
 explanation.

You get "color with VT100" by running a terminal (or emulator)
 that supports colors, and by setting up the terminal description
 so that ncurses knows how to perform basic operations (setting
 the foreground and background colors). See <a href="#no_color">My
 terminal doesn't recognize color</a>.

Ncurses does not "know" that your terminal does support color.
 You must tell it. Some terminals (e.g., the higher models of
 DEC's VTxxx series) provide status information to a host on the
 capabilities supported by a terminal. Unix hosts do not interpret
 this information to set your $TERM environment variable. Instead,
 $TERM is set based on the connection which you make with your
 computer (e.g., a device listed in the /etc/gettydefs or
 /etc/gettytab files). You can override this by setting $TERM to a
 correct value or setting $TERMINFO to a <a href=
 "#home_terminfo">private database</a>.

Ncurses does not by itself know that vt100's do not do color.
 The standard reference for VT100 is its reference manual. There
 is a copy of that on <a href="http://vt100.net/">vt100.net</a>
 (look for EK-VT100-TM-003_Jul82). There are of course contrary
 sources of information about color-vt100's. The earliest one that
 I have seen is <a href=
 "http://graphcomp.com/info/specs/ansi_col.html">this</a> (a copy
 of which is also found <a href=
 "http://www.termsys.demon.co.uk/vtansi.htm">here</a>), which
 (besides the misinformation in its opening paragraph) has more
 than one point where it differs from vt100:

<ul>
 <li>Terminal Setup. The example given refers to modes,
 saying that few are implemented correctly. But the example
 itself is incorrect, because it omits the "?" character which
 tells that the sequence is a DEC private mode. For comparison,
 see DECAWM in xterm's <a href=
 "/xterm/ctlseqs/ctlseqs.html">control sequences</a>
 document.</li>

<li>
 Fonts

<ul>
 <li>
 vt100 has one "font". Bold is not a font, but one of
 the video attributes which include blink, reverse,
 underline.
 </li>

<li>
 The term "font" is inappropriate; "character set" is
 correct.
 </li>

<li>
 The control sequence cited for switching fonts omits
 the character which identifies the character set to
 use.
 </li>
 </ul>
 </li>

<li>
 Cursor Control

<ul>
 <li>
 Force Cursor Position is the standard
 HVP (Horizontal and Vertical
 Position).
 </li>

<li>
 Save Cursor is from ANSI.SYS rather than
 VT100.
 </li>

<li>
 Restore Cursor is from ANSI.SYS rather than
 VT100.
 </li>
 </ul>
 </li>

<li>
 Erase Screen states that the cursor moves to
 home. This is a well-known difference between ANSI.SYS and
 VT100. With VT100, the cursor does not move. Incidentally,
 ISO&nbsp;6429 (ECMA-48) does not mention either behavior.
 </li>

<li>
 Define Key is not a VT100 feature. It is a
 feature of ANSI.SYS.
 </li>

<li>
 Set Display Attributes.

<ul>
 <li>
 Bright. The standard and VT100 documentation
 refer to "Bold".
 </li>

<li>
 Dim is not a VT100 feature.
 </li>

<li>
 Hidden is a VT220 feature, not found in
 VT100.
 </li>
 </ul>
 </li>
 </ul>

There are several copies of this document (usually missing the
 copyright notice) available around the Internet.

After Andrey Chernov added vt100-color and others to
 the <a href=
 "http://www.freebsd.org/cgi/cvsweb.cgi/src/share/termcap/">FreeBSD
 termcap file</a> in 2002, I discussed this with him, pointing out
 that while there are terminal emulators which do this, the
 hardware did not. He resolved the issue by adding a comment:

<blockquote class="code-block">
 <pre>
# For color-enabled terminal emulators
</pre>
 </blockquote>

(See revisions 1.117 and 1.119).

Even with that clarification, there are others who conflate
 "ANSI", "VT100" and color, for example

<ul>
 <li><a href=
 "http://www.nyangau.org/terminfo/terminfo.htm">Andys Terminfo
 Package</a></li>

<li><a href=
 "http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=241717">Debian
 #241717</a></li>
 </ul>

Ncurses does not have terminal descriptions such as
 vt100-color because those are inevitably an
 oversimplification. For related discussion, see my comments on
 <a href=
 "/xterm/xterm.faq.html#xterm_terminfo">"xterm-color"</a>.

<h4 id="no_color-id"><a name="no_color" id="no_color">My terminal
 doesn't recognize color</a></h4>

Check the terminal description, to see if it is installed
 properly (e.g., in <code>/usr/share/terminfo</code>) by looking
 at the output of <code>infocmp</code>.

<dl>
 <dt>It should contain the following capabilities (shown with
 infocmp -L):</dt>

<dd>
 <blockquote>
 <code>orig_pair</code> or <code>orig_colors</code> 
 and <code>max_colors</code> 
 and <code>max_pairs</code>
 </blockquote>
 </dd>

<dd>
 <blockquote>
 <code>set_foreground</code> and
 <code>set_background</code> 
 or <code>set_a_foreground</code> and
 <code>set_a_background</code> 
 or <code>set_color_pair</code>
 </blockquote>
 </dd>
 </dl>

The <code>orig_pair</code> and <code>orig_colors</code>
 capabilities are not required in ncurses 5.0 (SVr4 curses works
 properly without them). For the rest:

<ul>
 <li>
 <code>max_colors</code> and <code>max_pairs</code> are
 numbers, which (at least) should be nonzero.

Prior to ncurses 6.1, those were signed 16-bit values.
 With <a href="announce-6.1.html">ncurses 6.1</a> (and using
 the recommended ABI 6), it is possible to use signed 32-bit
 values except in interfaces where the requirement for binary
 compatibility would prevent this.

Several of the terminal descriptions were updated in
 ncurses 6.1 to take advantage of this, e.g., using larger
 values for <code>max_pairs</code>. Older toolsets may not
 accept the larger values. Solaris and NetBSD curses assign
 any out-of-range capability a zero value.

Like ncurses 6.0, ncurses 6.1 can be compiled to support
 the older ABI 5. In that case, <code>tic</code> and
 <code>infocmp</code> check for overflow, and limit values to
 16 bits. Because of these added checks, it is possible to
 have ncurses ABI 5 and 6 on the same machine. In contrast,
 older ncurses releases use only the low-order 16
 bits of the value, which may have the same effect as Solaris
 and NetBSD (the low-order 16 bits of 0x10000 is zero).
 </li>

<li>
 The other capabilities are strings, used for the
 escape sequences that tell the terminal what colors it should
 use.
 </li>
 </ul>

The most common complaint is that "I can see
 colors using <code>ls</code>, but not with ncurses applications".
 This is due to not having installed the <code>terminfo</code>
 database. GNU <code>ls</code> (in contrast to FreeBSD
 <code>ls</code>) uses its own data (with hardcoded SGR numbers
 and ignoring the <a href=
 "/xterm/xterm.faq.html#xterm_terminfo">back color erase (bce)</a>
 feature supported in terminfo/termcap), which is unrelated to
 other applications. To be fair, GNU ls does have an environment
 variable (<code>LS_COLORS</code>) which can be used to customize
 the numbers a little, but in practice very few users modify
 it.

Other libraries or applications may
 recognize other environment variables to supplant the terminal
 description (see for example the discussion of <a href=
 "#no_colorterm">COLORTERM</a>), or even assign special meaning to
 different values of the <code>$TERM</code> environment variable,
 but ncurses uses only the terminal description.

<h4 id="bce_mismatches-id"><a name="bce_mismatches" id=
 "bce_mismatches">My terminal shows some uncolored spaces</a></h4>

Even if your terminal supports the <a href=
 "/xterm/xterm.faq.html#xterm_terminfo">back color erase (bce)</a>
 feature, you may still have problems using it.

We can blame the standard (ISO-6429 aka ECMA-48 aka "ANSI").
 It is too simplistic, being descriptive rather than
 prescriptive. What is that, you might ask? It means that the
 standards writers used existing simplified descriptions by
 developers for the features rather than decomposing the features
 into distinct parts&mdash;and then making precise descriptions of
 those for testing compatibility of various implementations. The
 standard does not go into enough detail to prescribe a
 particular behavior.

Back color erase covers many (usually) related features,
 because there are many control sequences which might affect
 color, depending on the implementation. Here is a list&mdash;for
 each item there is probably at least one terminal type which
 differs from ncurses' assumptions. The list shows terminfo names
 with the standard control name, if any, in parentheses:

<dd>
 The standard notes that this is "implementation-defined",
 and "cancels the effect of any preceding occurrence of
 SGR".

For practical purposes, ncurses has to assume that the
 cancellation applies to the normal video attributes such as
 bold, underline, reverse, but possibly not affecting color.
 Color came into the picture fairly late in the history of
 terminals, and there have been some (such as <a href=
 "/xterm/xterm.faq.html#bug_color_xterm">color_xterm</a>)
 where <tt>SGR 0</tt> had no effect on color.
 </dd>

<dd>
 The erase-display feature is assumed to fill the cleared
 display (or part of the display) with the current background
 color.

Examples of terminals which do not do this include dtterm
 and teraterm 2.3.
 </dd>

<dd>
 The erase-line feature is assumed to fill the cleared line
 (or part of line) with the current background color.

There is no standard requiring that the behavior of
 <tt>EL</tt> and <tt>ED</tt> should affect color in the same
 way. It is a design choice; ncurses assumes this.
 </dd>

<dd>
 Erasing a character (which does not move any text on the
 screen) is assumed to fill the erased cells with the
 background color.

Rxvt does not do this, though its behavior matches most of
 the other assumptions.

For xterm, <tt>ECH</tt> is a VT220 feature. VT220s (and
 for that matter, none of DEC's VTxxx series before the VT525)
 did not implement ANSI color (or any style of color which you
 would use with ncurses). Rxvt's behavior was based on its
 developer's interpretation of DEC's documentation. XTerm's
 behavior on the other hand was based on (generally) agreeing
 with the developer of the Linux console.
 </dd>

<dd>
 Inserting an empty cell at the current position
 is assumed to fill the empty cell with the background
 color.

Note that this control does not erase as one
 would infer from back color erase.
 But it creates an empty cell.
 </dd>

<dd>
 Deleting the cell at the current position shifts the
 remainder of the line left, introducing an empty
 cell on the right end of the line.

Note that this control does not erase as one
 would infer from back color erase.
 But it creates an empty cell.
 </dd>

<dd>
 Inserting an empty line at the current position
 is assumed to fill the new empty line with the background
 color.

Note that this control does not erase as one
 would infer from back color erase.
 But it creates an empty row.
 </dd>

<dd>
 Deleting the line at the current position shifts the
 remainder of the screen up, introducing an empty row
 on the bottom of the screen.

Note that this control does not erase as one
 would infer from back color erase.
 But it creates an empty row.
 </dd>

<dt><tt>ind</tt>, <tt>indn</tt>
 (nonstandard)</dt>

<dd>
 Scrolling the text up is assumed to fill the newly empty
 line (at the bottom of the display) with the current
 background color.

Indexing, aka "scrolling" is not covered in the standard.
 The behavior is completely implementation-dependent. This is
 a commonly used feature of XTerm.

There are subtle differences in behavior still possible,
 beyond filling and not filling. For instance, if the
 scrolling was in response to printing text which wrapped,
 that may or may not (depending on implementation) fill the
 newly empty line with the current background color.

For instance, a change in this area the Linux 2.6.26
 console driver in <a href=
 "http://lkml.org/lkml/2008/4/26/305">April 2008</a> was
 reverted <a href="https://lkml.org/lkml/2008/10/3/66">6
 months later</a>. That was a bug for Linux console, because
 it broke more than <a href=
 "ncurses-slang.html#cause_bce_linux">16 years</a> of
 predictable behavior (since 0.96b in June 1992). Likewise it
 would be a bug for XTerm (and I have rejected patches to
 "improve" its behavior in this detail). But other terminals
 could do this intentionally as I observed of the
 &ldquo;VT340&rdquo; emulator bundled with the OnNet TCP/IP
 product of FTP Software in the mid 1990s.

The standard, by the way, takes it for granted that text
 will wrap and force the display to scroll. But it uses the
 terms &ldquo;wrap&rdquo; and &ldquo;scroll&rdquo; in only a
 few obscure comments, and doubtless left much leeway for
 implementation-dependent behavior.

XTerm of course implements scrolling because it was based
 on DEC's VT100, which does this. While a real VT340 did not
 support ANSI colors, one model of DEC's last terminal (the
 VT525) did. Its documentation indicates that it was released
 September 30, 1994 (more than two years after Linux's
 &ldquo;new color model&rdquo; was available). The DEC
 documentation mentions a setting (<tt>DECECM</tt>):

<blockquote>
 The Erase color
 selection controls the background color used when text is
 erased or new text is scrolled on to the screen.
 Screen background causes newly erased
 areas or scrolled text to be written using color index
 zero, the screen background. This is VT and DECterm
 compatible. Text background causes erased
 areas or scrolled text to be written using the current text
 background color. This is PC console compatible and is the
 factory default.
 </blockquote>

While Linux console (and XTerm) treat the insert/delete
 character/line operations as an erasure, according to more
 than one developer with firsthand experience, the omission by
 DEC in this summary was intentional: their terminals did not
 do this.
 </dd>

<dt><tt>ri</tt>, <tt>rin</tt>
 (nonstandard)</dt>

<dd>
 Scrolling the display down is assumed to fill the newly
 empty line (at the top of the display) with the background
 color.

Again, there is no standard behavior. There are only
 design choices and assumptions.
 </dd>

<dt><tt>csr</tt> (nonstandard)</dt>

<dd>
 Setting scrolling margins is assumed to limit erasures
 (and filling with background color) to those margins.

This is a commonly used feature of XTerm. Vertical
 scrolling margins are not the only feature of this sort.
 There are also horizontal margins (VT320) and origin mode
 (VT100) which can affect the ability to move the
 cursor&mdash;and erase text. But ncurses does little at
 present with those, since (aside from XTerm), most terminal
 emulators support those features poorly or not at all.
 </dd>
 </dl>

The ncurses library for the most part assumes that
 bce means the particular set of choices made for
 Linux console and <a href="/xterm/xterm.html">xterm</a>. Not all
 popular terminals match those choices. The standard is silent on
 "correct" behavior. There are two ways that ncurses may handle
 these differences:

<ol>
 <li>
 provide a terminal description which omits the specific
 capability which differs from the assumption, or
 </li>

<li>
 in the library itself, be more pessimistic in optimizing
 the given capabilities.
 </li>
 </ol>

The former is preferred, of course. The library should not be
 cluttered up with special cases.

<h4 id="no_xterm_color-id"><a name="no_xterm_color" id=
 "no_xterm_color">My xterm program doesn't recognize color</a></h4>

Another specific problem lies with the terminfo description
 <a href="/xterm/xterm.faq.html#xterm_terminfo">xterm</a>. There
 are several xterm- and xterm-like terminfo entries in ncurses'
 terminfo database, corresponding to different terminal emulators.
 Only one is named "xterm", and that corresponds to the standard
 one. A fresh install of ncurses provides a choice between the
 current and previous standard ones:

<dl>
 <dt>xterm-new</dt>

<dd>that is based on XFree86 xterm (plus updates since 2006, of
 course). See xterm's <a href="/xterm/xterm.log.html">change
 log</a> for details.</dd>

<dt>xterm-old</dt>

<dd>This is the same as <code>xterm-r6</code>, e.g., the X11R6
 xterm. That program (X11R6 xterm) does not support ANSI text
 color.</dd>
 </dl>

For either flavor, your packager may have customized the
 definitions for backspace and delete to match the conventions of
 your system.

A fresh install of xterm on top of ncurses installs its
 terminfo entry as "xterm".

<h4 id="no_colorterm-id"><a name="no_colorterm" id=
 "no_colorterm">Why doesn't ncurses use
 <code>$COLORTERM</code>?</a></h4>

<code>$COLORTERM</code> is an environment variable originally
 used by some applications developers who were constructing
 programs that run on <code>rxvt</code>, a terminal emulator,
 using the <a href="ncurses-slang.html"><code>slang</code></a>
 library:

<ul>
 <li>
 It tells the <code>slang</code> library to ignore the
 terminal description, using a set of built-in capability
 strings which produce ISO 6429 (aka "ANSI") color on that
 emulator.

The choice of capability strings may work for other
 emulators, but in general does not (e.g., terminals which
 lack the back color erase capability, such as <a href=
 "http://hp.vector.co.jp/authors/VA002416/teraterm.html">Tera
 Term</a>, <a href="/xterm/xterm.faq.html#bug_dtterm">CDE
 dtterm</a> and <a href=
 "/xterm/xterm.faq.html#bug_nxterm">nxterm</a>).
 </li>

<li><code>slang</code>'s <a href=
 "ncurses-slang.html#env_COLORTERM">documentation</a> does not
 give useful information on what values are accepted for the
 variable. The only way you will know how to use it is by
 reading the source code of the program (or using cut/paste from
 someone who did that).</li>
 </ul>

Viewed as a fallback, <code>$COLORTERM</code> is perhaps
 acceptable (ncurses can be configured with built-in predefined
 terminal descriptions), but as a modifier to existing terminal
 descriptions it only leads to confusion, since most emulators
 that support color differ in minor details from the model which
 is supported by <code>slang</code>.

For example, dtterm nominally emulates a DEC vt220. Someone
 who knows this, but is also told that it supports color may
 (based on the usual misinformation available in newsgroups) try
 setting $TERM to "ansi", "linux" or <a href=
 "/xterm/xterm.faq.html#xterm_terminfo">"xterm-color"</a>. (Use
 <code>infocmp</code> to see why this is uniformly bad advice).
 This figure illustrates what goes wrong when following that
 advice:

There is a <code>dtterm</code> terminfo entry which provides
 correct behavior.

