Copyright © 1997-2021,2022 by Thomas E. Dickey

Here is the latest version of this file.

XTerm – Frequently Asked Questions (FAQ)

What is XTerm?

From the manual page:

The xterm program is a terminal emulator for the X Window System. It provides DEC VT102/VT220 and selected features from higher-level terminals such as VT320/VT420/VT520 (VTxxx). It also provides Tektronix 4014 emulation for programs that cannot use the window system directly. If the underlying operating system supports terminal resizing capabilities (for example, the SIGWINCH signal in systems derived from 4.3bsd), xterm will use the facilities to notify programs running in the window whenever it is resized.

That is, xterm (pronounced "eks-term") is a specific program, not a generic item. It is the standard X terminal emulator program.

This FAQ presents various useful bits of information for both the specific program as well as other programs that imitate it.

As a stylistic convention, the capitalized form is "XTerm", which corresponds to the X resource class name. Similarly, uxterm becomes "UXTerm".

Who wrote XTerm?

I've been working on xterm since early 1996 (see my changelog for details).

But the program is much older than that:

A Prehistory Perspective

A lot of people, cited at the bottom of the manual page wrote the original xterm program, maintained by the X Consortium (later part of The Open Group – I'm well aware of the distinction, but am citing when the work was done, not who the current owner may be). There is no changelog, and it is not clear who did what. Email from Jim Gettys (September 1998) provides some background:

Cast of thousands...

To give a bit of history, xterm predates X!

It was originally written as a stand-alone terminal emulator for the VS100 by Mark Vandevoorde, as my coop student the summer that X started.

Part way through the summer, it became clear that X was more useful than trying to do a stand alone program, so I had him retarget it to X. Part of why xterm's internals are so horrifying is that it was originally intended that a single process be able to drive multiple VS100 displays. Don't hold this against Mark; it isn't his fault.

I then did a lot of hacking on it, and merged several improved versions from others back in.

Notable improvements include the proper ANSI parser, that Bob McNamara did.

The Tek 4010 support came from a guy at Smithsonian Astrophysical Observatory whose name slips my mind at the moment.

Ported to X11 by Loretta Guarino.

Then hacked on at the X Consortium by uncounted people.

Email from Doug Mink (October 1999) provides more background:

I was checking out the newly revised AltaVista search engine to see what was on the net about xterm, and I found your pages. I can add to the FAQ in that I was the "guy at the Smithsonian Astrophysical Observatory" Jim Gettys refers to. I am listed at the end of the man page under authors. What happened was that I was hired by SAO (after leaving the research staff at MIT) in October 1985 to write analysis software for the Spacelab 2 Infrared Telescope which was to fly on the Space Shuttle in 1985 less than six months after I was hired. I came with a tar tape full of software I had written for Unix and Tektronix terminals, but I was presented with a VS100 terminal which had an early version (X6 or so) of xterm, with no graphics capabilities. SAO is at Harvard, across Cambridge from MIT, where Jim Gettys was detailed from DEC to the X project, and Jim had connections with SAO, having worked here after college (MIT, where we had both worked at the observatory at various times); he was still sharing an apartment with an SAO colleague of mine, too. Anyway, everyone decided that since I knew Tektronix commands pretty well, and our group desparately needed the graphics capabilities, it would be a good use of my time to implement a Tektronix terminal emulator under X. So I set to work learning more C--I had only written a couple of wrappers to C I/O routines so I could use them with my Fortran software--and wrote a Tektronix emulator. The only X documentation at the time was the code itself. While I was at it, I wrote an improved Tektronix emulator for our Imagen laser printer which used the full resolution of that 300 dpi printer instead of the effective 100 dpi (i.e. jaggy) emultator distributed with the printer. The original xterm Tek emulator shared a window with the VT100 emulator, much like on the VT240 terminals which I had been using at MIT before I came to Harvard. With a VAX 750 running several VS100's, window creation was sloowww, so sharing a window was the quickest way to do things, and all of my software was written for that mode of operation, anyway. While I wrote the emulator so that my software would work on it, it was tested by the X group against a BBN graphics package, the name of which slips my mind right now.

Anyway, 15 years later, I am still using xterm and some of the same mapping software I wrote the emulator for. And I am still at the Smithsonian Astrophysical Observatory.

Looking for more perspective

VS100 refers to the VAXstation 100, introduced by DEC in 1984 (see manual).

There is a git repository here which gives some more of xterm's prehistory. But it has no usable data for X10 (see initial revisions in 1985/1986). The X developers did not really start checking in their code until 1987.

The earliest version of xterm which is available is from X10R3. According to the xc-historical repository, that was February 1, 1986 (a change to the rgb.c file). Even that date is suspect because the X10R3 tarball contains xterm files later than that date:

  17024 May 17 1986    xterm/util.c
  31545 May 17 1986    xterm/main.c
   1377 Feb 10 1986    xterm/Makefile
  10804 Feb  3 1986    man/xterm.1
    660 Feb  2 1986    xterm/terminfo
    606 Feb  2 1986    xterm/termcap
   1178 Feb  2 1986    xterm/tabs.c
  10525 Feb  2 1986    xterm/screen.c
   2907 Feb  2 1986    xterm/resize.c
   6699 Feb  2 1986    xterm/ptyx.h
   2720 Feb  2 1986    xterm/input.c
   2959 Feb  2 1986    xterm/esctable.h
   4482 Feb  2 1986    xterm/cursor.c
   7661 Feb  2 1986    xterm/chartable.h
  23543 Feb  2 1986    xterm/charproc.c
  13990 Feb  2 1986    xterm/button.c
   1122 Feb  2 1986    xterm/buf.c
   3932 Feb  2 1986    xterm/ansi.c
  15662 Feb  2 1986    xterm/Tplot.c
   2382 Feb  2 1986    xterm/README
    234 Feb  2 1986    cursors/xterm_mask.cursor
    234 Feb  2 1986    include/X/cursors/xterm_mask.cursor
    261 Feb  2 1986    cursors/xterm.cursor
    261 Feb  2 1986    include/X/cursors/xterm.cursor
   1335 Dec 21 1985    xterm/icon.ic
   1350 Dec 21 1985    xterm/icon_mask.ic

But since the X10R3 tarball contains only 10 “later” files out of 974, the 3-4 xterm files are just some after-the-fact correction whose reason is long lost.

A posting by Andreas Motl to nfo.links.computing in March 2003 recapitulates the content of a page Linux desktops (GUIs, widgets, window managers, etc) (Jorn Barger, November 2002). That mentions X10 (e.g., “X v10”):

1985: Sep: X v9 distributed as free [cite] cf Andrew [Gettys]
1985: X/Open Portability Guide [cite]
1985: Oct: Amiga 1.0 [info]
1985: 20Nov: Windows 1.0 ships [tour]
1985: Dec: Clayton Elwell's Portable User Interface Library project [GooJa]

1985: Dec: X v10r2 [annc]

  "X is a portable network transparent window system for bitmap displays.
Applications can be run from any machine in a local area network via stream
connections (currently TCP and Unix domain are supported). X is a fully
hierarchic window system intended to be very flexible. Manipulation of existing
windows is done with an external window manager; if you don't like ours, write
your own. Applications include a terminal emulator (~VT102 and Tek 4010), load
monitor, clock, imagen previewer, and several window managers among other

no-date: Windows 2.0 clones Mac interface; withdrawn and tweaked to avoid
lawsuit? [cite]

1986: Jan: DEC announces VAXstation with X
1986: Feb: X v10r3 [annc]

September 1985 might be accurate for X version 9. However, the first line of the xterm manual page from X10R3 says

.TH XTERM 1 "1 January 1985" "X Version 10"

Jim Gettys might be able to explain that. Several of the files in that version of xterm have copyright dates from 1984, making that the likely beginning of development for xterm. Other parts of the X10R3 source tree have similar clues. For example, Xlib may have begun development in 1984, but since it was unreleased until the end of 1985, the actual date is indeterminate.