<h4 id="xterm_color-id"><a name="xterm_color" id=
 "xterm_color">Why not just use TERM set to "xterm-color"?</a></h4>

ncurses has a terminal description named
 <code>xterm-color</code>. Users assume that means it will work
 properly for "any" xterm. That is incorrect. It combines
 the simplest form of ANSI colors with the older X11R6 xterm.
 Originally, <code>xterm-color</code> corresponded to the
 color_xterm from the mid-1990s. That was superseded by
 XFree86 xterm in 1996. That is better than nothing, however using
 it in modern xterm (or anything accurately claiming to be
 compatible), these problems would occur:

<ul>
 <li>
 selections from colored backgrounds will add unnecessary
 spaces to the pasted text.
 </li>

<li>
 the function keys do not necessarily match (F1-F4 for
 instance), and will be incomplete (for example keys modified
 with shift-, control-).
 </li>

<li>
 scrolling will be noticeably slower.
 </li>

<li>
 some visual features (flash, hidden cursor) are not
 available.
 </li>

<li>
 it will not work properly in <code>luit</code>.
 </li>
 </ul>

Some terminal emulators may set this value; however it is
 unlikely that any current emulators implement this particular set
 of limited features. It is more likely that a more capable
 description exists or (given suitable documentation) that one
 could be constructed.

For instance, it has been reported that Mac OS X's bundled
 terminal emulator uses this value. However, reliable reports of
 its actual capabilities say that there are differences, which are
 addressed in ncurses as the <code>nsterm</code> entry. In this
 case, infocmp shows

<blockquote class="code-block">
 <pre style="font-family: monospace; font-size: 10pt;">
comparing xterm-color to nsterm.
 comparing booleans.
 hs: F:T.
 km: T:F.
 npc: F:T.
 xon: F:T.
 comparing numbers.
 colors: 8, 16.
 ncv: NULL, NULL.
 pairs: 64, 256.
 wsl: NULL, 50.
 comparing strings.
 acsc: '``aaffggiijjkkllmmnnooppqqrrssttuuvvwwxxyyzz{{||}}~~', '``aaffggjjkkllmmnnooppqqrrssttuuvvwwxxyyzz{{||}}~~'.
 blink: NULL, '\E[5m'.
 civis: NULL, '\E[?25l'.
 clear: '\E[H\E[2J', '\E[H\E[J'.
 cnorm: NULL, '\E[?25h'.
 dsl: NULL, '\E]2;\007'.
 el1: NULL, '\E[1K'.
 enacs: '\E)0', '\E(B\E)0'.
 flash: NULL, '\E[?5h$&lt;200/&gt;\E[?5l'.
 fsl: NULL, '^G'.
 hpa: NULL, '\E[%i%p1%dG'.
 ich: NULL, '\E[%p1%d@'.
 ich1: NULL, '\E[@'.
 invis: NULL, '\E[8m'.
 is2: '\E[m\E[?7h\E[4l\E&gt;\E7\E[r\E[?1;3;4;6l\E8', NULL.
 ka1: NULL, '\EOq'.
 ka3: NULL, '\EOs'.
 kb2: NULL, '\EOr'.
 kbs: '^H', '\177'.
 kc1: NULL, '\EOp'.
 kc3: NULL, '\EOn'.
 kend: NULL, '\E[F'.
 kent: NULL, '\EOM'.
 kf1: '\E[11~', '\EOP'.
 kf18: '\E[32~', '\E[22~'.
 kf2: '\E[12~', '\EOQ'.
 kf3: '\E[13~', '\EOR'.
 kf4: '\E[14~', '\EOS'.
 kfnd: '\E[1~', NULL.
 khome: NULL, '\E[H'.
 kich1: '\E[2~', NULL.
 kmous: '\E[M', NULL.
 kslt: '\E[4~', NULL.
 meml: '\El', NULL.
 memu: '\Em', NULL.
 op: '\E[m', '\E[0m'.
 rmam: NULL, '\E[?7l'.
 rs2: '\E[m\E[?7h\E[4l\E&gt;\E7\E[r\E[?1;3;4;6l\E8', '\E&gt;\E[?3l\E[?4l\E[?5l\E[?7h\E[?8h'.
 setab: '\E[4%p1%dm', '\E[%?%p1%{8}%&lt;%t%p1%'('%+%e%p1%{92}%+%;%dm'.
 setaf: '\E[3%p1%dm', '\E[%?%p1%{8}%&lt;%t%p1%{30}%+%e%p1%'R'%+%;%dm'.
 setb: NULL, '%p1%{8}%/%{6}%*%{4}%+\E[%d%p1%{8}%m%Pa%?%ga%{1}%=%t4%e%ga%{3}%=%t6%e%ga%{4}%=%t1%e%ga%{6}%=%t3%e%ga%d%;m'.
 setf: NULL, '%p1%{8}%/%{6}%*%{3}%+\E[%d%p1%{8}%m%Pa%?%ga%{1}%=%t4%e%ga%{3}%=%t6%e%ga%{4}%=%t1%e%ga%{6}%=%t3%e%ga%d%;m'.
 sgr: NULL, '\E[0%?%p6%t;1%;%?%p2%t;4%;%?%p1%p3%|%t;7%;%?%p4%t;5%;%?%p7%t;8%;m%?%p9%t\016%e\017%;'.
 sgr0: '\E[m', '\E[m\017'.
 smam: NULL, '\E[?7h'.
 tsl: NULL, '\E]2;'.
 vpa: NULL, '\E[%i%p1%dd'.
</pre>
 </blockquote>

Additionally, Mac OS X 10.7 is reported to use
 <code>xterm-256color</code> as a default <code>$TERM</code>
 value. This differs from <code>xterm-color</code> in several
 ways, in particular, the support for <a href=
 "/xterm/xterm.faq.html#xterm_terminfo">bce</a>. It also differs
 from the recommended <code>nsterm-256color</code> (infocmp
 reports 111 differences).

<h4 id="xterm_generic-id"><a name="xterm_generic" id=
 "xterm_generic">Why not just use TERM set to "xterm"?</a></h4>

It is easy to find people recommending just setting
 <code>$TERM</code> to "xterm", noting that several developers
 have used that as a default value for their terminal emulators.
 However, this is usually (except for xterm itself) a bad idea.
 The reason that those terminal emulators use "xterm" is not
 because they actually match the behavior, but

<ul>
 <li>
 it seems too much work to install the correct terminal
 definition on each machine (this was the reason given for
 rxvt's having done so for several years), or
 </li>

<li>
 it seems inconvenient to the user to put up with the
 absence of a given terminal description when making remote
 connections (which pass <code>$TERM</code> to the remote
 system).
 </li>
 </ul>

However, "xterm" refers to a <a href=
 "#no_xterm_color">specific</a> terminal emulator. Some of its
 details are widely emulated, while others (such as the function
 key definitions) are not.

Ncurses has provided accurate descriptions for the different
 terminal emulators for many years. While each terminal may
 implement escape sequences which are not in the others (<a href=
 "/xterm/xterm.faq.html#compare_versions">none</a> implement as
 much as half of xterm), what is important for curses applications
 is what can be represented in the terminal description. Using
 <a href="/ncurses/man/infocmp.1m.html">infocmp</a>, here is the
 amount of difference seen by ncurses for commonly used terminal
 emulators as of January 2012:

<blockquote>
 <table border="1" width="380pt" summary=
 "Counting differences from xterm with infocmp">
 <caption>
 Counting differences from xterm with infocmp
 </caption>

<tr>
 <th>Description</th>

<th>Count</th>
 </tr>

<tr>
 <td>ansi (for reference)</td>

<tr>
 <td>konsole</td>

<tr>
 <td>linux</td>

<tr>
 <td>mlterm</td>

<tr>
 <td>nsterm (Apple Terminal)</td>

<tr>
 <td>putty</td>

<tr>
 <td>screen</td>

<tr>
 <td>screen.xterm-new</td>

<tr>
 <td>vte (e.g., gnome-terminal)</td>

<tr>
 <td>xterm-color</td>

<tr>
 <td>xterm-new (current standard)</td>

<tr>
 <td>xterm-old (old standard)</td>

<tr>
 <td>vt100 (for reference)</td>

Noting that infocmp lists 182 capabilities for xterm-new,
 entries with more than a small number of differences demonstrate
 lack of compability of xterm "me-too's" versus xterm itself.

That only covers the features which are in the terminal
 description, and does not address differences between terminal
 emulators. For example, in <a href=
 "terminfo.src.html#t20170729">mid-2017</a>, an update to the
 xterm terminal description added the ECMA-48 <code>REP</code>
 (repeat character) control. It was part of xterm since January
 1997, but a terminal description using the feature was part of
 xterm only (not ncurses). After adding it to
 ncurses, it was observed that:

<ul>
 <li>rxvt is unaffected, since it does not use
 <code>TERM=xterm</code>.</li>

<li>mlterm and OSX Terminal supported <code>REP</code>, using
 <code>TERM=xterm</code>.</li>

<li>VTE, konsole, PuTTY, iTerm2 did not support this standard,
 longstanding feature, of <a href="/xterm/xterm.html">xterm</a>
 although they set <code>TERM</code> to
 <code>xterm</code> or
 <code>xterm-256color</code>.</li>

<li>screen and tmux use a different <code>TERM</code> setting,
 and seemed to work (although tmux had some other issue with the
 test).</li>
 </ul>

Here are screenshots showing the problem, on the left correct
 and on the right incorrect:

<h4 id="xterm_256color-id"><a name="xterm_256color" id=
 "xterm_256color">Why not make "xterm" equated to
 "xterm-256color"?</a></h4>

It breaks things, in more than one way. Seriously, no one
 should be asking this question, but see <a href=
 "#xterm_generic">Why not just use TERM set to "xterm"?</a>. Still
 (writing in mid-October 2012), I see that the answer is
 needed.

First, some background is needed:

<ul>
 <li>
 SVr4 curses supported color terminals, defining names for
 the 8 ANSI colors.
 </li>

<li>
 SVr4 curses used color pairs to reflect
 limitations of some hardware terminals which could not render
 all possible combinations of color. The color pair value was
 a 6-bit field (defined by <code>A_COLOR</code>) in the
 <code>chtype</code> values, which allowed for all
 combinations of those 8 ANSI colors.
 </li>

<li>
 PDCurses 2.0 (November 23, 1992) set aside 7 bits for
 <code>A_COLOR</code>, but used only 6-bits.
 </li>

<li>
 ncurses began using an 8-bit field for color pairs
 starting (probably) early in 1993, providing for all
 combinations of 16-colors (e.g., for CGA displays).
 </li>

<li>
 X/Open started standardizing and extending curses shortly
 after. The result is a hybrid, with most of SVr4 curses
 present, but adding a new data-type <code>cchar_t</code> to
 hold wide characters. Incidentally, they provided an
 alternate way to encode colors. None of the Unix vendors did
 anything in this area other than store the same 8 ANSI colors
 inside the <code>cchar_t</code>, but they did provide room
 for more.
 </li>

<li>
 At that time (the mid-1990s), it was not apparent how this
 additional room might be useful, and in developing ncurses we
 simply stored the color in <code>cchar_t</code> in the
 attribute member. That allowed only 8 bits (16 colors), just
 as in the <code>chtype</code> data.
 </li>

<li>
 The xterm 256-color feature started with Todd Larason's
 patch, which I added in mid-1999 in <a href=
 "/xterm/xterm.log.html#xterm_111">patch #111</a>.
 </li>

<li>
 For the record, ANSI X3.64 did not define 256
 colors, nor even 16 colors. 
 Those (256- and 16-colors) are extensions defined by
 xterm and aixterm,
 respectively. 
 There is no such thing as &ldquo;ANSI 256 colors.&rdquo; File
 that away with &ldquo;<a href="#vt100_color">VT100
 colors</a>.&rdquo; 
 An extension is just what it sounds like: something
 compatible with, but not part of the
 standard.
 </li>

<li>
 Though we (including Steve Wall's changes for 88-colors)
 continued to refine the feature for some time, it was not at
 first used much by applications. For instance, though I added
 the terminal description to ncurses, it was only partly
 supported by ncurses until I added the extended color
 features beginning in <a href=
 "/ncurses/NEWS.html#t20050101">January, 2005</a>.
 </li>

<li>
 The ncurses extended color feature added a new structure
 member to <code>cchar_t</code> which allows for as many color
 pairs as one can store in an integer.
 </li>

<li>
 The change to <code>cchar_t</code> makes the library no
 longer binary compatible with ncurses ABI&nbsp;5. It is part
 of ncurses ABI&nbsp;6
 </li>

<li>
 Even with that change, there are several features in the
 application interface using 16-bit signed integers, limiting
 color pairs to 32767, but allowing 256 colors for practical
 purposes.
 </li>
 </ul>

Here are screenshots showing the ncurses test-program with
 extended colors, using menu entry "C", added in that time period.
 The first uses normal-weight text:

and the second uses bold font:

Other terminal emulators adapted the feature before any
 noticeable use by applications. After noticing these and
 verifying their behavior, I added corresponding entries to
 ncurses' terminfo.src file. Here are relevant changes:

<blockquote>
 <table border="1" summary="88/256-color Terminal Descriptions">
 <caption>
 88/256-color Terminal Descriptions
 </caption>

<th>Change</th>
 </tr>

<td>add xterm-88color, xterm-256color</td>
 </tr>

<td>add rxvt-16color, ibm+16color</td>
 </tr>

<td>update pairs for xterm-88color and xterm-256color to
 reflect the ncurses extended-color support</td>
 </tr>

<td>add aixterm-16color to demonstrate 16-color
 capability</td>
 </tr>

<td>add nsterm-16color entry</td>
 </tr>

<td>add xterm+256color building block, and add
 gnome-256color, putty-256color, rxvt-256color</td>
 </tr>

<td>add konsole-256color entry</td>
 </tr>

<td>add Eterm-256color, Eterm-88color, and
 rxvt-88color</td>
 </tr>

<td>add linux-16color; add ccc and initc capabilities to
 xterm-16color</td>
 </tr>

<td>add mrxvt-256color</td>
 </tr>

<td>add mlterm+256color entry</td>
 </tr>

<td>correct order of use-clauses in st-256color</td>
 </tr>

<td>add nsterm-256color, make this the default nsterm</td>
 </tr>
 </table>
 </blockquote>

Things started to change around 2005-2006 as you can see by
 looking at the dates in the table. <a href=
 "https://bugzilla.redhat.com/show_bug.cgi?id=175684">Red Hat
 #175684</a> gives some clue, stating that the "next version of
 Emacs has support for 256 color terminals". Emacs on the console
 by the way uses termcap, not terminfo and certainly not
 (n)curses.

I chose to add new entries rather than change xterm (and
 others) because they are incompatible. There is more than one
 aspect to consider:

<ul>
 <li>
 some applications on the same machine will blindly attempt
 to use all of the colors without being ready to handle more
 than 8/16 colors. There was for example more than one
 bug-report with Fedora of this type. The symptom was that
 text would be invisible (black on black). The bug reports
 were eventually closed without correcting the problem.
 </li>

<li>
 connections to/from remote machines with a different
 notion of xterm will behave inconsistently.
 </li>

<li>
 termcap-style (or the rare terminfo-style) applications on
 a given machine are the only ones that can use the -256color
 or -88color flavors. But aliasing those to xterm
 will break ncurses applications.
 </li>

<li>
 not all termcap applications will work properly, depending
 on what definitions they see for the <code>AF</code> and
 <code>AB</code> (or the <code>Sf</code> and <code>Sb</code>)
 capabilities. That is because an optimal capability
 for these (as provided from ncurses' terminal database) is a
 conditional expression&mdash;something like this:

Some applications (such as <a href=
 "http://vim.wikia.com/wiki/256_colors_in_vim">vim</a>) may
 allow (or even embed, hard-coded) the much simpler, but less
 efficient variant corresponding to

but doing that on an application-by-application basis is
 pointless. If a termcap application uses the ncurses library
 to send the escape sequences (using <a href=
 "/ncurses/man/curs_termcap.3x.html">tgoto</a>, for example),
 it is also redundant, because ncurses will return a usable
 terminfo string, and interpret it correctly.
 </li>

<li>
 there is already a widely-distributed terminal entry with
 the desired features. Since it has proven to be inconvenient
 to use the xterm-256color name everywhere, renaming
 it to xterm will simply shift the problem without
 providing a solution.
 </li>

<li>
 making incompatible changes to a widely-used terminal
 description produces bug-reports. I have limited changes of
 that type in xterm to describing new features, not changing
 existing features. Even with that, there have been complaints
 for each change. Consider

<dd><a href=
 "http://www.tomsarazac.com/tom/opinions/xterm-problems.html">
 rant by an end-user</a>, <a href=
 "/xterm/xterm.log.html#xterm_28">twelve years</a> after the
 change in question (viewer discretion advisory). David
 Dawes was the first to point out that this change
 introduced a problem with remote systems. After some
 discussion we decided to keep it.</dd>

<dt>Function-key modifiers
 </dt>

<dd>In xterm's <a href=
 "/xterm/xterm.log.html#xterm_94">patch #94</a> I added
 function-key modifiers. In <a href=
 "/xterm/xterm.log.html#xterm_167">patch #167</a> I modified
 the protocol to address a problem with Emacs (though making
 it configurable). In the meantime, KDE and GNOME had copied
 the earlier feature into their terminal emulators. Ten
 years later, neither had updated to the 2002 version of
 xterm's function-keys. There were several reports, the
 first that I noticed appeared in Red Hat's system. For
 example see <a href=
 "https://bugzilla.redhat.com/show_bug.cgi?id=127773">#127773</a>.
 <a href=
 "https://bugs.launchpad.net/ubuntu/+source/xterm/+bug/96676">
 Ubuntu #96676</a> is another example (still unresolved in
 2012).</dd>

<dd><a href=
 "http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=444250">Debian
 #444250</a>. Essentially the problem here was that adding a
 flag to the description telling how to enable/disable
 xterm's meta mode exposed a differing interpretation of
 <a href="#bash_meta_mode">meta mode</a> in bash which dates
 back to around 1990.</dd>
 </dl>
 </li>
 </ul>

Although ncurses (starting with release 5.5 in 2005) can be
 configured to support 256 colors, it seems that no widespread
 distribution is using the extended colors feature (certainly not
 Red Hat or Debian &mdash; perhaps OpenSuSE does). That (and the
 extended mouse feature) are the main features for the
 "ABI&nbsp;6", which could easily coexist with the usual
 "ABI&nbsp;5" ncurses.

Without widespread use of ABI&nbsp;6, there is no reason to
 change the primary flavor of any of ncurses' terminfo entries to
 the -256color or -88color flavor.

In the Debian distribution there was <a href=
 "http://lists.debian.org/debian-devel/2008/09/threads.html#00435">
 an attempt in 2008</a> on the part of the ncurses package
 maintainer to provide a migration path to ABI&nbsp;6. That effort
 died, and does not at this time have a clear successor.

Just because it is not viable does not prevent others from
 assuming it is. For example:

<ul>
 <li>
 <a href=
 "http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=230990">Debian
 #230990</a> asks for the extended mouse support. It was
 opened in February 2004.
 </li>

<li>
 <a href=
 "http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=405602">Debian
 #405602</a> took 18 months to move the xterm-256color entry
 from the ncurses-term to the ncurses-base package in
 mid-2008.
 </li>

<li>
 <a href=
 "http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=532537">Debian
 #532537</a> asks for a change of xterm to
 xterm-256color. It was opened in June 2009. As
 justification, it stated that "all Debian systems have been
 supporting the 256-colour interface for as long as anyone can
 remember". The package changelog for xterm states that the
 feature was added in January 2006, which when compared with
 the Debian release cycle makes it very far from "all Debian
 systems".
 </li>

<li>
 In mid-2012, I noticed that Fedora&nbsp;18 would have
 xterm aliased to xterm-256color. 
 Some notes are in Fedora's wiki <a href=
 "http://fedoraproject.org/wiki/Features/256_Color_Terminals">Features/256
 Color Terminals</a>. However, the proponents of the
 change place too much emphasis on just changing it and pay
 less attention to the ensuing problems.
 </li>

<li>
 Even in early 2015, postings such as <a href=
 "http://stackoverflow.com/questions/13060139/how-can-i-get-more-colors-curses-python-terminal">
 How can I get more colors? [Curses] [Python]
 [Terminal]</a> are common. 
 As I noted (and because my correction was twice deleted,
 promoted it to this FAQ):

<blockquote>
 <pre class="code-block">
Not exactly (curses will not do what was indicated). See the ncurses faq: Why
not make "xterm" equated to "xterm-256color"?. The terminal database will
return appropriate strings usable by low-level applications, but curses
applications will not use them.
</pre>
 </blockquote>

Specifically, this part of the "answer" is untrue:

<blockquote>
 If you run your program using
 <code>env TERM=xterm-256color
 yourprogram</code>, curses will enable 256 colors,
 and it will work fine as long as your terminal actually
 supports it!
 </blockquote>

Because (unlike this FAQ), StackExchange postings are
 <a href=
 "http://www.merriam-webster.com/dictionary/ephemeral">ephemeral</a>,
 here also are my followup comments:

<ul>
 <li>
 In this context, "low-level" applications include any
 which do their own screen optimization without using the
 curses high-level interface.
 </li>

<li>
 Urwid has display modules for curses and "raw". It
 uses hard-coded escape sequences in the latter to provide
 256 colors. (If it had used the terminal database to
 obtain the escape sequences, I would call the "raw"
 interface a low-level application).
 </li>

<li>
 The reason why the curses high-level applications do
 not display 256 colors is because that requires an ABI
 change - making packagers reluctant to introduce it.
 </li>

<li>
 The links for "Fabulous" are dead. It is unlikely that
 it can be mixed with ncurses calls; there has been no
 discussion on bug-ncurses mailing list.
 </li>
 </ul>
 </li>
 </ul>

Finally (<a href="/ncurses/announce-6.0.html">August
 2015</a>), we have ncurses6 (ncurses 6.0). Releasing a new
 version is a different matter from getting it accepted and in
 wide use. Once packagers have been providing it for some time,
 the compatibility issues will be of less concern.

<h4><a name="xterm_16MegaColors" id="xterm_16MegaColors">Why only
 16 (or 256) colors?</a></h4>

Before getting to the point, some background is needed.

Curses applications run in terminals (or terminal emulators).
 They set colors for the screen and parts of it by sending
 additional characters (escape sequences) to the screen.

The ANSI standard described 8 colors, did not in any sense
 allude to more colors. This is a palette-based set of colors;
 there is <a href="/xterm/xterm.faq.html#dont_like_blue">no
 standard</a> for how those colors are composed.

A few terminals recognize the aixterm
 16-color extension (not part of any standard), e.g., xterm
 <a href="/xterm/xterm.log.html#xterm_39">patch #39</a>. Again,
 there is no agreement between different terminals on the actual
 colors used. (While xterm is configurable, some of the others are
 not).

Starting with xterm <a href=
 "/xterm/xterm.log.html#xterm_111">patch #111</a>, there is the
 &ldquo;<a href=
 "terminfo.src.html#tic-xterm-256color">xterm-256color</a>&rdquo;
 terminal type. Using SGR 38 and 48 in an arguably nonstandard
 manner (at the time there was no free access to ISO&nbsp;8613-6),
 it provides a predefined palette of 256 colors which can be
 modified through escape sequences. With the terminals which
 imitate this (none matches xterm's behavior exactly), there are
 two problems:

<ul>
 <li>
 the colors differ
 </li>

<li>
 the interpretation of the escape sequences is generally
 incomplete, e.g., not recognizing the sequences for modifying
 the palette or not accepting only one color per escape
 sequence.
 </li>
 </ul>

These deficiencies are well known, easily demonstrated.

Sometime after xterm patch #111, a freely accessible version
 of ISO&nbsp;8613-6 (ITU&nbsp;T.416) was made available. Some
 (mostly secondhand) discussion of the ramifications of this is in
 KDE <a href=
 "https://bugs.kde.org/show_bug.cgi?id=107487">#107487</a>.

A small minority of terminals support an additional
 interpretation of SGR 38 and 48 alluded to in ISO&nbsp;8613-6 as
 direct colour. The term &ldquo;direct colour&rdquo; has
 been interpreted as &ldquo;24-bit mapping of RGB&rdquo; (it is an
 interpretation, since the ISO standard referenced is not
 specific). What the standard actually says is:

<blockquote class="code-block">
 The first parameter element indicates a choice between:

<blockquote>
 0 implementation defined (only applicable for the character
 foreground colour)

1 transparent.

2 direct colour in RGB space.

3 direct colour in CMY space.

4 direct colour in CMYK space.

5 indexed colour.
 </blockquote>

...

If the first parameter element has the value 5, then there
 is a second parameter element specifying the index into the
 colour table given by the attribute &ldquo;content colour
 table&rdquo; applying to the object with which the content is
 associated.

...

If the first parameter element has the value 2, the
 parameter elements 3, 4 and 5, are three integers for red,
 green, and blue colour components. Parameter 6 has no
 meaning.
 </blockquote>

There is no &ldquo;24-bit&rdquo; in the standard.

As is often the case with the ISO terminal standards, there is
 little or no prior art to use as a basis for standardization of
 these features. Accordingly, the existing implementations
 differ:

<ul>
 <li>
 ISO&nbsp;8613-6 prescribes a different syntax than that
 used for xterm patch #111.
 </li>

<li>
 ISO&nbsp;8613-6 does not specify the &ldquo;colour
 table&rdquo; it only mentions it.
 </li>

<li>
 xterm provided a color table, with additional control
 sequences for modifying it.
 </li>

<li>
 Konsole used the xterm syntax for its 256-color
 (indexed color) and direct color features, as noted
 in a <a href=
 "https://cgit.kde.org/konsole.git/tree/doc/user/README.moreColors?id=a593f29e2441158ade667992cbf36900727bbb08">
 README</a> file.

The example script, by the way, demonstrates the former
 (256-color feature) and was copied/renamed from
 xterm's sources, as shown <a href=
 "https://cgit.kde.org/konsole.git/commit/tests?id=3dc500cb801d1d945a97ced531213b828517d3d4">
 here</a>. As of February 2017, there is still no
 corresponding script for direct colour in Konsole's
 <a href=
 "https://cgit.kde.org/konsole.git/tree/tests?id=a593f29e2441158ade667992cbf36900727bbb08">
 source repository</a>.

The direct color feature was largely ignored for
 a few years. There were occasional comments, such as <a href=
 "https://groups.google.com/forum/?hl=en#!searchin/vim_use/%22max$20colors$20in$20text$20based$20terminals%22|sort:relevance|spell:true/vim_use/RP4iLbpyFNw/yJjhasUWI6AJ">
 this</a> on the vim-users discussion group.
 </li>

<li>
 Some attempts were made by GNOME developers to apply
 similar changes to other terminals; no coherent description
 was found summarizing the success of that endeavor.
 </li>

<li>
 After some reflection, I provided in xterm (with <a href=
 "/xterm/xterm.log.html#xterm_282">patch #282</a>, September
 2012) support for the direct color feature:

<blockquote class="code-block">
 modify <code>SGR&nbsp;38</code> and
 <code>SGR&nbsp;48</code> to accept RGB index, matching the
 closest entry in xterm's palette.
 </blockquote>

Both that and the earlier 256-color feature are usable
 with either the ISO&nbsp;8613-6 (colon) syntax, or the xterm
 patch #111 (semicolon) syntax. While xterm will accept either
 syntax indefinitely, the terminal description may change
 depending on circumstances:

<ul>
 <li>The <a href=
 "terminfo.src.html#tic-xterm_256color">xterm+256color</a>
 building block is used for legacy
 applications</li>

<li>The <a href=
 "terminfo.src.html#tic-xterm_256color2">xterm+256color2</a>
 building block is used for standard
 applications</li>
 </ul>
 </li>

<li>
 On beginning this item (January 2014), none of the other
 terminals which I tested accepted the ISO&nbsp;8613-6 syntax
 (there were some <a href=
 "https://gitlab.com/gnachman/iterm2/issues/218">unverified</a>
 reports).
 </li>

<li>
 There are other syntax issues. In the changes for xterm
 patch #282 (and the earlier KDE changes), no one noticed or
 commented on this additional sentence from ISO 8613-6:

<blockquote class="code-block">
 If the first parameter element has the value 2, 3, or 4,
 the second parameter element specifies a colour space
 identifier referring to a colour space definition in
 the document profile.
 </blockquote>

The ISO document offers no help in deciding what values
 that identifier might have or how it might be used, but
 omitting it may be a nuisance. Interestingly enough, some
 developers copied the feature from xterm while
 citing the ISO document as the source.
 </li>
 </ul>

Konsole's README is not specific (mentioning only
 &ldquo;3-byte RGB color space&rdquo;), but the current
 source-code refers to truecolor which is
 different from direct color. See for example the
 discussion of <a href=
 "https://tronche.com/gui/x/xlib/window/visual-types.html">Visual
 Types</a> in the X11R5 library documentation (1991):

<blockquote class="code-block">
 <ul>
 <li>For DirectColor, a pixel value is decomposed into
 separate RGB subfields, and each subfield separately indexes
 the colormap for the corresponding value. The RGB values can
 be changed dynamically.</li>

<li>TrueColor is treated the same way as DirectColor except
 that the colormap has predefined, read-only RGB values. These
 RGB values are server-dependent but provide linear or
 near-linear ramps in each primary.</li>
 </ul>
 </blockquote>

Using that escape sequence for TrueColor makes it a
 non-standard implementation. To be fair, <a href=
 "https://bugs.kde.org/show_bug.cgi?id=107487#c22">KDE's 2006
 changes</a> to provide 256-colors (and incidentally direct
 color) did not hint at TrueColor. That came from other
 people. Konsole mentions &ldquo;truecolor&rdquo; only in one
 place, setting the environment variable <a href=
 "#no_colorterm"><code>COLORTERM</code></a> from <a href=
 "https://bugs.kde.org/show_bug.cgi?id=371919">KDE #371919</a>,
 copying John Davis' magic variable for overriding
 <code>TERM</code>, and triggering whatever special feature he has
 in mind for s-lang. With one value or another, it (and other
 magic variables to amend this) dates back to mid-1995
 (rxvt-2.11). Here is the complete documentation in s-lang
 mentioning the variable as of February 2017:

<blockquote>
 <pre class="code-block">
 Of course not all terminals are color terminals. If the S-Lang global
 variable SLtt_Use_Ansi_Colors is non-zero, the terminal is assumed to
 be a color terminal. The SLtt_get_terminfo will try to determine
 whether or not the terminal supports colors and set this variable
 accordingly. It does this by looking for the capability in the
 terminfo/termcap database. Unfortunately many Unix databases lack this
 information and so the SLtt_get_terminfo routine will check whether or
 not the environment variable COLORTERM exists. If it exists, the
 terminal will be assumed to support ANSI colors and
 SLtt_Use_Ansi_Colors will be set to one. Nevertheless, the
 application should provide some other mechanism to set this variable,
 e.g., via a command line parameter.
</pre>
 </blockquote>

Beyond the issue of escape sequences, konsole and xterm use
 different library calls for implementing direct color.
 This makes no difference to an application which might use those
 escape sequences, but bears some comment:

<ul>
 <li>
 konsole uses Qt. In its source code (see
 <a href=
 "https://projects.kde.org/projects/kde/applications/konsole/repository">
 KDE repository</a>), it stores the color information in a
 <a href=
 "http://qt-project.org/doc/qt-5.0/qtgui/qcolor.html">QColor</a>
 class. The documentation mentions

<blockquote class="code-block">
 QColor is platform and device independent. The QColormap
 class maps the color to the hardware.
 </blockquote>

Because konsole's source code does not refer explicitly to
 QColormap, it relies on default behavior of the class. 
 The documentation for <a href=
 "http://qt-project.org/doc/qt-5.0/qtwidgets/qcolormap.html">QColormap</a>
 gives half a clue, in its comment about modes. Qt's
 documentation is vague in many respects (in sharp contrast to
 MSDN), leaving developers no recourse other than reading its
 source code. 
 Here is a starting point: <a href=
 "https://qt.gitorious.org/qt/qt/source/57756e72adf2081137b97f0e689dd16c770d10b1:src/gui/painting">
 Qt 4.8 src/gui/painting</a>.
 </li>

<li>
 In the 4.8 branch, we find that the QColormap class
 <a href=
 "https://qt.gitorious.org/qt/qt/source/57756e72adf2081137b97f0e689dd16c770d10b1:src/gui/painting/qcolormap_x11.cpp">
 for X11</a> was doing essentially what xterm does, i.e.,
 color maps.
 </li>

<li>
 In the 5.0 source (still in development at the
 time the vim_use discussion took
 place&mdash;released 6 months later), there are some changes
 for <a href=
 "https://qt.gitorious.org/qt/qtbase/source/a4e703764e5f174fc6ca3f919e3acf44203dad91:src/widgets/util/qcolormap.cpp">
 QColormap</a>. The class initialization distinguishes between
 24-bit and 8-bit color. The X11-relevant details have been
 moved away from this class.
 </li>

<li>
 Distinguishing between 8-bit and 24-bit color is a useful
 simplification for X11, since it trades off between storing
 an 8-bit value in one of the columns for the R/G/B structure
 versus storing three 8-bit values.

XTerm checks for this special case to allow exact color
 matches. Otherwise, it uses the closest color match.
 </li>

<li>
 Because xterm allows the application to set the palette
 which is used for either style of color (256-color or
 direct), providing &ldquo;lot of colors&rdquo; would have
 been pointless.
 </li>

<li>
 Konsole has no provision for modifying the palette, in
 either case.
 </li>
 </ul>

The implication of the direct colour feature is
 &ldquo;lots of colors&rdquo;. On the surface, that may seem
 attractive. However, there are several considerations:

<li>where will they use it?</li>

<li>how many colors do they need?</li>

<li>what are the performance tradeoffs?</li>
 </ul>

To answer the question, bear in mind that this is an area of
 special interest to only a handful of developers. End-users read
 &ldquo;16 million colors&rdquo; and uncritically accept arguments
 that they can effectively use this. However, there are only (at
 most) a few thousand characters on a terminal's screen at a time.
 That number of colors exceeds by a couple of orders of magnitude
 the ability of anyone to discern the differences and rely on
 those distinctions to aid them in viewing text.

Moving past the first couple of points,
 the availability of direct color depends on how the
 application uses the information. As a terminfo
 description, that works for applications which use the terminal
 database directly. The principal ones which come to mind are text
 editors (emacs and vi-clones). Generally their developers have
 chosen to make their task more difficult by remaining with
 termcap (which cannot express the relevant
 escape sequences). So they incorporate some hard-coded behavior.
 For instance, in a change to emacs in <a href=
 "http://git.savannah.gnu.org/cgit/emacs.git/commit/?id=e463e5762bbe628be3d15da066a90f079a8468b3">
 February 2017</a>, the developer

<ul>
 <li>used the ncurses user-defined capability feature to define
 single-parameter strings named <code>setf24</code> and
 <code>setb24</code> which transform the parameter into the
 three parameters needed for direct color, and</li>

<li>because emacs uses termcap and normally cannot read those,
 the developer added a call to the terminfo function <a href=
 "man/curs_terminfo.3x.html"><code>tigetstr</code></a> (which
 probably only works with ncurses, since only the <a href=
 "man/curs_termcap.3x.html">termcap</a> interface is
 initialized) to get the information, and</li>

<li>if successful, replaces the capability read from termcap
 for normal coloring, and</li>

<li>passes the result to the emacs-specific <a href=
 "https://github.com/emacs-mirror/emacs/blob/master/src/tparam.c">
 tparam</a> (which by the way accepts four parameters which
 could have been used for R/G/B as most developers would
 expect).</li>
 </ul>

Working through these obstacles, it is possible to
 &ldquo;use&rdquo; termcap to provide &ldquo;lots of colors&rdquo;
 without affecting the ncurses library. Termcap applications do
 not use much of ncurses; nor do their developers contribute
 significantly to ncurses. The emacs workaround was unsuitable for
 ncurses because emacs interprets the data in its own way, and
 displays the result independently of ncurses.

Some mailing-list comments indicate that the existing
 capabilities are not well-understood by developers, e.g., this
 discussion in <a href=
 "https://lists.gnu.org/archive/html/bug-ncurses/2013-10/msg00007.html">
 October 2013</a>. Taking the discussion into account, consider
 the <a href="/ncurses/terminfo.src.html#tic-linux-c">linux-c</a>
 terminal description dating from 1996. That sets up a palette
 using a color index value mapped to R/G/B components. Other
 longstanding uses in ncurses include these:

<blockquote>
 <a href=
 "/ncurses/terminfo.src.html#tic-linux-c-nc">linux-c-nc</a>,
 <a href="/ncurses/terminfo.src.html#tic-putty">putty</a>,
 <a href=
 "/ncurses/terminfo.src.html#tic-xterm-16color">xterm-16color</a>,
 and <a href=
 "/ncurses/terminfo.src.html#tic-xterm_256color">xterm+256color</a>
 </blockquote>

Some additional comments on mapping colors using
 <code>initc</code> and <code>initp</code> are found in <a href=
 "http://lists.schmorp.de/pipermail/rxvt-unicode/2013q3/thread.html#1826">
 this discussion</a>.