A comment by Jim Fulton on slashdot in 2000 gives more details on the early versions of X:

There weren't really ten full releases prior to X11R1, however there were 10 incompatible revs of the protocol. Most of the early versions were primarily used within MIT (Athena and LCS) and friendly commercial R&D labs. Here's some of the pre-history based on cryptic notes and blurry memory: X1 - summer 1984 - the first version, based on a substantial rearchitecting of the UNIX port of the W Window System (originally developed for the V Kernel). X3 - fall 1984 - used internally at MIT as the initial basis of various plotting packages for coursework. X6 - spring 1985 - first version licensed by MIT to various companies (including Cognition, MASSCOMP, and Digital) for use in commercial products. It cost $100 and if you wanted you could stop off at the (very small) licensing office to pick up your own magtape. X8/X9 - fall 1985 - added color (X8 lasted all of about a week; X9 was quickly released to fix a protocol alignment problem that impacted ports on the IBM PC/RT). Many organizations began developing ports (including a version to the Lexidata 9000 display card for VAXen that was used at the Autofact tradeshow in late 1985 to show a prototype of the first 3rd party application: a mechanical engineering design system). X.V10R1 - spring 1986 - first version released by MIT that did not require signing a license agreement. Also the first version to have a DOS Xserver developed. X.V10R[234] - fall 1986 & spring 1987 - an explosion of ports done on a variety of platforms. X.V11R1 - Sep 15, 1987 - major overall done in collaboration with folks from Digital, Sun, IBM, and other companies. Formed the basis of core protocol used today. Companies and organizations releasing X-based products used this release as a starting port for incorporating into their own distributions. X.V11R2 - March 1, 1988 - first version released under the auspices of the newly-formed MIT X Consortium. The MIT X Consortium continued to put out releases of X11 for a number of years. Then in the mid-90s, it was spun off into a separate not-for-profit organization (simple the X Consortium). As has been noted, that eventually folded into various organizations that became X.ORG. The rest is history. :) Jim Fulton

My Involvement

My involvement with xterm through XFree86 began at the end of 1995. This website has been "here" since 2001/6/5, replacing my ClarkNet page. I started the ClarkNet page 1996/12/31, as a followup to the release of ncurses 4.0) which featured xterm as one of the 16 programs I was involved with. From the outset, the page provided a link to a snapshot of the current source. Copies of patches which I sent to XFree86 were available on the ftp area.

XFree86 had its sources in CVS, but (like others in that era), were not directly visible to random developers. That came later. I started by downloading the sources (30Mb of compressed tar-files on a 56Kb phone connection took about 6 hours) and updating them with patches from the XFree86 mailing list.

Like the other programs that I worked on with others (vile, tin, lynx), I set up an RCS archive to track my changes locally before sending patches to the development list. As the XFree86 developers issued new patches, I would re-synchronize my archive. Later, XFree86 provided CVS (initially readonly). I was granted commit privileges on this in November 2000, and stopped mailing patches after #149.

Throughout this period, my work on xterm was released as part of XFree86. It was rare for a separate package to be provided. That was due to the potential conflict between the install procedures. Users of the downloads from my web/ftp site were predominantly individual developers.

There were exceptions. Christian Weisgerber proposed a package for FreeBSD ports later in 1999 (ports/15545: new port: x11/xterm, followup in March 2000). However, that was an exception. None of the Linux distributions provided a separate package before 2003 (when Mike Harris created a package of patch #177 for Red Hat). Again that is more of an exception than a rule:

Given that context (sources distributed via XFree86 CVS, releases via XFree86), the statement made by an Xorg hacker early in 2005 asserting that “It has not been maintained by anyone within the XFree86 or trees for many years” was at best misleading.

After the “fork” (sic) of Xorg in 2004, I continued to commit changes for xterm in XFree86 CVS until patch #216 in mid-2006. I stopped at that point because it was not possible to incorporate changes into xterm which were not sent to me first. I still send patch announcements to both the XFree86 and Xorg mailing lists, of course.

At that point (mid-2006), the XFree86 CVS was no longer the primary development repository for xterm. My RCS archive filled that need. Later (starting in 2016), I provided Git snapshots using the RCS labels which I make in development:

Focus of this FAQ

This FAQ is oriented toward the version of xterm originally distributed with XFree86 (more commonly known as modern, or "new xterm", with a corresponding terminal description "xterm-new"), which was based on the X11R6.3 xterm, with the addition of ANSI color and VT220 controls.

What is a VT220?

Why a VT220?

The manual page mentions a VT220. Most terminal emulators documentation talk about VT100. But a VT100 is a rather limited subset of what people expect:

Initially, I was only interested in making colors workable for curses programs.

Later, I noticed that xterm had some support for what would now be termed as ISO-2022. That was a VT220 feature which preceded ISO-2022 called National Replacement Character sets. In any case, it was not a VT100 feature. There were some missing pieces. So I decided to fill in those pieces and make xterm a VT220 emulator. (VT220s do not do ANSI color either—the missing pieces were in other areas).

XTerm also provides features that are in neither VT100 nor VT220, which are used by other programs as "xterm emulation".

By the way, the control string used for setting the titles was not in a standard format:

ECMA-48 (the standard) does not describe this particular control, but prescribes its format (an operating system command). It does not use a BEL.

I revised that area starting in 1996,

In addition to implementing the VT220's National Replacement Character sets (see vttest screenshots), I added other features to emulate the successive models of DEC terminals. The decTerminalID resource (in 1996) lets users select the emulation to use. Because many of my changes were extensions (features not in any of DEC's terminals) and because well-behaved VT100 applications would not use features from higher-level terminals it was not initially important to prevent use of those by applications which assumed they were using just a VT100. Knowledgable users could easily configure xterm to emulate a VT220. In 2012, I changed the default from VT100 to VT420.

What is a State Table?

That was mentioned regarding the title strings. XTerm uses a state machine to handle incoming characters. That is essentially what a real terminal does. Other "xterm" terminal emulators typically do not do this, which makes them not do well with vttest.

Why not emulate VT320?

You could do that (by changing decTerminalID, but the results were not that interesting). In retrospect, the VT320 was a stopgap implementation designed to bridge between the VT200 series and the VT420. It provided a standard codepage (for ISO Latin-1).

While it had other features not found in the VT200-series, most of those are less useful in a terminal emulator. I did adapt the ECMA-48 scrolling operations which the VT320 interpreted as panning the visible display in the terminal's memory. Expect some difference there (if you can find an application on VMS which used the feature).

The VT320 was popular with developers of commercial terminal emulators, whose literature referred to it as supporting ANSI color. It did not do this.

Why emulate VT420?

The VT420 was interesting because it provided two features that could be useful:

A VT420, of course, supports all of the features in VT320, in turn all of the features in VT220, and in turn VT100. Users would not lose features by changing the default emulation to VT420. By changing the default emulation, most users would automatically be able to use applications (such as tmux) that could perform better if the left/right margin feature is available. I changed the emulation to VT420 in 2012 for this reason.

XTerm does not emulate some esoteric features (such as dual sessions) because those require hosts using special software, and no publicly-available documentation was available.

Why not emulate VT520?

Again, the VT500-series is less interesting because most of the features which are not hardware-specific (such as reporting transmission rate) are less useful.


As for the other features, most are not useful in emulation (since they are hardware-specific). Additionally, these less-used features are not documented precisely and since the only point of providing them would be for successful interoperability with legacy applications, some reverse-engineering would be needed to provide a faithful emulation. To date there are no known terminal emulators which do that.

What platforms does it run on?

XTerm runs in all of the implementations of X11. As of 2000, I had built and run these since I started working on xterm in 1996:

The older configurations have X11R5 libraries. Only minor changes are needed to make xterm work on those systems. However, X11R6 provided better locale support, as well as new features such as the active icon. X11R7... not much to say there.

Since 2000, there have been many changes (including new platforms such as MacOS, NetBSD, OpenBSD, etc., as well as QNX, Cygwin, and Minix).

What is the latest version?

The most recent (and well supported) version of xterm is the one that I maintain:

What versions are available?

There are several other versions of xterm, based on xterm's source. These include

There are similar programs not based on xterm's source, which are compatible to different degrees. These include

Some of these use the VTE widget. Since that supplies most of the terminal emulation, the remaining differences between programs using VTE tend to be at the level of the window manager (menus, borders, etc.). Other (older) programs which are based on reusable widgets include dtterm and emu.

(I am aware of a few others, such as xcterm, but have not seen a working version of these).