Applications which use ncurses are a different
 matter. Unlike termcap applications, curses libraries (and
 analogous ones such as s-lang) use combinations of the terminal
 capabilities to allow their applications to work on a wide
 variety of terminals. There are a few limitations, due to the
 early implementation of terminfo:

<ul>
 <li>
 4096 size-limit on compiled entries
 </li>

<li>
 16-bit limit for numbers (to keep them small), which are
 all signed values
 </li>

<li>
 Because some terminal types did not provide a way to set
 foreground and background colors independently, the concept
 of color pairs was introduced.
 </li>
 </ul>

Using the ncurses 5 ABI, you have available 16 colors, or 256
 pairs of colors. Using the ncurses 6 ABI, you <a href=
 "#xterm_256color">have</a> 256 colors, or 32767 pairs (the limit
 for a signed 16-bit number). That limit is good enough for
 realistic applications, which could not have that many character
 cells on a screen simultaneously (unless of course, using 1-pixel
 fonts to pretend to draw graphics, e.g., <a href=
 "http://aa-project.sourceforge.net/aalib/">AA-lib</a>). Since
 ncurses is used to draw text, that is not a valid
 issue.

S-lang does not implement color-pairs (or <a href=
 "#need_padding">padding</a>, etc.), However (reading the source
 code), it also is limited by the terminal descriptions, using a
 maximum color index value 32767 (15-bits&mdash;signed
 short). No method for changing 15 into 24 was noted in <a href=
 "http://lists.jedsoft.org/lists/slang-users/2014/index.html#00002">
 this discussion</a>. However, that led to a bug report with a
 <a href=
 "http://www.midnight-commander.org/attachment/ticket/3145/slang-truecolor-demo.patch">
 patch</a> which <a href=
 "http://www.midnight-commander.org/ticket/3145#comment:11">isn't</a>
 fully "ABI compatible". John Davis was able to work around the
 obvious problems in this patch by splitting up the extended color
 field to use some space in his library's data structure.
 Otherwise, users would have had to recompile their applications
 to use the library, since the patch altered these features of the
 interface:

<blockquote>
 <pre class="code-block">
#define SLTT_BOLD_MASK 0x01000000UL
#define SLTT_BLINK_MASK 0x02000000UL
#define SLTT_ULINE_MASK 0x04000000UL
#define SLTT_REV_MASK 0x08000000UL
#define SLTT_ALTC_MASK 0x10000000UL
#define SLTT_ITALIC_MASK 0x20000000UL
</pre>
 </blockquote>

If you read the <a href=
 "http://www.jedsoft.org/snapshots/">source code</a>, you will see
 that he did not solve the problem of changing 15 into 24 (as of
 February 2017, it still used a signed 16-bit limit for the number
 of colors).

No matter how one uses the limits based on the terminal
 database, scaling is needed to map onto the actual device.

Bypassing the limits which can be derived from the terminal
 database (and veering off into the swamp of hard-coded behavior),
 the idle spectator might ask &ldquo;why not add a special ad hoc
 interface to (somehow) just do it?&rdquo; One of VTE's developers
 asked for that (see mailing list thread from <a href=
 "http://lists.gnu.org/archive/html/bug-ncurses/2014-01/msg00010.html">
 January 2014</a>).

The problem is that you cannot &ldquo;just do
 it&rdquo;&mdash;both ncurses and <a href=
 "http://www.jedsoft.org/slang/">s-lang</a> do optimization to
 address their legitimate users (those who are using the superior
 performance due to working with characters, especially via remote
 connections). They have longstanding APIs which expose some of
 the related information. Adding an ad hoc interface will either
 break compatibility, or require duplicating information. Either
 way, most of the existing users would be adversely affected.

Some hint of this is given in GNOME <a href=
 "https://bugzilla.gnome.org/show_bug.cgi?id=704449#c29">#704449</a>.
 Given the source of the comment, the effect is understandably
 understated. The usual reason given for wanting this feature has
 been for making pretty color schemes for text editors (e.g.,
 <a href=
 "http://www.vimninjas.com/2012/08/26/10-vim-color-schemes-you-need-to-own/">
 this</a> and <a href=
 "http://stackoverflow.com/questions/16711226/to-fix-the-color-in-vim-or-to-take-the-plunge">
 this</a>).

More generally, the way to &ldquo;solve&rdquo; the problem
 would be to abandon compatibility with <a href=
 "https://www2.opengroup.org/ogsys/catalog/C094">X/Open
 Curses</a>, using <code>int</code> where
 <code>short</code> is now used, and provide a
 new (incompatible) terminal database format to store bigger
 numbers. Incidentally, this would break applications which use
 the existing terminal database format, such as <a href=
 "http://www.jedsoft.org/slang/">s-lang</a>.

This extended item was prompted by an advertisement <a href=
 "http://lists.gnu.org/archive/html/bug-ncurses/2014-01/msg00008.html">
 early in 2014</a> posted to the ncurses mailing list. That was
 while I was working on <a href="ncurses-mapsyms.html">versioned
 symbols</a>, needed to gain packager's acceptance for <a href=
 "announce-6.0.html">ncurses6</a>. A hypothetical future
 release aimed solely at providing &ldquo;lots of colors&rdquo;
 could do each of the items mentioned in the previous paragraph.
 But it had no place in the ncurses6 plan, which required only
 recompiles, no source-changes.

Beginning with a fallacious premise (using a source <a href=
 "#misinfo-ansiseq">known</a> to be erroneous), its authors (the
 two individuals mentioned in <a href=
 "http://lists.gnu.org/archive/html/bug-ncurses/2014-01/msg00010.html">
 this thread</a>) selected specific instances in an
 attempt to persuade the reader that a majority of terminals
 implement TrueColor (&ldquo;16 million colors&rdquo;), and has a
 discussion page devoted to expanding this:

<ul>
 <li>
 Several entries (such as <a href=
 "http://worldwidemann.com/finally-terminated/">FinalTerm</a>)
 are bogus.
 </li>

<li>Some others involve a stretch, e.g., suggesting that the
 existence of some patch for PuTTY is going to make it
 do 24-bit colors for you.</li>

<li>
 All of the VTE-based terminals are listed as if they were
 independent. Presenting the information this way makes it
 (perhaps) more convincing. Certainly, if you get to vote as
 many times as you choose, you can win any dispute.
 </li>
 </ul>

Much of the discussion on the page is in the same vein, e.g.,
 talking about standards, while at the same time
 discarding the technical details of ITU&nbsp;T.416 and
 ECMA-48 which began this story.

Besides the discussion on that page, there are several related
 items such as <a href=
 "https://bugs.kde.org/show_bug.cgi?id=371919">KDE #371919</a>,
 repeating some of the invented facts from that page, e.g.,

<blockquote class="code-block">
 Unfortunately the maintainer of ncurses and terminfo is not
 interested at all in adding this possibility to terminfo, see
 e.g.
 https://lists.gnu.org/archive/html/bug-ncurses/2016-08/msg00036.html.
 </blockquote>

After the release of ncurses 6.1, the page's authors made
 minor, but not substantiative changes. In contrast, the changes
 made for <a href="announce-6.1.html">ncurses 6.1</a> to implement
 and integrate extended numbers (and colors) took about a year to
 develop (see <a href="ncurses-6.0-20170218-to-6.1.html">this
 page</a> for the full diffstat):

<blockquote>
 <pre class="code-block">
# 602 files changed, 54948 insertions(+), 35233 deletions(-)
</pre>
 </blockquote>

Those people are not contributors to ncurses. While comments
 are useful (sometimes) none of it was written by them.

Without significant revision, there is nothing more to
 discuss.

<h4 id="white_black-id"><a name="white_black" id=
 "white_black">Ncurses resets my colors to white/black</a></h4>

This is the way SVr4 curses works. From the XSI Curses
 specification

<blockquote class="code-block">
 The start_color() function also restores the colors on the
 terminal to terminal-specific initial values. The initial
 background color is assumed to be black for all terminals.
 </blockquote>

If your terminal description does not contain the
 <code>orig_pair</code> or <code>orig_colors</code> capability,
 you will be unable to reset colors to the pre-curses state. This
 happens, for example, with <code>aixterm</code>.

However, if your terminal does support resetting colors to a
 default state, ncurses will use this when exiting Curses mode.
 Terminal types that do this include the Linux console,
 <code>rxvt</code> and the XFree86 <code>xterm</code>.

Ncurses 4.1 provides an extension
 <code>use_default_colors()</code> which allows an application
 running on a terminal which supports resetting colors to mix the
 default foreground and background colors with the 8 defined
 curses colors.

<h4 id="interchanged_colors-id"><a name="interchanged_colors" id=
 "interchanged_colors">Why are red/blue interchanged?</a></h4>

You may notice if you are porting an application from SVr4
 curses to ncurses (or the other way), that some versions of
 ncurses have some pairs of colors interchanged with respect to
 SVr4 curses. This is a bug in ncurses (sorry). Someone made an
 error translating terminal descriptions, and confused the
 setaf/setab terminal capabilities with the setf/setb
 capabilities.

The 8 colors black, red, green, yellow, blue, magenta, cyan,
 white are coded in the ANSI (setaf/setab) convention with red=1,
 green=2 and blue=4, while the older (setf/setb) uses red=4,
 green=2 and blue=1. SVr4 curses accommodates either,
 interchanging colors in the setf/setb to match the setaf/setab
 style. Ncurses' terminfo database incorrectly renamed the
 setaf/setab capabilities to setf/setb, making it incompatible
 with the SVr4 curses library.

This was corrected in ncurses 4.1, but incorrect in all
 preceding versions.

<h3 id="problems_display-id"><a name="problems_display" id=
 "problems_display">Other Display Problems</a></h3>

<h4 id="no_line_drawing-id"><a name="no_line_drawing" id=
 "no_line_drawing">Line-drawing characters come out as x's and
 q's</a></h4>

The x's and q's correspond to a table (from terminfo/termcap)
 which tells ncurses how to map the "alternate" character set to
 the terminal's set of graphic characters. The reference for this
 table comes from the vt100. If the unmapped characters appear,
 then the terminal emulator does not recognize the escape sequence
 for switching between normal and alternate fonts that is given in
 the terminfo description.

There are several cases of note:

<ul>
 <li>
 Terminal emulators which use a different escape sequence
 or different range for mapping the resulting characters. For
 instance the so-called vt100-compatibles such as Linux
 console and Tera Term.
 </li>

<li>
 Terminal emulators which are locale-sensitive. Again,
 Linux console is a problem area when running in UTF-8 mode,
 since its nominal vt100-compatibility is further lessened by
 ignoring the escape sequences dealing with fonts. The
 <code>screen</code> utility also has the same problem;
 whether to make the implementation simple or to copy the
 Linux console. It ignores vt100-style font switching when the
 locale is a UTF-8 flavor.
 </li>

<li>
 If you happen to be using Solaris, it is often configured
 to prefer its terminal database to ncurses, even when ncurses
 is installed. However, its terminal description for xterm
 omits the <code>enacs</code> which is used to enable
 line-drawing. This does not work well with applications such
 as <code>screen</code> and <code>luit</code>.
 </li>
 </ul>

For the first case, you simply have to find the correct
 terminfo description. Fixing the latter is harder, since the
 damage is done outside ncurses. (Though one can easily make
 things compatible enough that this particular issue would never
 appear, that style of solution is not deemed proper by some
 coders).

The normal ncurses libraries support 8-bit characters. The
 ncurses library can also be configured
 (<code>--enable-widec</code>) to support wide-characters (for
 instance Unicode and the UTF-8 encoding). The corresponding
 wide-character <code>ncursesw</code> libraries are
 source-compatible with the normal applications. That is,
 applications must be compiled and linked against the
 <code>ncursesw</code> library.

The ncurses 5.3 release provides UTF-8 support. The special
 cases of Linux console and <code>screen</code> were addressed in
 development patches completed at the end of 2002.

<h4 id="bad_line_drawing-id"><a name="bad_line_drawing" id=
 "bad_line_drawing">Line-drawing characters come out as Latin-1
 characters</a></h4>

Those capital A's with dots on top. Yes, that is almost always
 a mismatch with the terminfo description.

There are some special cases such as Tera Term which are
 related to the font.

<h4 id="utf8_line_drawing-id"><a name="utf8_line_drawing" id=
 "utf8_line_drawing">Line-drawing characters do not appear</a></h4>

There is no character at all where the line-drawing should
 appear, and other characters may be shifted around the screen.
 This may happen on the Linux console, but also on some other
 terminal emulators. The line-drawing characters do not appear
 because the terminal emulator is treating them as illegal
 characters.

ncurses starts by assuming that the terminfo is correct, and
 overrides it for some special cases which are known to misbehave.
 See the discussion of <code>NCURSES_NO_UTF8_ACS</code> in the
 ncurses manpage for details.

<h4 id="emacs_cvvis-id"><a name="emacs_cvvis" id=
 "emacs_cvvis">Why does my cursor blink in Emacs?</a></h4>

Depending on Emacs, or other programs, the answer differs.
 Most often this question is raised by people using the Linux
 console.

There are three capabilities defined in terminfo which affect
 the cursor:

<blockquote>
 <table border="1" summary="Terminfo Cursor Capabilities">
 <caption>
 Terminfo Cursor Capabilities
 </caption>

<th>Terminfo</th>

<th>Termcap</th>

<th>Description</th>
 </tr>

<tr>
 <td>cursor_invisible</td>

<td>civis</td>

<td>make cursor invisible</td>
 </tr>

<tr>
 <td>cursor_normal</td>

<td>cnorm</td>

<td>make cursor appear normal (undo civis/cvvis)</td>
 </tr>

<tr>
 <td>cursor_visible</td>

<td>cvvis</td>

<td>make cursor very visible</td>
 </tr>
 </table>
 </blockquote>

The three capabilities appear in the terminfo database 128,
 178 and 103 times respectively. Some of those terminfo entries
 are building blocks. For the entries reported by <a href=
 "/ncurses/man/toe.1m.html">toe</a>, there are 670 which modify
 the cursor, out of 1629 entries. Some of these allow only between
 invisible/normal or normal/visible. There are 159 which allow all
 three possibilities, as well as 27 which allow only one
 choice.

The cursor blinks because the application (Emacs for instance)
 uses the <code>cvvis</code> capability. "Very visible" for a
 terminal cursor generally means that it blinks. Various tradeoffs
 are possible, depending on the terminal between underline, block,
 blinking.

Applications modify the cursor either via <a href=
 "/ncurses/man/curs_kernel.3x.html">curs_set</a> (in ncurses) or
 by sending the appropriate escape sequence to the terminal using
 termcap or terminfo calls. While <code>curs_set</code> refrains
 from substituting one escape sequence for another missing one,
 this is not true of termcap/terminfo applications. Frequently
 they replace a missing <code>cvvis</code> with an available
 <code>cnorm</code>. After a while, the programmers forget the
 difference, and surprise their users when someone upgrades a
 terminal description.

The Linux console tie-in happened with ncurses <a href=
 "terminfo.src.html#t19990703">in 1999</a>, by adding
 <code>cvvis</code> to the "linux" terminal description.

Outside of Emacs, the replacement of a missing
 <code>cvvis</code> with <code>cnorm</code> is the most common
 cause of surprise.

But Emacs is different. It intentionally uses the "very
 visible" blinking cursor by default. There is documentation
 <a href=
 "http://www.gnu.org/software/emacs/manual/html_node/emacs/Cursor-Display.html">
 here</a>. The earliest available version of <a href=
 "http://git.savannah.gnu.org/cgit/emacs.git/log/src/term.c">the
 code</a> from 1991 uses <code>TS_visual_mode</code> in this
 way.

<h4><a name="multithread" id="multithread">Why does (fill in the
 blank) happen when I use two threads?</a></h4>

If you have a program which uses curses in more than one
 thread, you will almost certainly see odd behavior. That is
 because curses relies upon static variables for both input and
 output. Using one thread for input and other(s) for output cannot
 solve the problem, nor can extra screen updates help. This FAQ is
 not a tutorial on threaded programming.

Starting with <a href="announce-5.7.html">ncurses 5.7</a>,
 this implementation of curses provides the ability to configure
 and compile the library to help solve the problem by:

<ul>
 <li>
 reducing the use of static variables (see <a href=
 "/ncurses/man/curs_sp_funcs.3x.html">curs_sp_funcs(3x)</a>),
 </li>

<li>
 adding mutexes around the low-level terminal I/O,
 </li>

<li>
 changing global variables such as <code>LINES</code> to
 "getter" functions (see <a href=
 "/ncurses/man/curs_opaque.3x.html">curs_opaque(3x)</a>,
 and
 </li>

<li>
 adding functions which an application can use to avoid
 those which rely upon global (or static) variables (see
 <a href=
 "/ncurses/man/curs_threads.3x.html">curs_threads(3x)</a>.
 </li>
 </ul>

Almost all programs can be recompiled to use the "ncursest" or
 "ncursestw" libraries. Just recompiling is not enough to make a
 program thread-safe. As usual, some (re)design effort is probably
 needed.

The test-programs provided with ncurses (<a href=
 "/ncurses/ncurses-examples.html">ncurses-examples</a>) include a
 few which demonstrate this alternate configuration of the
 library: ditto, rain, worm.

<h3 id="problems_keyboard-id"><a name="problems_keyboard" id=
 "problems_keyboard">Keyboard Problems</a></h3>

<h4 id="cursor_appmode-id"><a name="cursor_appmode" id=
 "cursor_appmode">My cursor keys do not work</a></h4>

For instance, you may press the up-arrow and see "OA" or "[A",
 rather than having your program recognize the up-arrow.

The terminal description may be correct, or the program may
 not be making proper use of it. ncurses (and other programs which
 use terminal descriptions) simply match the characters sent by
 the terminal against a list of strings defined in the terminal
 description.

vt100's can send different strings for cursor-keys (and
 keypad-keys) depending on how they are initialized. Terminal
 emulations which provide vt100 functionality behave the same way.
 There are two modes:

<dl>
 <dt>normal mode
 </dt>

<dd>This corresponds to the power-up state of the terminal.
 Cursor keys send escape sequences (strings beginning with the
 ASCII escape character) that mimic cursor-movement control
 sequences. Those begin with <code>escape&nbsp;[</code>. An
 up-arrow is <code>escape&nbsp;[&nbsp;A</code> (also expressed
 as <code>CSI&nbsp;A</code>.</dd>

<dt>application mode
 </dt>

<dd>This is the programmed mode of the terminal. Cursor keys
 send escape sequences that can be easily distinguished from
 cursor-movement control sequences. Those begin with
 <code>escape&nbsp;O</code>. An up-arrow is
 <code>escape&nbsp;O&nbsp;A</code> (also expressed as
 <code>SS3&nbsp;A</code>.</dd>
 </dl>

The terminal description describes only one mode. Usually that
 is the application mode.

The terminal description tells how to initialize the terminal,
 e.g., to make the cursor-keys send application mode escape
 sequences:

<ul>
 <li>Long ago, when there were many types of terminals in use,
 some would send nothing for special keys unless the terminal
 was initialized in this way.</li>

<li>Other terminals could move the cursor in local
 mode (not sending anything to the host computer).</li>

<li>Finally, there were some which would send
 cursor-movement sequences unless initialized.</li>
 </ul>

To accommodate all types of terminals, the convention was that
 a program would initialize the cursor keys when it needed them,
 i.e., to use the programmed mode of the terminal. That works well
 for full-screen programs. Terminal descriptions (usually)
 describe the application mode for this reason.

This convention does not work well for command-line programs,
 which do not initialize the terminal. Keys which differ between
 normal and application modes are not available for programs which
 do not initialize the terminal.

For example, the <code>readline</code> library (which was
 originally hardcoded, and did not use termcap or terminfo),
 exploited the feature of vt100's that their cursor-keys transmit
 something in normal mode.

The <code>readline</code> library (which is used by
 <code>bash</code>) did this for several years before someone
 connected it to a termcap library. Around that point, someone
 made a terminal description for the Linux console which "solved"
 the problem of initializing the terminal for <code>bash</code> by
 not initializing it at all. Cursor keys always send normal-mode
 strings with this arrangement.

Granted, that works&mdash;provided that the terminal is
 initialized consistently. If your terminal happens to "solve" the
 problem in a different way, e.g., by starting in application
 mode, and if your environment sets <code>TERM</code> to "linux",
 you may have problems with your cursor-keys.

Depending on the application, users expect an "alt" key to do
 one of two things:

<ul>
 <li>
 send an escape character before special keys such as
 cursor-keys, or
 </li>

<li>
 act as an extended shift, allowing you to enter codes for
 Latin-1 values from 160 to 255.
 </li>
 </ul>

xterm supports both features, e.g., with the
 <code>eightBitInput</code> and related resources.

Bash users have an extra problem in xterm when using a recent
 terminal description (since <a href=
 "/xterm/xterm.log.html#xterm_216">xterm patch #216</a>). For
 completeness, I noted in the terminal description how an
 application can set/reset "meta" mode. It turns out that bash
 turns on meta mode if it is available. It is not an optional
 feature in bash; bash's developers assumed that you would use it.
 That is an odd assumption, since the feature is rarely
 implemented. There are (counting xterm) only three terminal
 descriptions in ncurses' terminal database providing this (the
 low-level blocks for Ann Arbor, SCO, and XTerm).

Making it configurable was proposed in <a href=
 "https://bugzilla.novell.com/show_bug.cgi?id=541379">this bug
 report</a>. Earlier, bash's maintainer had <a href=
 "http://lists.gnu.org/archive/html/bug-ncurses/2007-04/msg00012.html">
 expressed reluctance</a> to address the issue. However, bash 4.1
 added an "enable-meta-key" feature, dated 9 October 2009. That
 makes it possible for users to disable it.

This is usually due to an incorrect terminal description. When
 it is not, it is due to the terminal emulator itself.

Application developers tend to hardcode references in their
 programs to the "home" and "end" key, forgetting that the
 presence of any particular function key depends on the
 terminal.

Part of the problem with terminal descriptions is due to the
 way xterm's terminal descriptions are used in ncurses. Because
 xterm emulates a VT220, the terminal descriptions provided with
 it follows the DEC keyboard convention, which does not provide
 home/end keys. Rather, it provides find/select keys. See for
 example the discussion of <a href=
 "/xterm/xterm.faq.html#how2_fkeys">function keys</a> and <a href=
 "/xterm/xterm.faq.html#xterm_keypad">keypad</a>.

On the other hand, the descriptions in ncurses reflect the Sun
 and PC keyboards which many people use. They define home/end keys
 for xterm.

However, some packagers for ncurses have pasted xterm's
 terminfo descriptions into the ncurses distribution. The effect
 is to remove the home/end key definitions from ncurses.

The problem is easily fixed by ignoring the packager's
 improvements and reinstalling the ncurses terminfo database.
 <a href="#which_terminfo">This section tells where to get
 it.</a>

<h4 id="modified_keys-id"><a name="modified_keys" id=
 "modified_keys">How can I use shift- or control-modifiers?</a></h4>

The standard response is "curses doesn't do that".

There are workarounds. Some explanation is needed first.

Most implementations of curses work with terminals that use
 serial communication. Generally those were inexpensive. Adding
 keys to the keyboard was probably less expensive than adding
 logic to handle different types of modifiers (such as shift).
 Terminals that could send different types of modifiers used to be
 rare.

However, the IBM PC provided a widely available platform with
 a keyboard that could provide modifier information. On the other
 hand, there was no standard protocol for sending the information
 on a serial line.

Most of the activity in exploiting this platform consisted of
 ad hoc implementations for computer consoles, such as for SCO,
 OS/2, etc., which associated the control-, shift- and alt-key
 modifiers with function-key numbering. That is, rather than
 attempting to extract information on the modifiers from function
 key expressed as a character string, the various combinations are
 numbered (using function-key number, multiplied by a code
 corresponding to the modifiers) and associated with the numbered
 function keys in the termcap/terminfo description.

For example, the <code>scoansi</code> description encodes the
 modifiers by an arbitrary sequence from the possible final
 characters for an ANSI control, e.g.,

<blockquote class="code-block">
 <dl>
 <dt>normal </dt>

<dt>F13-F24 are shifted F1-F12 </dt>

<dt>F25-F36 are control F1-F12 </dt>

<dt>F37-F48 are shift+control F1-F12 </dt>

Some combinations are missing (kf14 for instance corresponds
 to a back-tab).

A different scheme is used by <code>rxvt</code> (which was
 influenced by xterm and different hardware portability
 tradeoffs):

kf1=\E[11~,&nbsp;kf10=\E[21~,&nbsp;kf11=\E[23~,&nbsp;kf12=\E[24~, 
 kf13=\E[25~,&nbsp;kf14=\E[26~,&nbsp;kf15=\E[28~,&nbsp;kf16=\E[29~, 
 kf17=\E[31~,&nbsp;kf18=\E[32~,&nbsp;kf19=\E[33~,&nbsp;kf2=\E[12~, 
 kf20=\E[34~,&nbsp;kf3=\E[13~,&nbsp;kf4=\E[14~,&nbsp;kf5=\E[15~, 
 kf6=\E[17~,&nbsp;kf7=\E[18~,&nbsp;kf8=\E[19~,&nbsp;kf9=\E[20~, 
 kf21=\E[23$,&nbsp;kf22=\E[24$, 
 kf23=\E[11\^,&nbsp;kf24=\E[12\^,&nbsp;kf25=\E[13\^,&nbsp;kf26=\E[14\^, 
 kf27=\E[15\^,&nbsp;kf28=\E[17\^,&nbsp;kf29=\E[18\^,&nbsp;kf30=\E[19\^, 
 kf31=\E[20\^,&nbsp;kf32=\E[21\^,&nbsp;kf33=\E[23\^,&nbsp;kf34=\E[24\^, 
 kf35=\E[25\^,&nbsp;kf36=\E[26\^,&nbsp;kf37=\E[28\^,&nbsp;kf38=\E[29\^, 
 kf39=\E[31\^,&nbsp;kf40=\E[32\^,&nbsp;kf41=\E[33\^,&nbsp;kf42=\E[34\^, 
 kf43=\E[23@,&nbsp;kf44=\E[24@, 
 
 </blockquote>

The pattern is not so obvious here. The developer who assigned
 the numbering chose certain combinations from a table which was
 too large to map into the available 60 numbered keys. Here is the
 complete table, noting that <code>F1</code> is the X11 code which
 is not necessarily synonymous with <code>kf1</code>:

<blockquote>
 <table border="1" summary="Rxvt Function-Key Modifiers">
 <caption>
 Rxvt Function-Key Modifiers
 </caption>

<th>Normal</th>

<th>Shift</th>

<th>Control</th>

<th>Shift+Control</th>
 </tr>

That is, the final character is again used to encode the
 modifiers. There are several terminals which use different
 final-character encoding schemes, e.g., dg (Data General),
 interix, and those based on <code>ansi.sys</code> or
 <code>cons25</code>.

xterm uses a different scheme, encoding the modifiers as a
 parameter. There are many combinations available. Here is a short
 example:

kf1=\EOP,&nbsp;kf13=\E[1;2P,&nbsp;kf25=\E[1;5P,&nbsp;kf37=\E[1;6P, 
 kf49=\E[1;3P,&nbsp;kf61=\E[1;4P, 
 
 </blockquote>

All of that focuses on function keys. There are other
 special keys on the IBM PC keyboard, e.g., the editing keypad and
 cursor keys.

Here, conventional terminfo/termcap provides just a little
 help, but providing names for some of the shifted keys. xterm
 defines some of these:

kDC=\E[3;2~,&nbsp;kEND=\E[1;2F,&nbsp;kHOM=\E[1;2H,&nbsp;kIC=\E[2;2~, 
 kLFT=\E[1;2D,&nbsp;kNXT=\E[6;2~,&nbsp;kPRV=\E[5;2~,&nbsp;kRIT=\E[1;2C, 
 
 </blockquote>

But xterm can also work for control-, alt-, and
 meta-modifiers. None of those are defined in conventional
 terminfo/termcap.

ncurses allows you to define extended descriptions, i.e., to
 make up your own names. The xterm terminfo descriptions do just
 that: they define names for the modified cursor- and
 editing-keypad keys which are just the xterm modifier code
 appended to the name for the shifted key. For example, the
 delete-key (kdc) can be represented as shown below:

<blockquote>
 <table border="1" summary="Example of Xterm Extended Key (DC)">
 <caption>
 Example of Xterm Extended Key (DC)
 </caption>

<tr>
 <th>Xterm Code</th>

<th>Modifier</th>

<th>Extended Name</th>
 </tr>

<tr>
 <td>1 (or missing)</td>

<td>Normal</td>

<td>Shift</td>

<td>Shift + Alt</td>

<td>Control</td>

<td>Shift + Control</td>

<td>Alt + Control</td>

<td>Shift + Alt + Control</td>

Other terminal descriptions can be (and in ncurses, have been)
 modified to use this naming scheme for extended keys:

<ul>
 <li>
 gnome-terminal and konsole (for more than ten years) used
 an older version of xterm's encoding (<a href=
 "/xterm/xterm-function-keys.html">deprecated in 2002</a>
 because the parameter can be mistaken for a repeat count).
 Although the encoding is different (and these terminals omit
 some combinations), the functionality is similar.
 </li>

<li>
 rxvt also provides distinct escape sequences for modified
 cursor- and editing-keys.
 </li>
 </ul>

By adding these to the terminal description, curses
 applications will see the keycode rather than individual
 characters when processing <a href=
 "/ncurses/man/curs_inopts.3x.html">keypad</a> mode.

There are a few limitations though:

<ul>
 <li>
 the keycodes for the extended keys are generated at
 runtime, so the application must ask what their values are
 using <a href=
 "/ncurses/man/key_defined.3x.html">key_defined</a>.
 </li>

<li>
 compiled terminfo descriptions, though larger than
 termcap's 1023 bytes, are still limited to 4096 bytes. Many
 applications (such as those built using <a href=
 "ncurses-slang.html"><code>slang</code></a>) rely on this
 limit. Applications can work around this by defining keys at
 runtime, using <a href=
 "/ncurses/man/define_key.3x.html">define_key</a>.
 </li>
 </ul>

<h4 id="visible_keys-id"><a name="visible_keys" id=
 "visible_keys">How can I see what my keyboard sends?</a></h4>

All of the trouble-shooting for keyboard problems relies on
 you being able to see what the keys send. 
 You can do this in more than one way:

<ul>
 <li>
 Some implementations of the <code>cat</code> program have
 a <code>-v</code> option which tells it to show non-printing
 characters in visible form. 
 These include all BSD and Linux-based systems, as well as
 <a href=
 "https://www-01.ibm.com/support/knowledgecenter/ssw_aix_71/com.ibm.aix.cmds1/cat.htm">
 AIX</a>, <a href=
 "https://www.freebsd.org/cgi/man.cgi?query=cat&amp;apropos=0&amp;sektion=1&amp;manpath=HP-UX+11.11&amp;arch=default&amp;format=html">
 HPUX</a> and <a href=
 "https://www.freebsd.org/cgi/man.cgi?query=cat&amp;apropos=0&amp;sektion=1&amp;manpath=SunOS+5.10&amp;arch=default&amp;format=html">
 Solaris</a>.

While all of the extent Unix systems appear to have this
 feature, <a href=
 "http://pubs.opengroup.org/onlinepubs/9699919799/utilities/cat.html">
 POSIX</a> does not document it.

You would use it by typing

It has a few limitations:

<ul>
 <li>it does not stop special characters such as
 control-C</li>

<li>the output format cannot be inverted to obtain the
 original inputs.</li>
 </ul>
 </li>

<li>
 You can suppress the terminal driver's interpretation of
 the first byte in a given key by first pressing
 <code>control-V</code> (the lnext or
 literal-next character).

In practice, this is "good enough" since most of the
 terminals you might want to use will only have a leading
 escape character in the special key strings.
 </li>
 </ul>

That makes it possible to see what your keyboard sends. There
 is a corresponding problem seeing what programs actually send to
 your terminal.

<ul>
 <li>
 A few platforms have <a href=
 "https://www.freebsd.org/cgi/man.cgi?query=vis&amp;apropos=0&amp;sektion=1&amp;manpath=FreeBSD+11-current&amp;arch=default&amp;format=html">
 <code>vis</code></a>, which first appeared in 4.4BSD. There
 is a corresponding <a href=
 "https://www.freebsd.org/cgi/man.cgi?query=unvis&amp;apropos=0&amp;sektion=1&amp;manpath=FreeBSD+11-current&amp;arch=default&amp;format=html">
 <code>unvis</code></a> which can put the string back together
 (unlike <code>cat -v</code>). You would use this by using
 <code>script</code> to capture all of the bytes sent to your
 terminal, e.g., in a file named <code>typescript</code>.
 </li>

<li>
 But <code>vis</code> is not available everywhere (e.g.,
 AIX and Solaris do not provide it, nor are you likely to find
 it with a Linux-based system).

At the <a href="/misc_tools/CHANGES.html#t19951215">end of
 1995</a>, I wrote a similar program named <a href=
 "/misc_tools/index.html#item:unmap"><code>unmap</code></a>
 based on a description of <code>vis</code>, and likewise
 <a href=
 "/misc_tools/index.html#item:map"><code>map</code></a> (like
 <code>unvis</code>) for completeness.
 </li>
 </ul>

<h3 id="problems_hardware-id"><a name="problems_hardware" id=
 "problems_hardware">Using Hardware (Real) Terminals</a></h3>

<h4 id="reset_logout-id"><a name="reset_logout" id=
 "reset_logout">Why does reset log me out?</a></h4>

This is a limitation of real hardware terminals: resetting
 them will break the communications to the host temporarily. Some
 terminal interfaces will tolerate this. Others (most) will
 interpret this as hanging up, and log you out.

The reset is specified in the terminal description, e.g.,

rs1=\Ec, 
 
 </blockquote>

That is the hardware reset escape sequence for vt100. Some
 terminals provide enough control over their features that a very
 complicated substitute could be concocted for the normal reset
 which does not perform a hardware reset. In practice, this is not
 easily done.

<h4 id="need_padding-id"><a name="need_padding" id=
 "need_padding">Why do I get trash on the screen?</a></h4>

This is a problem of real hardware terminals. Cheap terminals
 and cheap interfaces do not do sophisticated flow control (e.g.,
 XON/XOFF). Instead, they rely on a host which does not send them
 characters too rapidly. Remote terminal servers may provide flow
 control; direct console or serial port connections often do not.
 (If you are asking this question, you probably have inexpensive
 hardware).

Terminfo descriptions designed for these inexpensive terminals
 have delay times specified in the control sequences which take
 extra time, such as clearing the screen. For example, the vt100
 description tells the application to wait 50msec after clearing
 the screen:

clear=\E[H\E[J$&amp;lt;50&amp;gt;, 
 
 </blockquote>

Note: the <a href="ncurses-slang.html"><code>slang</code></a>
 library does not implement delay times, and is not suitable for
 applications which require direct connection to a hardware
 terminal. The author of that library states that no one uses
 hardware terminals any more, suggesting that I add this
 information to the FAQ.