Finally of course, there are a multitude of programs which set TERM to "xterm", in the hope that applications will treat them the same as xterm. For example,

Each of the programs noted here which are well-established and which are known to differ markedly from xterm have their own terminal descriptions in ncurses, to which TERM should be set. Otherwise, bug-reports are misdirected to ncurses which should have been addressed by the respective developers of these programs. These include

Comparing versions, by counting controls

Several of these programs are claimed (either by their developers, or their users) to emulate "most" of xterm. To me, "most" would be something quantifiable, e.g., 80 percent. To satisfy my curiousity, I wrote a script to extract the control sequence information from ctlseqs.txt. This counts each control sequence, as well as the variations such as setting bold, color, inverse video. Then I (laboriously) inspected these terminal implementations:

As of mid-November 2010, these were the latest implementations. I included data for the vt220 and vt102 to be able to contrast the various terminal emulators against those as well as xterm. There were:

For each control, there are three possibilities:

  1. "yes" — the terminal implements it, matching xterm. If xterm implements it, and it is a feature of vt220 or vt102, then in turn xterm's behavior must match vt220 or vt102.
  2. "partial" — the terminal implements it, but its behavior does not match the reference noted above.
  3. "no" — the terminal does not implement the control.

The control sequences document lists a few controls which xterm does not (completely) implement, e.g.,

Those are not implemented in xterm because all X applications share the same keyboard (see manpage), and xset is the proper tool for changing auto-repeat. The X10R4 xterm did implement key-repeat, but the feature was removed in X11R1. None of the other terminal emulators implements those either.

Comparing against the control sequences document
yes partial no program
488 4 6 xterm-new
154 6 338 xterm-r6
188 5 305 vt220
104 0 394 vt102
204 3 291 rxvt
219 3 276 urxvt
191 2 305 putty
170 3 325 konsole
184 6 308 vte

Modern xterm implements 188 primary controls. In this table, vte ranks lower than PuTTY because it does not support vt52 emulation. This is not unusual, since the rxvt-based emulators do not, either. However, all vt100's provide this feature; programs lacking this are not really a vt100 emulator. On the other hand, PuTTY (which is not a vt100 emulator due to its incompatible wrapping behavior) supports this feature.

Aside from that, the various emulators implement much the same features from xterm. None implements as many as half of xterm's controls.

Comparing against xterm
yes partial no program
488 0 0 xterm-new
154 6 328 xterm-r6
182 2 304 vt220
98 0 390 vt102
204 3 281 rxvt
219 3 266 urxvt
189 2 297 putty
170 3 315 konsole
184 6 298 vte

DEC VT220 implements 96 primary controls. Modern xterm (as documented), implements most of the VT220. VTE implements fewer than half. The others are a little better. None of the others could be used as a real VT220.

Comparing against vt220
yes partial no program
182 0 6 xterm-new
78 6 104 xterm-r6
188 0 0 vt220
104 0 84 vt102
101 3 84 rxvt
106 3 79 urxvt
107 2 79 putty
100 3 85 konsole
88 6 94 vte

DEC VT102 (the actual flavor used for "vt100" in most cases), implements 68 primary controls. Again, VTE fares worst, and the others a little better.

Comparing against vt102
yes partial no program
98 0 6 xterm-new
70 6 28 xterm-r6
104 0 0 vt220
104 0 0 vt102
79 2 23 rxvt
81 2 21 urxvt
86 2 16 putty
85 3 16 konsole
60 1 43 vte

I have continued to add features to xterm:

The other programs change far more slowly. As a result they implement a smaller fraction of xterm's repertoire in 2019 than in 2010. Even for cases where they implement a function, it may not work properly (see for example the screenshot of VTE in the vttest NRCS examples).

In summary, none of the other terminal emulators emulates "most" of xterm. Instead, they implement the most commonly-used control sequences, and there are differences between them.

How do I ...

Not really problems, but frequently asked questions (the point of this, after all):

How do I change the font size?

XTerm uses fonts given as resource settings. You can switch between these fonts at runtime, using a menu. This is documented in the manpage, in the MENUS section.

X Consortium xterm provides popup menus, by pressing the control key together with the mouse button. Control right mouse button pops up the VT FONTS menu, from which you can select fonts that are specified in xterm's resources. Usually these are in increasing order of size.

Modern xterm provides the menu, plus a feature adapted from rxvt: pressing the shifted keypad plus or minus keys steps through the font menu selections, in order of their size.

XTerm's manpage does not document the syntax for X resources; it is done in the X documentation. If you are instead asking about a problem displaying a given font, it may be due to a problem with your resource settings.

How do I print the screen?

That depends on why you want to print it.

If you want a trace of an interactive session, you should use the script program. It records every character sent to the screen, recording them in a file typescript. There are two drawbacks to this approach:

Well, what about logging? Some versions of xterm support logging to a file. In fact modern xterm does. Logging was dropped from X Consortium xterm during X11R5 due to security concerns. Those were addressed, but logging was not reinstated (in fact there is a related bug in xterm). Some people prefer this, because it is convenient: you can start and stop logging a popup menu entry. However

Both script and logging are useful for recording, but they require interpretation to make sense of the trace. You probably would not send that trace to a printer (not twice, anyway).

If you want to print the contents of the screen, modern xterm implements, as part of the VT100 emulation, an "attached" printer.

There are limitations and tradeoffs using the "attached" printer, because it is an emulation:

If you use the popup menu to print the screen, this will close the printer pipe unless it was already opened by the application running in xterm.

How do I set up function keys?

With modern xterm, this is relatively simple. So I'll answer that first.

With X Consortium xterm, you had partial support for DEC VTxxx function keys. Function keys F1 to F12 correspond to DEC's F1 to F12 (sort of). Actually, DEC's VT220 terminals do not have codes for F1 through F5. They are reserved for local functions. And the VT220 (and up) terminals have 20 function keys. So you cannot do anything with the F13 through F20 (i.e., DO, HELP and SELECT). Finally, though xterm is reputed to be VT100-compatible, it has no support for the VT100 keypad (PF1 to PF4, and the "," key).

Modern (XFree86) xterm changed the X Consortium codes for F1 to F4 to match the VT100 PF1 to PF4, except when the emulation level is VT220 and up. In this case, it generates the same F1 to F4 codes as X Consortium xterm. Moreover, it adds a new resource sunKeyboard, which tells the program whether it has only 12 function keys (i.e., a Sun or PC keyboard). If so (this is selectable from the popup menu), you can use the control key with F1 to F12 to get F13 to F24, and use the "+" key on the keypad as an alias for "," (comma).

The emulation level for modern xterm is set via the resource decTerminalID, e.g., to 220 for a VT220. Once set, applications can set the emulation level up or down within that limit. DEC's terminals are configured in much the same way by a setup option.

That is the simple way, using a couple of new resources. The traditional way to get function keys involves translations. I have seen a few postings on the newsgroups that do this. Here is one from Bruce Momjian <> for a VT220:

xterm $XTERMFLAGS +rw +sb +ls $@ -tm 'erase ^? intr ^c' \
        -name vt220 -title vt220 -tn xterm-220 "$@" &

with the corresponding resources:

XTerm*VT100.translations: #override \n\
<Key>Home: string(0x1b) string("[3~") \n \
<Key>End: string(0x1b) string("[4~") \n
vt220*VT100.translations: #override \n\
<Key>F1: string(0x1b) string("OP") \n \
<Key>F2: string(0x1b) string("OQ") \n \
<Key>F3: string(0x1b) string("OR") \n \
<Key>F4: string(0x1b) string("OS") \n \
<Key>F5: string(0x1b) string("[16~") \n \
<Key>F6: string(0x1b) string("[17~") \n \
<Key>F7: string(0x1b) string("[18~") \n \
<Key>F8: string(0x1b) string("[19~") \n \
<Key>F9: string(0x1b) string("[20~") \n \
<Key>F10: string(0x1b) string("[21~") \n \
<Key>F11: string(0x1b) string("[28~") \n \
<Key>F12: string(0x1b) string("[29~") \n \
<Key>F1: string(0x1b) string("[23~") \n \
<Key>F2: string(0x1b) string("[24~") \n \
<Key>F3: string(0x1b) string("[25~") \n \
<Key>F4: string(0x1b) string("[26~") \n \
<Key>F5: string(0x1b) string("[K~") \n \
<Key>F6: string(0x1b) string("[31~") \n \
<Key>F7: string(0x1b) string("[31~") \n \
<Key>F8: string(0x1b) string("[32~") \n \
<Key>F9: string(0x1b) string("[33~") \n \
<Key>F10: string(0x1b) string("[34~") \n \
<Key>F11: string(0x1b) string("[28~") \n \
<Key>F12: string(0x1b) string("[29~") \n \
<Key>Print: string(0x1b) string("[32~") \n\
<Key>Cancel: string(0x1b) string("[33~") \n\
<Key>Pause: string(0x1b) string("[34~") \n\
<Key>Insert: string(0x1b) string("[2~") \n\
<Key>Delete: string(0x1b) string("[3~") \n\
<Key>Home: string(0x1b) string("[1~") \n\
<Key>End: string(0x1b) string("[4~") \n\
<Key>Prior: string(0x1b) string("[5~") \n\
<Key>Next: string(0x1b) string("[6~") \n\
<Key>BackSpace: string(0x7f) \n\
<Key>Num_Lock: string(0x1b) string("OP") \n\
<Key>KP_Divide: string(0x1b) string("Ol") \n\
<Key>KP_Multiply: string(0x1b) string("Om") \n\
<Key>KP_Subtract: string(0x1b) string("OS") \n\
<Key>KP_Add: string(0x1b) string("OM") \n\
<Key>KP_Enter: string(0x1b) string("OM") \n\
<Key>KP_Decimal: string(0x1b) string("On") \n\
<Key>KP_0: string(0x1b) string("Op") \n\
<Key>KP_1: string(0x1b) string("Oq") \n\
<Key>KP_2: string(0x1b) string("Or") \n\
<Key>KP_3: string(0x1b) string("Os") \n\
<Key>KP_4: string(0x1b) string("Ot") \n\
<Key>KP_5: string(0x1b) string("Ou") \n\
<Key>KP_6: string(0x1b) string("Ov") \n\
<Key>KP_7: string(0x1b) string("Ow") \n\
<Key>KP_8: string(0x1b) string("Ox") \n\
<Key>KP_9: string(0x1b) string("Oy") \n
!       <Key>Up: string(0x1b) string("[A") \n\
!       <Key>Down: string(0x1b) string("[B") \n\
!       <Key>Right: string(0x1b) string("[C") \n\
!       <Key>Left: string(0x1b) string("[D") \n\
*visualBell:    true
*saveLines:    1000
*cursesemul:    true
*scrollKey: true
*scrollBar: true

Note that real VT220 terminals use shifted function keys to mean something different: the user-programmable keys (i.e., DECUDK). Modern xterm supports this, but the translations do not (they're using shift to select F13 to F20).

Here's another one, from Robert Ess <>:

#               vax
# 09-17-96      Bob Ess      - initial creation
# 09-26-96      Shig Katada  - Additional keybindings
#               Script file to incorporate keybindings and command line
#               options for connecting to a VAX node
# Usage statement
        echo " Usage  : vax -options"
        echo " Options: -80   for 80 column terminal"
        echo "          -132  for 132 column terminal"
        echo "          -fg colorname"
        echo "          -bg colorname"
        echo "          -fn fontname"
        echo "          -fb bold fontname"
        echo "          -host [altair] [devel] [leonis] [castor]"
        echo ""
        echo " Example: \"vax -80 -fg white -bg black -fn 9x15 -fb 9x15b -host castor\""
        echo "          Starts a VAX session with an 80 column terminal"
        echo "          with a black background, white foreground, a normal"
        echo "          font of 9x15 and a bold font of 9x15b, and connects"
        echo "          to the node 'castor'"
        echo "          If you need additional help, please call Workstation"
        echo "          Services at x92396."
        exit 1
# Default to a black foreground with a white background.
# Use the 9x15 and 9x15bold fonts. Connect to castor by default.
# Parse the command line arguments
while [ $# != 0 ];
        case $1 in
                -80)    COLS=80
                -132)   COLS=132
                -fg)    shift
                -bg)    shift
                -fn)    shift
                -fb)    shift
                -host)  shift
                -help)  Usage;;
                *)      Usage;;
xterm  -title "VAX" -sb -sl 1200 -geo ${COLS}x24 -fg ${FG} -bg ${BG} \
        -cr red -fn ${FONT} -fb ${BFONT} -xrm \
        'XTerm*VT100.translations:     #override \n\
        <Key>Insert:            string(\001) \n\
        Shift <Key>Up:          scroll-back(1,lines) \n\
        Shift <Key>Down:        scroll-forw(1,lines) \n\
        Shift <Key>Right:       string(0x1b) string("f") \n\
        Shift <Key>Left:        string(0x1b) string("b") \n\
        Shift <Key>Delete:      string(0x1b) string(0x08) \n\
        Shift <Key>Tab:         string(0x1b) string("*") \n\
        <Key>0x1000FF0D:        scroll-back(1,page) \n\
        <Key>0x1000FF0E:        scroll-forw(1,page) \n\
        <Key>0x1000FF09:        string(\010) \n\
        <Key>0x1000FF0A:        string(\005) \n\
        <Key>BackSpace:         string(0xff) \n\
        <Key>Select:            select-start() \n\
        <Key>0x1000FF02:        select-end(PRIMARY,CUT_BUFFER0) \n\
        Meta <Key>0x1000FF02:   select-end(CLIPBOARD) \n\
        <Key>0x1000FF04:        insert-selection(PRIMARY,CUT_BUFFER0) \n\
        Meta <Key>0x1000FF04:   insert-selection(CLIPBOARD) \n\
        <Key>F1:                string(0x1b) string("OP") \n\
        <Key>F2:                string(0x1b) string("OQ") \n\
        <Key>F3:                string(0x1b) string("OR") \n\
        <Key>F4:                string(0x1b) string("OS") \n\
        <Key>F5:                string(0x1b) string("OA") \n\
        <Key>F11:               string(0x1b) string("[23~") \n\
        <Key>F12:               string(0x1b) string("[24~") \n\
        <Key>KP_0:              string(0x1b) string("Op") \n\
        <Key>KP_1:              string(0x1b) string("Oq") \n\
        <Key>KP_2:              string(0x1b) string("Or") \n\
        <Key>KP_3:              string(0x1b) string("Os") \n\
        <Key>KP_4:              string(0x1b) string("Ot") \n\
        <Key>KP_5:              string(0x1b) string("Ou") \n\
        <Key>KP_Divide:         string(0x1b) string("OP") \n\
        <Key>KP_Multiply:       string(0x1b) string("[29~") \n\
        <Key>KP_Enter:          string(0x1b) string("OM") \n\
        <Key>KP_Subtract:       string(0x1b) string("Om") \n\
        <Key>KP_Add:            string(0x1b) string("Ol") \n\
        <Key>KP_Decimal:        string(0x1b) string("On") \n\
        <Btn1Down>:             select-start() \n\
        <Btn1Motion>:           select-extend() \n\
        <Btn1Up>:               select-end(PRIMARY,CUT_BUFFER0) \n\
        Button1<Btn2Down>:      select-end(CLIPBOARD) \n\
        Button1<Btn2Up>:        ignore()'
        -e telnet $HOST &

Finally (for the moment) is a further modification of Robert Ess's script by Erik Ahlefeldt, <>. From his readme file, for vmsterm:

This script is for people who wish to connect from a Linux or Unix computer to a VMS computer using telnet and get a good VT100 or VT220 emulation. The key mappings have been specifically designed to emulate the VT terminal auxiliary numeric keypad, so that you can use VMS EDT and TPU editors, as well as the many VMS applications use keys PF1 to PF4. The script should work with any recent version of Xterm using a standard extended IBM PC keyboard or a Sun keyboard.

About the keymappings. First the auxiliary numeric keypad. My prime objective with these mappings was to produce a setup that I could use with the EDT and TPU editors which make extensive use of the numeric keypad. The top row of keys PC numeric keypad (Num Lock, Divide, Multiply, Subtract) are where you find PF1, PF2, PF3, PF4 on a VT keyboard, so I have mapped them to PF1 thru PF4. The PC numeric keypad Add key (+) takes up the space of two keys which are Minus and Comma on the VT keyboard – I have mapped it to Comma (Delete Character in the EDT editor). I have then used the PC Pause key to map to VT key Minus (Delete Word in the EDT editor). The remaining keys on the auxiliary numeric keypad are the same for PC and VT.