<h3 id="problems_programs-id"><a name="problems_programs" id=
 "problems_programs">Interaction with Other Programs</a></h3>

<h4 id="handle_piping-id"><a name="handle_piping" id=
 "handle_piping">Redirecting I/O to/from a Curses application</a></h4>

In principle, you should be able to pipe to/from a curses
 application. However, there are caveats:

<ul>
 <li>
 Some (very old) curses implementations did not allow
 redirection of the screen. Ncurses, like Solaris curses,
 consistently writes all output to the standard output. You
 can pipe the output to a file, or use <code>tee</code> to
 show the output while redirecting.
 </li>

<li>
 Ncurses obtains the screen size by first using the
 environment variables LINES and COLS (unless you have
 disabled it with the <code>use_env</code> call), then trying
 to query the output file pointer, and then (finally) the
 terminal description. If you are redirecting output, then a
 query based on the file pointer will always fail, resulting
 in the terminal description's size.
 </li>

<li>
 Similarly, you can redirect input to an ncurses
 application. However, I have observed that the use of
 <code>setvbuf</code> (for better output buffering) interferes
 with the use of stream I/O on Linux-based systems (and
 possibly other platforms). Invoking <code>setvbuf</code> may
 (depending on the implementation) cause buffered stream input
 to be discarded. Ncurses does not use buffered input, however
 you may have an application that mixes buffered input with a
 curses session.
 </li>
 </ul>

<h4 id="readline_library-id"><a name="readline_library" id=
 "readline_library">What about readline?</a></h4>

The <code>readline</code> library appeals to a number of
 people, who would like to use it within an ncurses
 application.

At first glance&mdash;but only for that instant&mdash;it
 appears that the library is flexible enough to substitute a
 different display driver, e.g., to output via something other
 than <code>tput()</code> and <code>putc()</code>. Close reading
 of its <code>display.c</code> file shows this is ultimately
 futile. Quoting (Brian Fox's comment from the early 1990s, still
 present in readline 6.1 in 2010) from that file gives some
 insight:

/*&nbsp;This&nbsp;is&nbsp;the&nbsp;stuff&nbsp;that&nbsp;is&nbsp;hard&nbsp;for&nbsp;me.&nbsp;&nbsp;I&nbsp;never&nbsp;seem&nbsp;to&nbsp;write&nbsp;good 
 &nbsp;&nbsp;&nbsp;display&nbsp;routines&nbsp;in&nbsp;C.&nbsp;&nbsp;Let's&nbsp;see&nbsp;how&nbsp;I&nbsp;do&nbsp;this&nbsp;time.&nbsp;*/

</blockquote>

John Greco suggests a different approach <a href=
 "http://lists.gnu.org/archive/html/bug-ncurses/2010-10/msg00041.html">
 here</a>, which discards ncurses' input functions (such as
 <code>wgetch</code>), using <code>rl_callback_read_char</code> to
 fill the <code>rl_line_buffer</code> array. That can be printed
 using ncurses.

<h4 id="setvbuf_once-id"><a name="setvbuf_once" id=
 "setvbuf_once">Problems with Output Buffering</a></h4>

Ncurses shares with SVr4 curses a limitation which is
 documented in the C standard. To attain good performance, they
 buffer output data, e.g., with the <code>setvbuf</code> function
 (or equivalent, depending on the platform). The performance gain
 is noticable.

However, if your application spawns a subprocess, it will
 inherit the output stream from ncurses&mdash;still buffered. On
 several platforms this results in odd behavior, since normally
 the standard output is line buffered, making the output flushed
 at the end of each line. To solve this problem, your application
 should disable <code>setvbuf</code> before invoking the
 subprocess and restore it when resuming. That is, it should, but
 often cannot&mdash;that is the problem. The standard says that
 <code>setvbuf</code> must be called only after opening a stream
 and before performing any reads or writes with that stream.

If your application calls <code>initscr</code>, it uses the
 standard output, which ncurses assumes has not been written to,
 to which ncurses then applies buffering. (Caveat: The
 standard writers neglected to provide a mechanism for determining
 if the stream is indeed buffered). Adding a call to
 <code>setvbuf</code> to disable buffering may work or not. In at
 least one implementation, the C library continues using the
 buffer after the buffer is disabled, even if an
 <code>fflush</code> is first given. That is, it will produce a
 core dump.

The fix? Use <code>newterm</code> to initialize ncurses and
 manage the input and output streams yourself. For instance, you
 may simply open <code>/dev/tty</code> for input and output, and
 leave the standard input and output alone.

<h3 id="problems_xwindows-id"><a name="problems_xwindows" id=
 "problems_xwindows">Mice and Windows</a></h3>

<h4 id="xterm_paste-id"><a name="xterm_paste" id="xterm_paste">I
 can't cut/paste in xterm</a></h4>

This is a general FAQ relating to xterm. When an application
 sets xterm to any of its mouse tracking modes, it reserves the
 unshifted mouse button clicks for the application's use. Unless
 you have modified the treatment of the shifted mouse button
 events (e.g., with your window manager), you can always do
 cut/paste by pressing the shift key while clicking with the
 mouse.

Ncurses sets the mouse tracking mode as a result of your
 application's calls to <code>mousemask</code>, which is an
 extension.

<h4 id="handle_resize-id"><a name="handle_resize" id=
 "handle_resize">Handling SIGWINCH (resize events)</a></h4>

It is possible to make an application resize when
 running in a windowing environment (e.g., in an
 <code>xterm</code>). This is not a feature of standard SVr4
 curses, though some curses implementations (e.g., HP-UX) support
 this.

Within ncurses, there are two ways to accomplish this. One
 relies on side-effects of the library functions, and is
 moderately portable. The other is an extension to SVr4
 curses.

<ul>
 <li>
 <code>endwin</code>/<code>refresh</code> when invoked will
 briefly exit curses and reinitialize the display, picking up
 the new screen size. Ncurses will reallocate the WINDOW data
 (e.g., curscr, stdscr) to reflect the new limits.
 </li>

<li>
 <code>resizeterm</code> can be invoked directly to make
 ncurses resize its WINDOW data. I use it in my directory
 editor <a href="/ded/ded.html">ded</a> to achieve
 flicker-free resizing via a signal handler for SIGWINCH. (The
 documentation for HP-UX curses implies that they use a
 similar approach; I have been unable to make it work.)
 </li>
 </ul>

Ncurses 5.0 can be configured to establish its own SIGWINCH
 hander. In this configuration, the <code>wgetch</code> function
 will return a special keycode <code>KEY_RESIZE</code> when a
 resizing event is detected. The signal handler also sets a flag
 which the library checks with <code>is_term_resized</code>, then
 calls <a href=
 "/ncurses/man/resizeterm.3x.html"><code>resizeterm</code></a>
 (Caveat: <code>malloc</code> and
 <code>free</code> are not guaranteed to be safe for use in a
 signal handler).

There is a known problem using bash's checkwinsize
 misfeature. See Novell <a href=
 "https://bugzilla.novell.com/show_bug.cgi?id=828877">#828877</a>,
 as well as the thread starting <a href=
 "http://lists.gnu.org/archive/html/bug-bash/2013-07/msg00054.html">
 here</a> on bug-bash, and of course the manpage for <a href=
 "/ncurses/man/curs_util.3x.html">use_env</a>. Shells which set
 <code>LINES</code> and <code>COLUMNS</code> might have been a
 good idea in 1995, but not as a rule more recently than that.

<h4 id="using_gpm_lib-id"><a name="using_gpm_lib" id=
 "using_gpm_lib">Linking with GPM (Linux console mouse
 library)</a></h4>

Ncurses works correctly with the Linux GPM (general purpose
 mouse) library. However, early on some distributors tampered with
 the library, making it not generally useful, by linking in a
 portion of the BSD curses library to satisfy references for
 <code>Gpm_Wgetch</code>. That prevented one from using ncurses'
 GPM support.

GPM provides limited support for xterm mouse control
 sequences. This is implemented in <code>GPM_Wgetch</code>, which
 makes some unreasonable assumptions about the curses library's
 internal behavior of <code>wgetch</code>. In particular, GPM
 interferes with the logic which combines characters into
 function-key codes. GPM also uses as part of its xterm control
 sequences a pair which save/restore the mouse mode (and are not
 actually handled by any of the other terminal emulators).

On writing this faq in 1999, there were no
 applications that used this misfeature. Later (in 2005) there
 were still no curses applications which do, however w3m contains
 some contorted code to exploit this, by abusing the library
 interface: it defines several symbols that conflict with ncurses
 to intercept calls to <code>wgetch</code>, while using other
 symbols from ncurses as is. (There is also documented
 <code>Gpm_Getch</code>, but it is no longer present in the GPM
 source code).

Since version 1.10 GPM comes with a configure script, which
 allows the system builder to suppress this from the shared
 library, e.g.,

configure&nbsp;--without-curses 
 
 </blockquote>

You should verify that the shared library does not use the
 symbol <code>wgetch</code>. Version 1.16 lacked the configure
 script option to suppress this hook; removing libcurses.o from
 the list of objects in GPM's Makefile worked just as well.
 Version 1.17 built correctly when I tested it, however, though
 the changelog does not mention the change. It would seem that the
 issue would be long resolved. However, it is not.

Starting with ncurses 5.5, the recommendation is still the
 same: build the GPM library without the <code>Gpm_Wgetch</code>
 interface. ncurses 5.5 can <a href=
 "NEWS.html#t20041009">dynamically load</a> the GPM library on
 Linux, and that eliminates any reason to have the ncurses library
 built with an explicit dependency upon GPM.

Some of the GPM fixes are based on a quirk of the library: if
 <code>Gpm_Open</code> is called when the <code>TERM</code>
 environment variable contains "xterm", it opens a connection
 which returns the raw character stream which might contain mouse
 escape sequences. It returns a special file descriptor (-2) which
 is easy to overlook in the normal checks on file descriptors
 which are valid only when positive. For quite a while as well,
 GPM and X could not co-exist. It was not uncommon to have some
 safeguards to turn off GPM when starting X. By around 2005, some
 fixes had been made to the X server to allow the two to run
 concurrently. That exposed some problems in ncurses which had not
 properly checked for the special file descriptors, and affected
 programs running in an xterm, e.g., <a href=
 "/dialog/dialog.html">dialog</a> and <a href=
 "/ded/ded.html">ded</a> which start/stop the mouse interface. For
 these, the GPM server would ultimately be locked up and not
 return from a call to <code>Gpm_Open</code>.

<h3 id="problems_packages-id"><a name="problems_packages" id=
 "problems_packages">Problems with Packages</a></h3>

<h4 id="packages_buggy-id"><a name="packages_buggy" id=
 "packages_buggy">Comments on packages</a></h4>

Packages are a good thing. Sometimes.

Not all distributions clearly distinguish between release
 versions of software, betas and alpha versions. (To be fair, not
 all producers distinguish these properly).

<h4 id="version4_or_not-id"><a name="version4_or_not" id=
 "version4_or_not">Ncurses 4/4.2/5.0</a></h4>

Ncurses 5.0 was not compatible with ncurses 4.2, however I
 frequently saw people advising others to "fix" programs that
 require the older library to make a symbolic link from the newer
 name to the older. That only works for simple applications, and
 not all of those.

Ncurses 4.0 and 4.2 were released respectively before and
 after X/Open finished their curses specification. Both were based
 on the draft specifications from 1995 and 1996. The released
 specification (available in 1997) differs in several places,
 mainly in the provisions for multi-byte character sets.

Late in 1998 through early 1999, I
 made corrections to the development version of ncurses to align
 it to the X/Open specification. Near the end of this, I realized
 that I had an opportunity to add an extension to ncurses which
 would make the terminfo format extensible, just as termcap is. It
 required a change to the <code>term.h</code> header. to allow the
 arrays for terminfo booleans, numbers and strings to be set at
 runtime. This had the effect of making programs not
 binary-compatible, but that was not a drawback, since it was
 already conceded that ncurses 5.0 would not be binary-compatible
 with ncurses 4.2 because of the X/Open changes. Only a few
 applications use <code>term.h</code>, and those would be fixed by
 a recompile.

One problem: Redhat packaged development versions of ncurses
 without distinguishing them from the release versions. We
 discussed the matter with them, but they did not wish to
 cooperate. Redhat 6.0 was released with almost all of the
 interface changes that comprised ncurses 5.0&mdash;as "ncurses
 4.2". When ncurses 5.0 was released, they did not bother to read
 the release notes, and released that as "ncurses 4.2". Somewhat
 later, they added to the confusion by calling it "ncurses 4.0".
 Until mid-2001, much of this information was still available in
 Bugzilla.

Redhat continues to distribute development versions of ncurses
 without distinguishing between release- and
 development-versions.

<h4 id="colorfgbg_bug-id"><a name="colorfgbg_bug" id=
 "colorfgbg_bug">rxvt's <code>$COLORFGBG</code> variable</a></h4>

Ncurses 5.2 added an experimental feature: support for rxvt's
 <code>$COLORFGBG</code> variable. This is a feature which tells
 the application what colors the default foreground and background
 correspond to. It is specific to rxvt: in general other terminal
 emulators assign colors for foreground and background which do
 not necessarily correspond to any of the ANSI colors. This
 feature was enabled in some rpm-based distributions, e.g.,
 Mandrake and Redhat.

It worked for the configuration on which I tested, however
 there are two configurations. The format of the
 <code>$COLORFGBG</code> variable is not documented; you must read
 the C code to find how rxvt sets it. The two configurations
 correspond to whether the xpm library is used or not. If xpm is
 used, ncurses 5.2 sees the wrong value for the background, and
 display black-on-black.

Quick fix: unset <code>$COLORFGBG</code>. 
 Better fix: update the ncurses rpm (Mandrake did).

<h3 id="no_urxvt-id"><a name="no_urxvt" id="no_urxvt">Where is
 rxvt-unicode?</a></h3>

See the <a href="ncurses-urxvt.html">rxvt(-unicode) in
 ncurses</a> page.

<h2 id="report_bugs-id"><a name="report_bugs" id=
 "report_bugs">How do I report bugs?</a></h2>

First, check to see if your problem is addressed in this FAQ.
 Read the INSTALL document, if you have not done so. However, it
 may not be a known problem. Read on.

<ul>
 <li><a href="#how_to_report">How should I report bugs?</a></li>

<li><a href="#problems_building">How do I report problems
 building ncurses?</a></li>

<li><a href="#dbupdate">Updates to terminal database</a></li>

<li><a href="#backlog">Why aren't my bugs being fixed?</a></li>

<li><a href="#patch_summary">How are patches organized?</a></li>
 </ul>

<h3 id="how_to_report-id"><a name="how_to_report" id=
 "how_to_report">How should I report bugs?</a></h3>

Contact the current maintainers at <a href=
 "mailto:bug-ncurses@gnu.org">bug-ncurses@gnu.org</a>.

To join the ncurses mailing list, please write email to
 <code>bug-ncurses-request@gnu.org</code> containing the line:

<blockquote>
 <pre>
subscribe &lt;name&gt;@&lt;host.domain&gt;
</pre>
 </blockquote>

This list is open to anyone interested in helping with the
 development and testing of this package.

There is an archive of the mailing list here:

<blockquote>
 <a href=
 "http://lists.gnu.org/archive/html/bug-ncurses/">http://lists.gnu.org/archive/html/bug-ncurses</a>
 (also <a href=
 "https://lists.gnu.org/archive/html/bug-ncurses/">https</a>)
 </blockquote>

Otherwise, you may email directly to the maintainers,
 currently:

<ul>
 <li><a href="mailto:dickey@invisible-island.net">Thomas E.
 Dickey &lt;dickey@invisible-island.net&gt;</a></li>
 </ul>

If you send email only to one of the other authors, I will
 probably not see it. I prefer that bug reports go to the mailing
 list. (Occasionally I get private email cc'ing Zeyd and
 Eric, neither of whom has contributed since before 2000).

I get some of my bug reports via the ncurses mailing list,
 some via bug-tracking systems, and the others via direct
 email.

The mailing list is useful. When it began, more than half of
 the changes that were introduced without review in the ncurses
 mailing list introduced a bug. So I find it necessary to review
 proposed changes.

Refer to the standard <a href=
 "/personal/bug-reports.html">reporting/patching</a> procedures,
 for a start.

When sending patches, specify the version of ncurses that you
 are patching. All versions of ncurses have major, minor and
 patch-date numbers. Those may not be the same as a package
 revision number; it helps to identify that, so that I can relate
 it to the source.

There is a defunct "help-ncurses" mailing list, which I
 requested to be shut down long ago. I am not subscribed to that
 list.

<h3 id="problems_building-id"><a name="problems_building" id=
 "problems_building">How do I report problems building
 ncurses?</a></h3>

This is a little different from reporting bugs. If you have a
 machine that I've not ported to, and have problems, I'll require
 the relevant information:

<ul>
 <li><code>config.log</code>
 </li>

<li><code>config.status</code>
 </li>

<li><code>include/ncurses_cfg.h</code>
 </li>

<li>log from running <code>configure</code>, with options</li>

<li>log from running <code>make</code>, with options</li>
 </ul>

A MIME attachment (preferably of the compressed
 files) is preferred, because logfiles can be awkward to send and
 receive via email. You may find the <a href=
 "/scripts/readme.html">scripts</a> which I use for building and
 saving logfiles useful.

If you're having trouble building on a known "good" platform,
 please make sure that you've got a current version of ncurses,
 and please read the installation instructions.

<h3 id="dbupdate-id"><a name="dbupdate" id="dbupdate">Updates to
 terminal database</a></h3>

I attempt to verify users's proposed changes to the terminal
 database. Most of those are for terminal emulators, although
 there are still (2018) occasional updates for hardware terminals.
 Calling them all &ldquo;applications&rdquo;, there are some
 limitations which must be mentioned, based on developers' past
 abuse of common-sense guidelines:

<ul>
 <li>
 Updates for applications which are not generally available
 in a finished form are not suitable for
 inclusion.

This is not the place for publishing in-development work
 which might change (or might not even be correct).

Start by stating which packaged version of an application
 should be used for validating the updates.
 </li>

<li>
 Terminal descriptions for applications which have some
 &ldquo;string attached&rdquo; (license fee, non-disclosure
 restriction, etc.), might be suitable for inclusion provided
 that there is no implication that they are the same as some
 less restrictively available application.

For instance, there will be no aliases added in the
 less-restrictive application's terminal description
 mentioning the more-restrictive one.
 </li>
 </ul>

Given those limitations, keep in mind that the terminal
 descriptions have to work with ncurses:

<ul>
 <li>
 Keep in mind that ncurses displays characters on
 terminals and get input (e.g., read characters) from
 the terminal. The predefined capabilities are used for one or
 both of those goals. Terminal descriptions may have
 user-defined capabilities, but if those capabilities are used
 to tell applications how to display on a terminal or
 get input from a terminal, they have be relevant to
 ncurses.
 </li>

<li>
 Applications which infer characteristics of a terminal
 from the name of a terminal description are not
 relevant to ncurses. No changes will be made to the terminal
 database to accommodate those applications.
 </li>
 </ul>

Furthermore:

<ul>
 <li>
 I use <a href="man/tic.1m.html">tic</a>'s consistency
 checks, <a href="tack.html">tack</a> and <a href=
 "/vttest/vttest.html">vttest</a>, as well as custom scripts
 for testing basic functionality.
 </li>

<li>
 When users provide custom scripts or programs which
 demonstrate a given feature, that helps a lot.
 </li>

<li>
 If the source-code for the application is published, I
 will study that to amend the proposed changes. Often,
 developers propose changes based on cut/paste from other
 descriptions. Testing and review eliminate most of that
 problem.
 </li>

<li>
 If a feature is present in the application but does not
 work as other implementations do, it will be removed or
 cancelled from the terminal description.
 </li>

<li>
 Applications whose terminal descriptions which simply
 repeat another description might be considered for inclusion
 as aliases, but only if those have wide-spread
 recognizably different appearance.
 </li>
 </ul>

<h3 id="backlog-id"><a name="backlog" id="backlog">Why aren't my
 bugs being fixed?</a></h3>

Sorry. This is a hobby. There's a large backlog. Some changes
 pass review quickly, others are difficult, because one fix may
 break other functionality. My criteria are less stringent if you
 provide a short program that demonstrates the problem, or if
 you're modifying something that you maintain.

In any case, I will incorporate patches into my beta version
 only if I have reviewed the patch, tested it (if the patch is not
 obvious), and repaired any omissions (e.g., portability
 constraints). Occasionally I have patches (including my own)
 which cannot pass immediate review; these constitute most of my
 backlog. A small part of my backlog consists of issues which
 highlight incompatibilities between ncurses and SVr4 curses;
 these are listed in the TO-DO file.

I use the following guidelines:

<ul>
 <li id="bugnot-svr4docs">
 extensions (deviations from SVr4 curses) are allowed only
 if they do not modify the documented/observed behavior of the
 API.
 </li>

<li id="bugnot-internals">
 extensions are feature changes which cannot be implemented
 without modifying the library.

For example, if you really need a function which
 interprets ANSI escape sequences, that can be done without
 modifying the library.
 </li>

<li id="bugnot-svr4testing">
 I test behavior using unix
 curses (Solaris and others), and use the vendor's manual
 pages in conjunction with the X/Open Curses
 documentation.
 </li>

<li id="bugnot-xpg4curses">
 The Solaris XPG4 curses implementation is known to be
 badly broken; its SVr4 curses is usable for comparisons.
 </li>

<li id="using_illumos">
 The OpenSolaris source code (available via <a href=
 "https://github.com/illumos/illumos-gate">Illumos</a>) is
 useful for gaining insight (and clarifying documentation),
 because it includes the Solaris SVr4 and XPG4 curses
 sources. 
 However, Solaris' XPG4 curses differs from other vendors'
 implementations.
 </li>
 </ul>

<h3 id="patch_summary-id"><a name="patch_summary" id=
 "patch_summary">How are patches organized?</a></h3>

Prior to version 4.0 I posted patches to the ncurses mailing
 list summarizing only my changes (after applying changes
 submitted by others). The intent was that people who followed the
 list closely could build developmental versions.

Generally (unless we find a serious error), I issue patches on
 Saturdays, since validating patches takes time.

Beginning with version 4.0, I maintain "complete" patches (my
 changes together with those that I have integrated). It is
 simpler, and does not require making complete snapshots as
 often.

<ul>
 <li>
 User-supplied patches that require no modification are
 always tagged with their name, making it simpler to follow
 the revision history.
 </li>

<li>
 User-supplied patches that require some modification are
 still tagged, but my patch will have followup changes.
 </li>
 </ul>

Most files have RCS identifiers. If you are maintaining
 ncurses in an RCS (or CVS, etc.) archive, you can keep in sync
 with this using the "-k" option of <code>ci</code>.

<h2 id="posix_compliant-id"><a name="posix_compliant" id=
 "posix_compliant">POSIX Compliant?</a></h2>

That depends.

The library and utilities use (where possible) POSIX
 interfaces.

However, X/Open Curses is not part of POSIX (the
 1003.x series, referred to as the Base
 Specifications). It is a different standard, updated less
 frequently, and bundled as part of the <a href=
 "http://www.unix.org/version4/overview.html">Single
 UNIX Specification</a>.

Quoting from Advanced Programming in the UNIX
 Environment, third edition (Stevens and Rago):

<blockquote class="code-block">
 The third version of the Single UNIX Specification (SUSv3)
 was published by The Open Group in 2001. The Base
 Specifications of SUSv3 are the same as IEEE Standard
 1003.1-2001 and are divided into four sections: Base
 Definitions, System Interfaces, Shell and Utilities and
 Rationale. SUSv3 also includes X/Open Curses Issue 4, Version
 2, but this specification is not part of POSIX.1.

...

In 2003, the Single UNIX Specification was updated,
 including corrections and new interfaces, removing obsolete
 interfaces, and marking other interfaces as being obsolescent
 in preparation for future removal. Additionally some previously
 optional interfaces were promoted to nonoptional status,
 including asynchronous I/O, barriers, clock selection, memory
 mapped files, memory protection, reader-writer locks, real-time
 signals, POSIX semaphores, spin locks, thread-safe functions,
 threads, timeouts, and timers. The resulting standard is known
 as Issue 7 of the Base Specifications, and is the same as
 POSIX-1.2008. The Open Group bundled this version with an
 updated X/Open Curses specification and released them as
 version 4 of the Single Unix Specification in 2010. We'll refer
 to this as SUSv4.
 </blockquote>

Additionally, ncurses

<ul>
 <li>provides extensions (features not addressed by
 X/Open Curses),</li>

<li>is sometimes different from X/Open Curses,
 and</li>

<li>provides extra libraries and programs (<a href=
 "man/form.3x.html">form</a>, <a href=
 "man/menu.3x.html">menu</a>, <a href=
 "man/panel.3x.html">panel</a> as well as <a href=
 "man/toe.1m.html">toe</a>, <a href=
 "man/tset.1.html">tset</a>) 
 not documented by X/Open Curses.</li>
 </ul>

<h2 id="y2k_compliant-id"><a name="y2k_compliant" id=
 "y2k_compliant">Y2K Compliant?</a></h2>

Certainly.

The ncurses library does not store or retrieve dates in any
 form that depends on the year. Ncurses' use of time information
 is limited to

<ul>
 <li>computing the elapsed times, in fractions of a second,</li>

<li>comparing file creation times, and</li>

<li>sample programs that display the current time of day.</li>
 </ul>

<hr>

<h2 id="terminology-id"><a name="terminology" id=
 "terminology">Terminology</a></h2>

Ncurses is used primarily with
 terminal emulators. A few people use it with
 hardware terminals.

<dl>
 <dt>Terminals</dt>

<dd>
 Short for data terminals,
 terminals provide users with the ability to
 enter and view data. The word terminal refers to
 something that is the endpoint.

Terminals are generally not part of the computer, nor are
 they necessarily near the computer. Originally, terminals
 were connected to the computer by cables, later by local
 networks. The terminal emulator running in your computer's
 desktop environment is a special case&mdash;still using a
 network connection, but perhaps sharing keyboard and display
 with other terminal emulators.
 </dd>

<dt>Emulators</dt>

<dd>
 An emulator of course is something which
 behaves like something else. In the case of terminals, these
 are programs which provide the same functionality as hardware
 terminals, or even other terminal emulator programs.
 </dd>

<dt>Control sequences</dt>

<dd>
 Often referred to as escape sequences because
 many begin with the ASCII escape-character, control
 sequences are sequences of characters sent to a
 terminal to make it perform some operation. Most terminals
 (the ones that ncurses deals with) are asynchronous,
 and accept control sequences which update parts of the
 display. There are others, e.g., the synchronous
 "block-oriented" terminals such as the IBM 3278.

The VT100's control sequences are used in many terminal
 emulators for example. In turn, most of those control
 sequences follow standards established by ANSI, ISO, ECMA.
 These are the most useful (and accessible):

<ul>
 <li><a href=
 "http://www.ecma-international.org/publications/standards/Ecma-035.htm">
 ECMA-35 - Character Code Structure and Extension
 Techniques</a></li>

<li><a href=
 "http://www.ecma-international.org/publications/standards/Ecma-048.htm">
 ECMA-48 - Control Functions for Coded Character Sets</a></li>
 </ul>

The standards do not give exact functional definitions.
 For that, ncurses development relies on vendor documentation,
 analysis and comparison testing.
 </dd>

<dt>Consoles</dt>

<dd>
 Originally console referred to the front
 panel of a computer, e.g., lights and switches, as
 well as the attached operator's terminal. The switches would
 do more than just turn the computer on and off.

<ul>
 <li>
 Some machines provided special purpose switches which
 could enter data or control the computer's operation.
 </li>

<li>
 Later, the front panel was not so important for this
 purpose, and the operator's terminal was treated as the
 console. Many operating systems have a special terminal
 device for the system console terminal, e.g,.

<li><code>/dev/console</code> on Unix-like
 systems.</li>

<li>
 Still later, with video terminals becoming more common
 than printing terminals, the console terminal became an
 integral part of the computer. Several operating systems
 implement their console terminal as part of the kernel,
 e.g., Solaris, FreeBSD, Linux. Many people are confused
 by this, and refer to these as hardware consoles
 though they are not. They are programs. Generally they
 are not referred to as emulators simply because
 they usually are not made to imitate another
 terminal.
 </li>

<li>
 The console terminal now is used mainly when
 installing or reconfiguring the computer, rather than for
 routine control of its programs. Some system calls treat
 it specially, by using it as the destination for messages
 to the operator.
 </li>
 </ul>

The X Window system can be run on many of the console
 terminals. Sun's workstations for example did this. Few
 people used the console terminal on that hardware (which by
 the way was not VT100-compatible). Since X took over the
 computer's display, some way was needed to make the operator
 messages visible. The <a href=
 "http://www.x.org/archive/X11R6.8.1/doc/xconsole.1.html">xconsole</a>
 program did that; desktop systems provide analogous
 solutions. It is not necessary to use a terminal emulator to
 display operator messages. A few terminal emulators (such as
 <a href="/xterm/">xterm</a>) can intercept console messages
 as an optional feature. The <a href=
 "/xterm/xterm.faq.html#bug_konsole">konsole</a> program on
 the other hand, does not perform this function,
 notwithstanding its name. Rather, it is a terminal
 emulator.

Microsoft Windows's treatment of console devices has
 changed:

<ul>
 <li>
 In MS-DOS, there was one display device.
 </li>

<li>
 In Windows, there can be several console windows, each
 with its own console device. You can see the
 characteristics of the corresponding device by entering
 the command <code>mode con</code> into a console window.
 They are referred to as console windows because they
 provide partial compatibility with the MS-DOS console
 commands (including the device name
 <code>CON:</code>).
 </li>

<li>
 They are not terminal emulators however: these console
 windows do not by themselves respond to control
 sequences. MS-DOS provided a special device driver
 for this purpose called <code>ANSI.SYS</code>.
 </li>

<li>
 Windows provides an application programming interface
 (API) by which programs can treat the console window as a
 display.
 </li>

<li>
 Several terminal emulators have been written using the
 Windows <a href=
 "http://msdn.microsoft.com/en-us/library/windows/desktop/ms682087(v=vs.85).aspx">
 console API</a> which run in one of these console windows
 (for example, the Cygwin terminal before <a href=
 "http://code.google.com/p/mintty/">mintty</a>).
 </li>

<li>
 There is also a port of ncurses using MinGW to these
 console windows. However, the ncurses MinGW port is a
 special case not using a terminal emulator. Like
 pdcurses, ncurses implements the Windows port using the
 <a href=
 "http://msdn.microsoft.com/en-us/library/windows/desktop/ms682073(v=vs.85).aspx">
 Windows Console API</a>. It does not rely upon
 <code>ANSI.SYS</code> or any of its successors.
 </li>
 </ul>

Like the Windows Console API, neither ncurses or pdcurses
 are terminal emulators. However, it is possible to write a
 terminal emulator using any of these application programming
 interfaces.
 </dd>
 </dl>

<h2 id="additional_reading-id"><a name="additional_reading" id=
 "additional_reading">Additional Reading</a></h2>

<ul>
 <li><a href="#for_reference">For reference</a></li>

<li><a href="#other_books">Dead trees</a></li>

<li><a href="#other_langs">Language bindings</a></li>

<li><a href="#other_impls">Other implementations</a></li>

<li><a href="#other_appls">Related applications</a></li>

<li><a href="#ref_obsolete">Technically obsolete, but often
 cited</a></li>

<li><a href="#ref_reallyold">Really old stuff</a></li>

<li><a href="#ref_misleading">Interesting but misleading</a></li>
 </ul>

<h3 id="for_reference-id"><a name="for_reference" id=
 "for_reference">For reference:</a></h3>

<ul>
 <li>
 The <a href="/ncurses/man/index.html">manpages</a>, of
 course.
 </li>

<li>
 X/Open Curses is not part of POSIX (the 1003.x
 series, referred to as the Base Specifications). It
 is a different standard, updated less frequently, and bundled
 as part of the <a href=
 "http://www.unix.org/version4/overview.html">Single
 UNIX
 Specification</a>.

Here are some useful links

<ul>
 <li>
 <a href=
 "http://www.opengroup.org/publications/catalog/C094.htm">19
 Apr 2013 &ndash; Issue 7</a>
 </li>

<li>
 <a href=
 "http://www.opengroup.org/publications/catalog/C610.htm">15
 Jul 1996 &ndash; Issue 4, Version 2</a>
 </li>

<li>
 <a href=
 "http://www.opengroup.org/onlinepubs/7908799/cursesix.html">
 X/Open Curses, Issue 4, Version 2, Reference Pages</a>
 </li>
 </ul>
 </li>

<li>
 These are based on ncurses:

<ul>
 <li>
 <a href=
 "http://tldp.org/HOWTO/NCURSES-Programming-HOWTO/">NCURSES-Programming-HOWTO</a>.
 (another copy <a href=
 "/ncurses/howto/NCURSES-Programming-HOWTO.html">here</a>
 and <a href=
 "/ncurses/howto/ncurses-howto-20050620.zip">source</a>
 for the SGML and programs)
 </li>

<li>
 <a href=
 "http://www.c-for-dummies.com/ncurses/">Programmer's
 Guide to NCurses</a> by Dan Gookin. Wiley, 2007. 
 (my involvement is mentioned <a href=
 "/personal/paperstuff.html#acknowledgements">here</a>).

The source code for the examples is available on
 <a href=
 "https://github.com/wkoszek/ncurses_guide">Koszek's
 page</a>.
 </li>
 </ul>
 </li>

<li>
 These are useful reading as background material:

<ul>
 <li>
 <a href=
 "https://docs.oracle.com/cd/E36784_01/html/E36880/curses-3curses.html#scrolltoc">
 Solaris 11 Reference Manual Collection, 
 manual pages section 3: Curses Library Functions</a>
 </li>

<li>
 <a href=
 "http://publibn.boulder.ibm.com/doc_link/en_US/a_doc_lib/aixprggd/genprogc/curses.htm">
 AIX General Programming Concepts: Writing and Debugging
 Programs, 
 Chapter 2: The Curses Library</a>
 </li>

<li>
 <a href=
 "http://uw714doc.sco.com/en/man/html.3ocurses/Intro.3ocurses.html">
 Unixware curses</a> (a rare reference to &ldquo;POSIX
 curses&rdquo;).
 </li>

<li>
 <a href=
 "http://polarhome.com/service/man/?qf=curses&amp;tf=2&amp;of=IRIX&amp;sf=3X">
 IRIX64 curses</a>
 </li>
 </ul>
 </li>
 </ul>

<h3 id="other_books-id"><a name="other_books" id=
 "other_books">Dead trees:</a></h3>

Intentionally ignoring the &ldquo;door stops&rdquo; which
 reprint manual pages, these books (long out of print) give useful
 information:

<ul>
 <li>
 <a name="book_goodheart" id="book_goodheart">UNIX
 Curses Explained</a>. Berny Goodheart. Prentice Hall of
 Australia Pty Ltd. 1991. 287 pages. ISBN: 0-13-931957-5.

Goodheart's book is useful for its historical perspective:
 he frequently mentions when a feature was added, and how it
 fit in.
 </li>

<li>
 <a name="book_strang" id="book_strang">termcap &amp;
 terminfo</a>. John Strang, Linda Mui and Tim O'Reilly.
 O'Reilly &amp; Associates, Inc., Sebastopol, CA, USA. Third
 edition with corrections, July 1992. 253 pages. ISBN:
 0-937175-22-6.

Originally published in 1985 to describe termcap
 and then updated to cover terminfo, this is useful,
 to a point: it omits any discussion of color, and the
 discussion of line-drawing is cursory (superficial).

Strang also wrote Programming with Curses (1986),
 but that is not useful because it describes only 4.3BSD
 curses.
 </li>

<li>
 <a name="book_gomez" id="book_gomez">Character User
 Interface Programming (UNIX SVR4.2)</a>. Edited by Claro
 Gomez. UNIX Press, Prentice-Hall, Inc. 1993. 720
 pages. ISBN: 0-13-042581-8.