The six keys between the main and numeric keypads on the PC (Insert, Home, Page Up, Delete End, Page Down) are usually mapped to the VT keys by either position or by (approximate) function. As I rarely use these keys I have mapped them by function as follows: PC key Insert to VT Insert Here, PC Home to VT Find, PC Page Up to VT Prev, PC Delete to VT Remove, PC End to VT Select, PC Page Down to VT Next.

Function keys.
There are 12 function keys on the PC keyboard and 20 on the VT keyboard, so I map PC F1 thru F12 to VT F1 thru F12 (except for F1 thru F5 as noted below) and PC Shift F1 thru Shift F10 to VT F11 thru F20.

The VT keys F1 thru F5 are local hardware function keys so there is nothing to emulate, however some PC to VT emulations in the past have mapped PF1 thru PF4 here, so I have done that too, even though they are already mapped on the auxiliary numeric keypad.

Xterm functionality.
You lose some xterm functions when you remap the keyboard, however this script implements a scroll back buffer of 1000 lines which you scroll through using Shift and Up (a.k.a. Up Arrow or Cursor Up key) or Shift and Down.

a summary of the keyboard mapping:

PC Key     maps to   VT Key.
------               ------
F1                   PF1
F2                   PF2
F3                   PF3
F4                   PF4
F5                   unused
F6                   F6
F7                   F7
F8                   F8
F9                   F9
F10                  F10
F11                  F11
F12                  F12
Shift F1             F11
Shift F2             F12
Shift F3             F13
Shift F4             F14
Shift F5             F15 (Help)
Shift F6             F16 (Do)
Shift F7             F17
Shift F8             F18
Shift F9             F19
Shift F10            F20
Shift F11            F11
Shift F12            F12
Print                Help (F15)
Cancel               Do   (F16)
Pause                Keypad Minus

Insert               Insert Here
Delete               Remove
Home                 Find
End                  Select
Prior                Prev
Next                 Next
BackSpace            BackSpace (sends DEL - ascii 127)

Num_Lock             PF1
KP_Divide            PF2
KP_Multiply          PF3
KP_Subtract          PF4
KP_Add               Keypad Comma
KP_Enter             Enter
KP_Decimal           Period
KP_0                 Keypad 0
KP_1                 Keypad 1
KP_2                 Keypad 2
KP_3                 Keypad 3
KP_4                 Keypad 4
KP_5                 Keypad 5
KP_6                 Keypad 6
KP_7                 Keypad 7
KP_8                 Keypad 8
KP_9                 Keypad 9
Up                   Up
Shift Up             Scroll Back
Down                 Down
Shift Down           Scroll Forward
Right                Right
Left                 Left

and the script:

#               vmsterm
#               from an original script by Bob Ess
#               key translations by Erik Ahlefeldt 
#               Script file using Xterm and telnet to connect to a VMS host
#               and give a decent vt220 emulation.
# Usage statement
        echo " Usage  : vmsterm -options"
        echo " Options: -80   for 80 column terminal"
        echo "          -132  for 132 column terminal"
        echo "          -bg colorname"
        echo "          -fg colorname"
        echo "          -fn fontname"
        echo "          -fb bold fontname"
        echo "          -host [] [earth] []" 
        echo ""
        echo " Example: \"vmsterm -80 -fg white -bg black -fn 9x15 -fb 9x15b -host earth\""
        echo "          Starts a VMS session with an 80 column terminal"
        echo "          with a black background, white foreground, a normal"
        echo "          font of 9x15 and a bold font of 9x15b, and connects"
        echo "          to the node 'earth'"
        echo ""
        echo " Example: \"vmsterm -host earth\""
        echo "          Starts a VMS session with default terminal settings "
        echo ""
        echo " Example: \"vmsterm -help\""
        echo "          Displays vmsterm options "
        exit 1
# Default to a black foreground with a white background.
# Use the 9x15 and 9x15bold fonts. Connect to by default.
# Parse the command line arguments
while [ $# != 0 ];
        case $1 in
                -80)    COLS=80
                -132)   COLS=132
                -fg)    shift
                -bg)    shift
                -fn)    shift
                -fb)    shift
                -host)  shift
                -help)  Usage;;
                *)      Usage;;
xterm   -title "VMSTERM" -sb -sl 1000 -geo ${COLS}x24 -fg ${FG} -bg ${BG} \
        -cr blue -fn ${FONT} -fb ${BFONT} -xrm \
        "XTerm*vt100.translations: #override \\
        ~Shift  <Key>F1:        string(0x1b)    string("
OP") \\
        ~Shift  <Key>F2:        string(0x1b)    string("
OQ") \\
        ~Shift  <Key>F3:        string(0x1b)    string("
OR") \\
        ~Shift  <Key>F4:        string(0x1b)    string("
OS") \\
        ~Shift  <Key>F5:        string("
Break") \\
        ~Shift  <Key>F6:        string(0x1b)    string("
[17~") \\
        ~Shift  <Key>F7:        string(0x1b)    string("
[18~") \\
        ~Shift  <Key>F8:        string(0x1b)    string("
[19~") \\
        ~Shift  <Key>F9:        string(0x1b)    string("
[20~") \\
        ~Shift  <Key>F10:       string(0x1b)    string("
[21~") \\
        ~Shift  <Key>F11:       string(0x1b)    string("
[23~") \\
        ~Shift  <Key>F12:       string(0x1b)    string("
[24~") \\
        Shift   <Key>F1:        string(0x1b)    string("
[23~") \\
        Shift   <Key>F2:        string(0x1b)    string("
[24~") \\
        Shift   <Key>F3:        string(0x1b)    string("
[25~") \\
        Shift   <Key>F4:        string(0x1b)    string("
[26~") \\
        Shift   <Key>F5:        string(0x1b)    string("
[28~") \\
        Shift   <Key>F6:        string(0x1b)    string("
[29~") \\
        Shift   <Key>F7:        string(0x1b)    string("
[31~") \\
        Shift   <Key>F8:        string(0x1b)    string("
[32~") \\
        Shift   <Key>F9:        string(0x1b)    string("
[33~") \\
        Shift   <Key>F10:       string(0x1b)    string("
[34~") \\
        Shift   <Key>F11:       string(0x1b)    string("
[28~") \\
        Shift   <Key>F12:       string(0x1b)    string("
[29~") \\
                <Key>Print:     string(0x1b)    string("
[28~") \\
                <Key>Cancel:    string(0x1b)    string("
[29~") \\
                <Key>Pause:     string(0x1b)    string("
Om") \\
                <Key>Insert:    string(0x1b)    string("
[2~") \\
                <Key>Delete:    string(0x1b)    string("
[3~") \\
                <Key>Home:      string(0x1b)    string("
[1~") \\
                <Key>End:       string(0x1b)    string("
[4~") \\
                <Key>Prior:     string(0x1b)    string("
[5~") \\
                <Key>Next:      string(0x1b)    string("
[6~") \\
                <Key>BackSpace: string(0x7f)    
                <Key>Num_Lock:  string(0x1b)    string("
OP") \\
                <Key>KP_Divide: string(0x1b)    string("
OQ") \\
                <Key>KP_Multiply: string(0x1b)  string("
OR") \\
                <Key>KP_Subtract: string(0x1b)  string("
OS") \\
                <Key>KP_Add:    string(0x1b)    string("
Ol") \\
                <Key>KP_Enter:  string(0x1b)    string("
OM") \\
                <Key>KP_Decimal: string(0x1b)   string("
On") \\
                <Key>KP_0:      string(0x1b)    string("
Op") \\
                <Key>KP_1:      string(0x1b)    string("
Oq") \\
                <Key>KP_2:      string(0x1b)    string("
Or") \\
                <Key>KP_3:      string(0x1b)    string("
Os") \\
                <Key>KP_4:      string(0x1b)    string("
Ot") \\
                <Key>KP_5:      string(0x1b)    string("
Ou") \\
                <Key>KP_6:      string(0x1b)    string("
Ov") \\
                <Key>KP_7:      string(0x1b)    string("
Ow") \\
                <Key>KP_8:      string(0x1b)    string("
Ox") \\
                <Key>KP_9:      string(0x1b)    string("
Oy") \\
        ~Shift  <Key>Up:        string(0x1b)    string("
[A") \\
        Shift   <Key>Up:        scroll-back(1,lines) 
        ~Shift  <Key>Down:      string(0x1b)    string("
[B") \\
        Shift   <Key>Down:      scroll-forw(1,lines) 
                <Key>Right:     string(0x1b)    string("
[C") \\
                <Key>Left:      string(0x1b)    string("
[D")" \
        -e telnet $HOST 

How do I set the title?

The control sequences for doing this are documented in

The usual context for this question is setting the title according to the current working directory. People post answers to this periodically on the newsgroups. Here is one that I have seen, from Roy Wright <>. In your /etc/profile after:

if [ "$SHELL" = "/bin/pdksh" -o "$SHELL" = "/bin/ksh" ]; then
        PS1="! $ "
elif [ "$SHELL" = "/bin/zsh" ]; then
        PS1="%m:%~%# "
elif [ "$SHELL" = "/bin/ash" ]; then
        PS1="$ "
        PS1='\u@\h:\w\$ '


if [ "$TERM" = "xterm" ]; then
        PS1="\033]2;\u@\h:\w\007bash$ "

The terminator "\007" is a problem area. XTerm historically uses this character, though it is non-ANSI. The "correct" character should be a "\233" string terminator, or "\033\\", which is the 7-bit equivalent. Modern xterm recognizes either (the "\007" or string terminator); waiting for the first of these.

You may have resource or environment problems that prevent you from setting the title at all. Newer xterms (starting somewhere in X11R5) use the $LANG variable. If your locale is incorrectly installed, you will be unable to set the xterm's title. As noted by Mikhail Teterin <>: Make sure that the locale (LANG and/or LOCALE environment variable) is known to X Window System. Check ${X11ROOT}/lib/X11/locale.* for it. If it is not listed in either one of the files, find the nearest match and add an alias to it. Restart X if you have made changes.

On a related note, some people want to know how to read the title from an xterm. This works for modern xterm and dtterm, but not for other variations:

# Echo the current X term title bar to standard output.
# Written by Icarus Sparry <> 11 Apr 1997
exec </dev/tty
old=$(stty -g)
stty raw -echo min 0  time ${1-10}
print "\033[21t\c" > /dev/tty
IFS='' read -r a
stty $old
print -R "${b%??}"

But it is possible to avoid escape sequences altogether (from Hemant Shah <>):

$ xprop -id $WINDOWID | grep WM_NAME
WM_NAME(STRING) = "this is my title"
current_title=$(xprop -id $WINDOWID | grep WM_NAME | cut -d= -f2)

Here's another source of information: Xterm-Title HowTo

Standard xterm does not implement a blinking cursor. Some of the variations do: dtterm, GNOME Terminal, and modern xterm (from mid 1999, patch 107).

Frequent problems

Starting xterm, or not

XTerm does not run (no available pty's)

Your copy of xterm may not have enough permissions to use existing pty's:

Perhaps your system does not have enough pty's, or (problems reported with newer Linux kernels supporting Unix98 pty's, beginning with RedHat 6.0) the major device numbers of the pty's may have changed during a kernel upgrade. (This is described in /usr/src/linux/Documentation).

See also the MAKEDEV script, which usually exists under /dev.

I need /etc/termcap

If you have a termcap version of xterm on a system with no termcap libraries, you may also be missing /etc/termcap.

A workaround is to copy /usr/X11R6/lib/X11/etc/xterm.termcap to /etc/termcap.

This is fixed another way starting with XFree86 3.3.1. If xterm cannot find the terminal description, it will accept that, though it will print a warning. If xterm does not find the termcap entry, it will not set the $TERMCAP variable.

Why does $LD_LIBRARY_PATH get reset?

If xterm is running setuid (which is needed on some systems which have no wrappers for opening pty's and updating utmp), newer systems automatically set or reset environment variables which are considered security problems. These include $PATH and $LD_LIBRARY_PATH, since they affect the choice of which programs are run if not specified via a full pathname.

This means, for example, that if you attempt to run

xterm -e foo

where foo is a program that uses shared libraries in /usr/local/lib, then the command will fail, because /usr/local/lib is not considered part of root's environment.

Modern Unix systems (such as recent Solaris and HPUX versions) do not require you to run xterm setuid. Some will result in odd malfunctions if you do this.

Why do the -e and -ls options not work together?

XTerm has two useful options for controlling the shell that is run:

tells xterm to execute a command using the remaining parameters after this option.
tells xterm to invoke a login shell, making it read your .login file, for instance.

The two are not compatible. If you specify both, xterm uses -e, and if that fails for whatever reason will fall through to the -ls option. It cannot (in general) combine the two, since some shells permit this (e.g., bash), and others do not (e.g., tcsh).

Why is my screen size not set?

Well, it may be set, but not correctly. You may notice these symptoms:

XTerm knows how big the screen is (of course), and tries to tell your applications (e.g., by invoking ioctl's and sending SIGWINCH). But sometimes it cannot:

Most full-screen applications such as vi are designed to use the ioctl calls that return the screen size. When they fail, the applications use the size defined in the terminal's terminfo or termcap description.

You may be able to use the resize program to issue the ioctl's that will notify your application of the actual screen size. This does not always work for the reasons just mentioned. Newer versions of stty let you specify the screen size, though it will not be updated if you resize the xterm window:

stty rows 24 columns 80

Most full-screen applications also check if the $LINES and $COLUMNS variables are set, using those values to override the terminal description:

setenv LINES 24
setenv COLUMNS 80

Why 65 lines? The standard xterm terminfo description specifies 65 lines, perhaps because someone liked it that way. Real VT100's are 24 lines. I once used (and wrote applications for) a Bitgraph terminal, which emulated VT100, but displayed 65 lines.

Why are the menus tiny?

Everything seems to work, except that the xterm menus (VT options, fonts, etc.) do not display properly; the menus pop up, but only with a tiny display area in which none of the options are visible (and only part of the menu title is visible).

You have specified the geometry for xterm too high in the hierarchy, and that 24x80 (or whatever the -geometry parameter happens to be) is applying to the menus in pixels. This resource makes the geometry apply to the menus as well as the VT100 widget:

XTerm*geometry: 80x24

while this applies only to the VT100 widget (which is probably what you intended):

XTerm.VT100.geometry: 80x24

or better yet (to allow for the toolbar option, which uses a level of widget hierarchy):

XTerm*VT100.geometry: 80x24

Font problems

My terminal doesn't show box characters

XTerm displays the 7-bit ASCII and VT100 graphic characters (including box corners) using specially arranged fixed-pitch fonts. The first 32 glyph positions (which would correspond to nonprinting control characters) are used to hold the VT100 graphic characters. Some fonts that otherwise look fine (such as courier) do not have glyphs defined for these positions. So they display as blanks. Use xfd to display the font.

Modern xterm can form its own line-drawing characters (see patch 90, for example). It does not draw all of the graphic characters, only those that may be done with straight lines. But those are the most used, making most of the fixed-pitch fonts useful for xterm.

You may also have a problem with the terminfo description. As distributed, the X11R6 terminfo for xterm does not have the acsc string defined, so most implementations of curses do not try to use the alternate character set.

Finally, some people confuse the VT100 graphic characters with the VT220 support for DEC technical character set. These are distinct (7-bit) character sets. Xterm currently does not support this.

The bold font is ugly

XTerm lets you directly specify one bold font, which is assumed to correspond to the default font. Older versions of xterm make a fake bold font for the other choices via the fonts menu by drawing the characters offset by one pixel. I modified xterm to ask the font server for a bold font that corresponds to each font (other than the default one). Usually that works well. However, sometimes the font server gives a poor match. Xterm checks for differences in the alignment and size, but the font server may give incorrect information about the font size. The scaled bitmap font feature gives poor results for the smaller fonts. In your X server configuration file, that can be fixed by disabling the feature, e.g., by appending ":unscaled" to the path:

FontPath        "/usr/lib/X11/fonts/100dpi/:unscaled"
FontPath        "/usr/lib/X11/fonts/75dpi/:unscaled"
FontPath        "/usr/lib/X11/fonts/misc/:unscaled"

You can suppress xterm's overstriking for bold fonts using the alwaysBoldMode and related resources. However, rendering ugly bold fonts is a "feature" of the font server. In particular, the TrueType interface provides less ability to the client for determining if a particular font supports a bold form.

I see little dots on the screen

Well, I do. Perhaps you do not. It depends on the fonts you choose, and how you use them.

Standard xterm has a "normal" font for which a bold font can be chosen, and several alternative fonts, useful for changing the font size. The alternative fonts do not have corresponding bold fonts. Xterm simulates bold fonts in this case by overstriking the character one pixel offset. That can make an bold character extend into the area that another character occupies. When erasing a bold character from the screen, xterm does not erase the extra pixel. This is corrected in modern xterm, subject to the available fonts (from late 1998, patch 85). For each font, it asks the font server for a corresponding bold font. Your font server may not have the bold font (or it may incorrectly report that it does). But it usually works.

My terminal doesn't display Cyrillic characters

Cyrillic encodings typically use characters in the range 128-159. For a VT220 (or any terminal that follows ISO 6429), those are treated as control characters. Still, some people want to use KOI8-R, etc. I modified xterm in patch 175 to add an option (-k8) and corresponding resource settings to allow them to customize their environment. Here is a sample script and resource file which I use for testing this configuration.

I see boxes instead of characters in uxterm

XTerm may show boxes instead of characters if the font that you have selected does not contain those characters. Normally you can fix most of that using the UTF-8 feature, with uxterm. However, your X resource settings may be the source of the problem.

One pitfall to setting X resources is that they allow you to specify wildcards, e.g., the "*" character. When you give a wildcard, the X resource matches any number of levels in the widget hierarchy.

XTerm has more than one widget matching "font" at different levels of the hierarchy. There are the popup menus, and there are the fonts used for uxterm. The latter is where an overbroad pattern can cause xterm to use a different font than you expect.

Suppose your resource setting includes this pattern

*VT100*font: fixed

It could be interpreted as this:

*VT100.font: fixed
*VT100.utf8Fonts.font: fixed

XTerm uses the utf8Fonts subresources to provide runtime-switchable fonts between IS0-8859-1 (Latin-1) and ISO-10646 (Unicode). Modifying the Unicode font to "fixed" will make most of the characters unavailable (i.e., shown as boxes). If instead your resource looks like

*VT100.font: fixed

it would be unambiguous, and not modify the utf8Fonts value.

I see question-marks instead of characters in uxterm

You may be seeing a question mark in a black diamond, like this:

That is called the Unicode replacement character. It is used when a program is told to display a character which is not in the Unicode system. The program replaces the illegal/invalid character with this symbol. That is different from boxes, which xterm displays when it has nothing in the current font for a valid Unicode character.

Unicode's rules are oriented toward handling input in different encodings, and displaying the corresponding Unicode value. Although Unicode has rules and a definition for the valid characters, different programs may handle invalid input in different ways. This page gives an overview of how xterm may differ from other programs in that regard.

The first popup menu is very slow

Some users report that when starting xterm, it is very slow, that their computer's CPU time increases, etc.

This is a longstanding bug in the X libraries. There is a workaround using a resource setting for xterm.


XTerm uses the Athena (Xaw) widgets to display popup menus. In the normal case, those are initialized one-by-one as they are first used. If you have configured xterm to use its toolbar configuration, they are all initialized on startup. In the latter, performance problems are more noticeable.

The Athena widgets XawInitializeWidgetSet function goes through several levels down to the X library _XlcAddUtf8LocaleConverters function to call create_tocs_conv and related functions to make a list of character sets from the locale, which is used in menus to get all possible fonts needed for a fontset.

If your current locale uses UTF-8 encoding, this will read a long list of bitmap fonts—everything whose encoding might be useful for displaying the menus. For example, this list (from lcUTF8.c) which dates from around 2000 is the core of the problem:

ISO10646-1, ISO8859-1, ISO8859-2, ISO8859-3, ISO8859-4, ISO8859-5, ISO8859-6, ISO8859-7, ISO8859-8, ISO8859-9, ISO8859-10, ISO8859-11, ISO8859-13, ISO8859-14, ISO8859-15, ISO8859-16, JISX0201.1976-0, TIS620-0, GB2312.1980-0, JISX0208.1983-0, JISX0208.1990-0, JISX0212.1990-0, KSC5601.1987-0, KOI8-R, KOI8-U, KOI8-C, TATAR-CYR, ARMSCII-8, IBM-CP1133, MULELAO-1, VISCII1.1-1, TCVN-5712, GEORGIAN-ACADEMY, GEORGIAN-PS, ISO8859-9E, MICROSOFT-CP1251, MICROSOFT-CP1255, MICROSOFT-CP1256, BIG5-0, BIG5-E0, BIG5-E1, ISO10646-1, ISO10646-1

However, xterm is going to use only the characters shown in the popup menus. It is unlikely that you need Chinese fonts for that.


XTerm's menuLocale resource can be set to an explicit value, e.g., "C" to override the current locale as seen by this initialization debacle.


The workaround does not prevent some hacker from "improving" the X libraries still further.

Keyboard problems

Why can't I input 8-bit characters?

You must have the eightBitInput resource set to do this.

Why doesn't my delete key work?

This seems to have begun as a problem with the older XFree86 release (3.1.2). I have picked up pieces of the story (xterm and the keyboard work as designed under XFree86 3.2 and up).

The underlying problem is that we've accumulated three things that are being equated as "Delete":

ASCII BS (backspace, code 8)
ASCII DEL (delete. code 127)
VT220 "remove" aka "delete" (ESC [ 3 ~)

You are probably talking about the backarrow key (on my keyboard, at the upper right of the QWERTY block), or the key labeled delete which is on the 6-key "editing keypad". Since xterm is emulating a VT100/VT220, the backarrow key should generate a 127 (often displayed as ^?). You would use a control/H to obtain a backspace on a real VT220.

The reason why BS and DEL are of special interest is that on Unix, the stty command and the underlying termios/termio system calls allow only single-byte codes to be assigned to special functions such as erase. For instance, you could see something like this on your terminal:

$ stty -a
speed 38400 baud; rows 40; columns 80; line = 0;
intr = ^C; quit = ^\; erase = ^H; kill = ^U; eof = ^D; eol = <undef>;
eol2 = <undef>; swtch = <undef>; start = <undef>; stop = <undef>; susp = <undef>;
rprnt = ^R; werase = ^W; lnext = ^V; flush = ^O; min = 1; time = 0;
-parenb -parodd cs8 -hupcl -cstopb cread -clocal -crtscts
-ignbrk brkint -ignpar -parmrk -inpck -istrip -inlcr -igncr -icrnl ixon -ixoff
-iuclc -ixany -imaxbel -iutf8
-opost -olcuc -ocrnl -onlcr -onocr -onlret -ofill -ofdel nl0 cr0 tab0 bs0 vt0 ff0
isig -icanon -iexten -echo -echoe -echok -echonl -noflsh -xcase -tostop -echoprt
-echoctl -echoke

Tastes differ. On Unix, people expect the backarrow key to generate a backspace (or not). As I understand it, at one point, XFree86 picked up the sense of the erase character during initialization, so that xterm would in effect use the same erase character as the console. The current scheme (X11R6) uses keyboard mapping tables that are independent of the environment.

Modern xterm (since patch #83 in 1998) provides a resource toggle backarrowKey (and an escape sequence from VT320) that changes this key between the two styles (backspace or delete).

With modern xterm patch 95 (also in the stable version as "88c"), you may have an xterm which can automatically initialize the backarrow key to backspace or delete depending on the pseudo terminal's sense, or based on the termcap setting of kbs (backspace key). This feature is controlled by the resource setting ptyInitialErase.

Why did my delete key stop working?

Well, something changed. You have to determine what did.

This may be because an upgrade introduced different X resource settings, or because you are using the newer xterm with the ptyInitialErase resource (or perhaps both). Use

appres XTerm

to see the X resources that you are using, in particular the translation (or Translation) resource for the vt100 widget.

One unexpected scenario came out of hiding when I was implementing the ptyInitialErase resource. When xterm is (by default) built to support this, it sets the pty's erase character to match the termcap entry. Xterm also sets the $TERMCAP environment variable to match. So everything is consistent, and everything defined. The stty erase character is either backspace (^H) or delete (^?).

The problem arises because there are two things called "delete", which were not well-defined: ASCII delete (127) and the PC-style adaptation of VT220 remove assigned to the key Delete.

However, the screen program prefers to make the termcap delete (kD) an <escape>[3~, which corresponds to the VT220 remove key. If $TERMCAP is set when starting screen, it will translate stty's erase character into the <escape>[3~, making most curses and termcap applications work. But stty still has the original erase character. So low-level applications which check stty will not work. I found that unsetting $TERMCAP before running would work, but this was not a good solution. Someone pointed out (see patch 129), that the problem really was because termcap kD should delete the character at the current position. So it cannot be the same as stty erase.

As a matter of fact, stty erase has to be a single character, so <escape>[3~ would not work anyway.

Well, how can I set my delete key?

When people first started asking this question in 1995-1996, it appeared in the context of making Netscape work. Netscape's use of the delete key running in X did not match user's expectations when compared to that other platform. They were commonly advised to use xmodmap, e.g.,

keysym BackSpace = Delete


keycode 22 = 0xff08

Either way is a bad technical solution – it works for some people but not others (on my keyboard at work, keycode 22 is the numeric keypad '9').

Alternatively, you can set resources. This works reasonably well for environments where you have different versions of xterm, e.g.,

XTerm*VT100.translations: #override \n\
        <Key>Delete: string(

I do not do that either, because it is not flexible. Not all programs use the same sense of stty erase; some use termcap or terminfo, and some are hardcoded. So I prefer to be able to switch the xterm's keyboard at runtime. You cannot do that with resources. (Or not really – xterm has a keymap() action which could support this if you provided a rather complex resource settings, but the X library support for that is broken in X11R6). Instead, I have added to xterm a set of resources (and popup menu entries) to allow simple switching between the different styles of keyboard, in particular for the backspace/delete issues. See the manual page for backarrowKey backarrowKeyIsErase and deleteIsDEL as well as sunKeyboard.

Why doesn't my keypad work?

A few people have commented that the keypad does not work properly. Aside from bugs (I have fixed a few), the most common problem seems to be misconception.

Here's a picture of the VT100 numeric keypad:

| PF1 | PF2 | PF3 | PF4 |
|  7  |  8  |  9  |  -  |
|  4  |  5  |  6  |  ,  |
|  1  |  2  |  3  |     |
+-----+-----+-----+ ENT +
|     0     |  .  |     |

and the similar Sun and PC keypads:

| NUM |  /  |  *  |  -  |
|  7  |  8  |  9  |     |
+-----+-----+-----+  +  +
|  4  |  5  |  6  |     |
|  1  |  2  |  3  |     |
+-----+-----+-----+ ENT +
|     0     |  .  |     |

Working in X11, the NUM (NumLock) key has better uses than an alias for PF1 (and is sometimes reserved). I use the F1 through F4 on the keyboard to implement PF1 through PF4, alias the keypad "+" to "," and use the existing "-" key.

VT220 emulation uses the VT100 numeric keypad as well as a 6-key editing keypad. Here's a picture of the VT220 editing keypad:

| Find   | Insert | Remove |
| Select | Prev   | Next   |

and the similar Sun and PC keypads:

| Insert | Home   | PageUp |
| Delete | End    | PageDn |

I chose to use keys that are mnemonic rather than in the "same" positions, though some emulators (e.g., Tera Term) use the same positions:

| Insert | Find   | Prev   |
| Remove | Select | Next   |

I test the keyboard (for VT52/VT100/VT220) using vttest. If you find (or think that you have found) a problem with the keyboard handling of xterm, please test it with vttest first.

Other arrangements of the keyboard are possible of course. If you prefer to use the top row of the numeric keypad as PF1 through PF4, you should do this using xterm's X resources.

In 2014, I noticed a comment, which relates to the PF1-PF4 assignment, but also to the use of function-key modifiers.
Because that is a digression, I have expanded it in a separate page.

How can my program distinguish control-I from tab?

Your program can use xterm's modifyOtherKeys feature, e.g., using a control sequence such as

\033 [ > 4 ; 2 m

to temporarily switch into a mode where shift, alt, control and meta modifiers applied to a key tell xterm to send an escape sequence which encodes all of that information. The tab key would send a tab, but control-i would send

\033 [ 2 7 ; 5 ; 1 0 5 ~

while control-shift-i would send

\033 [ 2 7 ; 6 ; 7 3 ~

Your program should turn that off when it is not needed; it is a real nuisance when you cannot type control-characters as they were meant to be used:

\033 [ > 4 m

In 2019, Bram Moolenaar asked for more details about modifying other keys; I have expanded that in a separate page.

Why can't I use the pageup/pagedown keys?

Some vendors, e.g,. Sun, added key translations which make the pageup and pagedown keys talk to the xterm's scrollbar instead of your application. They did the same thing for the home and end keys, thereby obscuring a bug in xterm.

You can override this by specifying your own translations in your resource file. The issue was first noted with Solaris 2.5, with the file given in two locations:


using a symbolic link to relate the two. Later releases of Solaris, e.g., 8-10 omitted the former location.
Solaris 11 provides modern xterm (patch #271), and does not have this problem.

As of February 2014, I was able to verify that AIX and HPUX have updated to modern xterm, e.g.,

Older AIX and HPUX releases distributed the X Consortium (1994) app-defaults file.

In updating this question in February 2014, I noticed that IBM added their copyright notice in AIX's copy of the app-defaults file in


There were no other changes to the file. Someone at IBM blundered.
In patch #252, I ensured that my copyright notice is on those files (I am the sole author, and can do that).

Use the translations in the system's app-defaults file as a guide. The relevant section of the app-default file looks like

*VT100.translations:    #override \
<Key>KP_0: string(0)\n\
<Key>KP_1: string(1)\n\
<Key>KP_2: string(2)\n\
<Key>KP_3: string(3)\n\
<Key>KP_4: string(4)\n\
<Key>KP_5: string(5)\n\
<Key>KP_6: string(6)\n\
<Key>KP_7: string(7)\n\
<Key>KP_8: string(8)\n\
<Key>KP_9: string(9)\n\
<Key>KP_Add: string(+)\n\
<Key>KP_Decimal: string(.)\n\
<Key>KP_Divide: string(/)\n\
<Key>KP_Enter: string(\015)\n\
<Key>KP_Equal: string(=)\n\
<Key>KP_Multiply: string(*)\n\
<Key>KP_Subtract: string(-)\n\
<Key>F16: start-extend() select-end(PRIMARY, CUT_BUFFER0, CLIPBOARD) \n\
<Key>F18: insert-selection(PRIMARY, CLIPBOARD) \n\
<Key>F27: scroll-back(100,page) \n\
<Key>R13: scroll-forw(100,page) \n\
<Key>Home: scroll-back(100,page) \n\
<Key>End: scroll-forw(100,page) \n

For example, a more-specific pattern for the resource name lets you override:

XTerm*VT100.translations:       #override \n\
<Key>Home:  string(\033[1~)\n\
<Key>End:   string(\033[4~)\n\
<Key>Prior: string(\033[5~)\n\
<Key>Next:  string(\033[6~)\n\
<Key>Prior:  scroll-back(1,page) \n\
<Key>Next:   scroll-forw(1,page) \n\
<Key>Home:   scroll-back(100,page) \n\
<Key>End:    scroll-forw(100,page) \n

makes the home/end and pageup/pagedown keys usable by your editor, while leaving their shifted equivalents available for the scrollbar.

Why can't I use the home/end keys?

This is a long story, unless you are referring to X Consortium xterm. That program is simply broken in this respect.

At the beginning, when the home/end keys were fixed for modern xterm (in early 1996), there was some discussion regarding what the escape sequences should be for those keys (for the 6-key editing keypad). Those were chosen as "PC-style" codes (like SCO "ansi"), i.e.,


for normal mode, and


for cursor application mode.

That style of coding fit easily into the existing logic of xterm. It was not my change, and (because xterm should be based upon standards