The SVr4.2 book Programming with UNIX System
 Calls (1992) refers to this book for details on FMLI
 (not implemented by ncurses) and Extended Terminal
 Interface (ETI). ncurses provides libraries which
 correspond to the latter. The book is presented in tutorial
 manner, not a technical reference as such. It refers the
 reader to the manual pages for details. Interestingly, the
 system calls book's section mentioning Character User
 Interface Programming illustrates a technical blunder by
 implying that these two lines are the same length:

<blockquote>
 <pre class="code-block">
ETI is a great package for creating forms and menus.
ETI is the best package for creating forms and menus.
</pre>
 </blockquote>

About 40% of the book (chapters 6 through 13) covers the
 ETI consisting of the <a href="man/panel.3x.html">panel</a>,
 <a href="man/menu.3x.html">menu</a>, and <a href=
 "man/form.3x.html">form</a> libraries as well as the soft-key
 support added to SVr4.2 curses. Only the last feature
 (<a href="man/curs_slk.3x.html">soft-keys</a>) was
 incorporated in <a href=
 "http://pubs.opengroup.org/onlinepubs/7908799/xcurses/slk_clear.html">
 X/Open Curses</a>.
 </li>
 </ul>

<h3 id="other_langs-id"><a name="other_langs" id=
 "other_langs">Language bindings:</a></h3>

<ul>
 <li>
 <a href="/ncurses/Ada95.html">Ada95 binding, from
 ncurses</a>
 </li>

<li>
 <a href=
 "http://www.pitman.co.za/projects/charva/index.html">CHARVA
 (Java binding)</a>
 </li>

<li>
 <a href=
 "http://common-lisp.net/project/cl-ncurses/">Common Lisp
 binding</a>
 </li>

<li>
 <a href=
 "http://efsa.sourceforge.net/archive/ecurses/ecurses.htm">Eiffel
 binding</a>
 </li>

<li>
 <a href=
 "http://spiderape.sourceforge.net/plugins/ncurses/">JavaScript
 binding</a>
 </li>

<li>
 <a href=
 "http://www.mtu-net.ru/aborovsky/lincrt/lincrt-lib.html">LinCRT
 (Kylix binding)</a>
 </li>

<li>
 <a href=
 "http://caml.inria.fr/cgi-bin/hump.en.cgi?contrib=100">OCaml
 binding</a>
 </li>

<li>
 <a href=
 "http://search.cpan.org/search%3Fmodule=Curses">Perl
 binding</a>
 </li>

<li>
 <a href="http://www.php.net/manual/en/ref.ncurses.php">PHP
 binding</a>
 </li>

<li>
 <a href=
 "http://docs.python.org/library/curses.html">Python
 binding</a>
 </li>

<li>
 <a href="http://ncurses-ruby.berlios.de/">Ruby binding</a>
 </li>

<li>
 <a href=
 "http://www.schwartzcomputer.com/tcl-tk/tcl-tk.html">Tcl/Tk
 Extensions &amp; Information Page -&gt; ncurses cTk</a>
 </li>
 </ul>

<h3 id="other_impls-id"><a name="other_impls" id=
 "other_impls">Other implementations:</a></h3>

<ul>
 <li>
 <a name="pdcurses" href=
 "http://pdcurses.sourceforge.net/" id="pdcurses">PDCurses</a>
 </li>

<li>
 <a name="netbsd" href=
 "/ncurses/ncurses-netbsd.html" id="netbsd">NetBSD</a>
 </li>
 </ul>

<h3 id="other_appls-id"><a name="other_appls" id=
 "other_appls">Related applications:</a></h3>

<ul>
 <li>
 <a href="http://xterminal.sourceforge.net/">Xterminal's
 Home Page</a>
 </li>

<li>
 <a href="/cdk/cdk.html">cdk</a> Curses Development Kit (a
 library of widgets)
 </li>

<li>
 <a href="/dialog/dialog.html">dialog</a> Script-driven
 curses widgets
 </li>
 </ul>

<h3 id="ref_obsolete-id"><a name="ref_obsolete" id=
 "ref_obsolete">Technically obsolete, but often cited:</a></h3>

<ul>
 <li>
 <a href=
 "https://web.archive.org/web/20060315053358/http://docs.freebsd.org/44doc/#psd">
 4.4BSD Documents -&gt; original curses</a>
 </li>

<li>
 <a href=
 "http://www.ditec.um.es/~piernas/manpages-es/otros/tutorial-ncurses.html">
 Escribir Programas con NCURSES</a>
 </li>

<li>
 <a href="/ncurses/ncurses-intro.html">Writing Programs
 with NCURSES</a>
 </li>

<li>
 <a href="hackguide.html">A Hacker's Guide to Ncurses
 Internals</a>
 </li>
 </ul>

<h3 id="ref_reallyold-id"><a name="ref_reallyold" id=
 "ref_reallyold">Really old stuff</a></h3>

<ul>
 <li>
 As mentioned, the <a href="#using_illumos">illumos</a>
 sources are useful for comparison&mdash;up to a point. Those
 are based on the OpenSolaris release in 2005. It was not a
 complete Solaris source-release. For instance, some parts
 (such as FMLI) were not included.

The copyright notices are helpful, but not definitive
 because they reflect only when someone recorded that a file
 was added to the AT&amp;T sources, rather than when the
 program was written.
 </li>

<li>
 For earlier implementations, the <a href=
 "http://www.mckusick.com/csrg/">CSRG cdroms</a> and Jonathan
 Gray's <a href=
 "https://github.com/jonathangray/csrg">repository</a>
 (converted from SCCS) are helpful. Both have problems: the
 2BSD snapshots include files dated later than 3BSD, while the
 SCCS history goes back only to the end of 1979 (with six
 checkins), and checkins were not always done in a timely
 manner.

The copyright dates in the source code are not helpful
 either, for a different reason. Early in 1985 (before the
 release of 4.3BSD (1986)), the developers added copyright
 notices for about half the source files, dating those to
 &ldquo;1980&rdquo; which is misleading. Many of those files
 were first published in 4.2BSD (1983), and did not exist in
 1980.
 </li>
 </ul>

<h3 id="ref_misleading-id"><a name="ref_misleading" id=
 "ref_misleading">Interesting but misleading:</a></h3>

<ul>
 <li id="misinfo-termhow2">
 <a href=
 "http://www.tldp.org/HOWTO/Text-Terminal-HOWTO.html">Text-Terminal-HOWTO</a>.

I find the deliberate misstatements in this one to be
 fascinating. Still, there is some useful information as
 well.
 </li>

<li id="misinfo-haskell">
 <a href=
 "http://hackage.haskell.org/cgi-bin/hackage-scripts/package/nanocurses">
 nanocurses (Haskell binding)</a> 
 <a href=
 "https://github.com/skogsbaer/hscurses/blob/f63d98aae747ef2aef41b4a2c35be451b14c7878/UI/HSCurses/Curses.hsc#L31">
 It says</a> (referring to the ncurses manpages):

<blockquote>
 Sections of the quoted documentation are from the
 OpenBSD man pages, which are distributed under a BSD
 license.
 </blockquote>

First noticed in September 2008, as of July 2019, this
 misattribution is still present, e.g., in the Debian <a href=
 "https://packages.debian.org/search?keywords=libghc-hscurses-dev">
 libghc-hscurses-dev</a> package (according to &ldquo;git
 blame&rdquo; the upstream comment was added in 2005).
 </li>

<li id="misinfo-openbsd">
 Aside from a few grammatical and spelling fixes, OpenBSD
 developers have not contributed to ncurses documentation.
 That does not mean that the ncurses documentation in OpenBSD
 is identical. OpenBSD has made some questionable changes,
 e.g., the <a href=
 "https://man.openbsd.org/reset#HISTORY">history</a> and
 <a href="https://man.openbsd.org/reset#AUTHORS">authors</a>
 sections in the <a href=
 "https://man.openbsd.org/reset">tset</a> manual page (compare
 with <a href="man/tset.1.html#h2-HISTORY">ncurses</a>).
 </li>

<li id="misinfo-interix">
 Even when the ncurses documentation is used without
 modification, it can still be misleading. For example,
 Interix (a &ldquo;client system&rdquo; or
 &ldquo;environment&rdquo; (like OS/2 EMX, U/Win, Cygwin,
 MSYS2) came with a port of ncurses.

In the first announced version (e.g., 3.0), that appeared
 to be <a href="ncurses-license.html#patch_961130">ncurses
 1.9.9g</a> based on the header files. However its
 documentation was a compressed help file &ldquo;.chm&rdquo;
 which used an earlier set of documentation from early 1996,
 e.g., <a href="ncurses-license.html#ncurses_1_9_9e">ncurses
 1.9.9e</a>. The disparity was evident due to the way
 resizeterm and wresize were <a href=
 "ncurses-license.html#ncurses_1_9_9">documented</a>.

The final version of Interix 3.5 (aka "SFU" or "SUA") with
 updates to 2008 came with ncurses 4.2 (19980228) (apparently
 completed by <a href=
 "https://groups.google.com/forum/#!topic/comp.lang.tcl/Uc4oo1Kco68">
 Rodney Ruddock</a> in September 1999). The documentation was
 not updated to match the software. Besides the <a href=
 "man/resizeterm.3x.html">resizeterm</a> issue, the TechNet
 page <a href=
 "https://technet.microsoft.com/en-us/library/bb463194.aspx">Curses
 Applications on Interix</a> for 3.0 mentions a problem
 which was fixed in <a href="NEWS.html#t960615">June
 1996</a>:

<blockquote>
 function calls to getbkgd()
 needs to be replaced by wgetbkgd()
 </blockquote>
 </li>

<li id="dead-bugreport">
 But Interix is dead, last seen in Windows 7 Ultimate
 (January 14, 2020). This <a href=
 "https://jdebp.eu/FGA/interix-terminal-type.html">page</a> is
 an example of a dead bug report. It was <a href=
 "NEWS.html#t20170729">addressed</a> in ncurses in 2017.
 </li>

<li id="misinfo-csharp">
 A C# binding for curses, which supports some of ncurses
 and PDCurses extensions, first noted in mid-2009 <a rel=
 "nofollow" href=
 "https://en.wikipedia.org/w/index.php?title=Ncurses&amp;diff=prev&amp;oldid=297445477">
 here</a> (apparently added anonymously by the program's
 author). Its <a href=
 "http://curses-sharp.sourceforge.net">page</a> claimed that
 "Virtually every function in curses has its managed
 counterpart." I noted that it implemented less than half of
 the functions in ncurses 5.7 (206/414). As of early 2013, it
 is (like most SourceForge projects), apparently dead.
 </li>

<li id="misinfo-ansiseq">
 The treatment of escape sequences <a rel="nofollow" href=
 "http://en.wikipedia.org/wiki/ANSI_escape_code">here</a> is a
 mish-mash of incorrect information from the DOS/Windows and
 Unix environments. There are several areas of confusion,
 e.g., mistaking common practice for standardization (and
 introducing confusion about the scope of the standards), as
 well as over-generalization. The worst offenders are of
 course <a href=
 "/personal/not-anonymous.html">anonymous</a>.
 </li>

<li id="misinfo-python">
 The Python binding and tutorial bears some comment. On
 reflection, the comment is involved (see <a href=
 "/ncurses/ncurses-python.html">this page</a>).
 </li>

<li id="misinfo-colorman">
 The accepted answer for <a href=
 "http://unix.stackexchange.com/questions/119/colors-in-man-pages">
 Colors in Man Pages</a> has some issues:

<ul>
 <li>It goes to some effort to state that termcap is better,
 and that <code>less</code> uses termcap, but
 explicitly uses the terminfo-based <a href=
 "/ncurses/man/tput.1.html">tput</a> utility.</li>

<li>It introduces some capabilities which are not supported
 by the <a href=
 "http://www.greenwoodsoftware.com/less/"><code>less</code></a>
 pager, i.e., for superscripts and subscripts.</li>
 </ul>
 </li>

<li id="misinfo-reset">
 The accepted answer for <a href=
 "http://unix.stackexchange.com/questions/79684/fix-terminal-after-displaying-a-binary-file/79686#79686">
 Fix terminal after displaying a binary file</a> has
 some issues:

<ul>
 <li>
 At the outset, a figure is shown which is actually
 irrelevant to the question. It shows the effect of
 sending a binary file to the terminal, but all of the odd
 characters are the Unicode <a href=
 "http://unicode.org/charts/PDF/UFFF0.pdf">replacement
 character</a>. That is

<blockquote>
 used to replace an incoming
 character whose value is unknown or unrepresentable in
 Unicode
 </blockquote>

One of the <a href=
 "https://www.cl.cam.ac.uk/~mgk25/ucs/man-utf-8.html">claimed
 advantages</a> of UTF-8 is that it is stateless.
 In this situation, there is no action needed by the user
 to get the terminal to stop printing replacement
 characters, since the next output character is checked
 without regard to the previous character.

If the terminal were not in UTF-8 mode, it would not
 print replacement characters, and users expect
 it to show line-drawing characters, e.g.,. as in <a href=
 "http://unix.stackexchange.com/questions/50752/how-to-fix-a-terminal-after-a-binary-file-has-been-dumped-inside-it">
 How to fix a terminal after a binary file has been
 dumped inside it?</a>, or (older) <a href=
 "http://tldp.org/HOWTO/Keyboard-and-Console-HOWTO-4.html">
 4. Resetting your terminal</a> in <a href=
 "http://tldp.org/HOWTO/Keyboard-and-Console-HOWTO.html">
 The Linux keyboard and console HOWTO</a>. But bear
 in mind that the HOWTO (2002) is fairly recent: the
 <a href=
 "/ncurses/man/tset.1.html#h2-HISTORY"><code>reset</code></a>
 program dates from around 1980, and has been an FAQ ever
 since.
 </li>

<li>
 After the interesting picture, two methods are
 suggested for fixing the terminal. For the first, the
 advice is to

<blockquote>
 1. Hit Control + C a couple of
 times (Ctrl+C) 
 2. Type the command reset and hit return
 </blockquote>

That is partly helpful. No reason is given for using
 control/C, but suppose the binary file had sent a control
 to the computer asking it to do something (without
 providing a newline to complete the command). The
 <code>reset</code> command does two (types of) thing:

<ol>
 <li>it resets the terminal modes (in the
 computer's terminal driver)</li>

<li>it resets the terminal itself (using escape
 sequences).</li>
 </ol>

Actually, the first part should not be needed, since
 the escape sequences possibly echoed by the binary file
 will not change the terminal modes. But typing
 <code>reset</code> is easy.
 </li>

<li>
 The other method says in effect to do each of the
 steps that <code>reset</code> would do, as separate
 commands. No explanation is given (nor would it help).
 The reason why one would use <code>stty
 sane</code> is to cleanup after a program which modified
 the terminal modes, e.g., from a text editor.
 That would be a different question.

However, running <code>stty sane</code> may not
 restore your terminal to a state that you expect. It uses
 settings built into the <code>stty</code> program which
 may have no relation to the settings established when you
 logged in. For instance:

<ul>
 <li>On some platforms (Solaris, Linux), it changes the
 <code>erase</code> character to ASCII DEL (127) &mdash;
 inconsistent with normal usage for the first, but
 consistent with the second.</li>

<li>But on HPUX, it changes <code>erase</code> to
 <code>#</code> (and it changes the
 <code>intr</code> character from control/C to ASCII
 DEL, and kills the <code>kill</code> character).</li>

<li>On Unix platforms, the same command also changes
 the parity handling, and turns off hardware tabs.</li>
 </ul>
 </li>

<li>
 Despite all of that, there was some value to the
 answer: it prompted a review of the documentation for
 <a href=
 "/ncurses/man/tset.1.html#h2-HISTORY"><code>reset</code></a>
 and <a href=
 "/ncurses/man/tput.1.html#h2-HISTORY"><code>tput</code></a>
 (and improvements to the programs, so that
 &ldquo;<code>tput reset</code>&rdquo; would have the same
 effect as &ldquo;<code>reset</code>&rdquo;).
 </li>
 </ul>
 </li>

<li id="misinfo-linfo">
 Noticed in addressing <a href=
 "/ncurses/ncurses-license.html#issues_expat">some issues</a>
 regarding the ncurses license, <a href=
 "http://www.linfo.org">this site</a> is (like <a href=
 "http://www.osscc.net/en/">another</a>) apparently an
 anonymous work by an individual posing as a group of people
 for the sole purpose of disseminating misinformation.
 </li>

<li id="misinfo-man7">
 On a lighter note (partly), while looking for a suitable
 webpage for online manual pages, I noticed <a href=
 "http://man7.org/linux/man-pages/man3/ncurses.3x.html">this
 page</a>. Unlike <a href=
 "http://linux.die.net/man/3/ncurses">another site</a>, its
 formatting is reasonable, and is reasonably complete.
 However:

<ul>
 <li>its ncurses-related manual pages are generated from a
 repository based on my development patches for ncurses,
 but</li>

<li>it does not process the sources to fill in
 configuration details (such as the actual name of the
 programs, the version information), and</li>

<li>it lacks most of the hyperlinks which are created by
 <a href="/scripts/man2html.html">man2html</a> (compare with
 <a href="man/ncurses.3x.html">my site</a>), and</li>

<li>the boilerplate at the bottom probably should point to
 this <a href="/ncurses/">ncurses page</a>, rather than
 merely to the release announcement. I put a lot of
 information and links into the release announcements, but
 "home page" it is not. At the time of release for ncurses
 6.1, according to FSF's <a href=
 "https://web.archive.org/web/20180129060116/https://directory.fsf.org/wiki/GNU">
 Free Software Directory</a>, my site was the homepage
 (dickey.his.com, which in turn forwards to
 invisible-island.net).</li>
 </ul>
 </li>

<li id="misinfo-homepage">
 Speaking of "home page", Eric Raymond at one point had
 what he termed an <a href=
 "https://web.archive.org/web/19970605130234/http://www.ccil.org/~esr/ncurses.html">
 "ncurses resource page"</a>. After the <a href=
 "ncurses-license.html">licensing issues</a> were resolved, he
 modified his terminfo page to state that "there is no
 resource page for ncurses", and a few years later, again
 modified his <a href=
 "https://web.archive.org/web/20010620080924/http://hurkle.thyrsus.com/~esr/terminfo/index.html">
 terminfo page</a> to point to the release announcement as the
 "ncurses home page".
 </li>
 </ul>

<h2 id="copyright_text-id"><a name="copyright_text" id=
 "copyright_text">Copyright</a></h2>

Permission to use, copy, modify, and distribute this software
 and its documentation for any purpose and without fee is hereby
 granted, provided that the above copyright notice appear in all
 copies and that both that copyright notice and this permission
 notice appear in supporting documentation, and that the name of
 the above listed copyright holder(s) not be used in advertising
 or publicity pertaining to distribution of the software without
 specific, written prior permission.

THE ABOVE LISTED COPYRIGHT HOLDER(S) DISCLAIM ALL WARRANTIES
 WITH REGARD TO THIS SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES OF
 MERCHANTABILITY AND FITNESS, IN NO EVENT SHALL THE ABOVE LISTED
 COPYRIGHT HOLDER(S) BE LIABLE FOR ANY SPECIAL, INDIRECT OR
 CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
 LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT,
 NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
 CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
</body>
</html>

cmd: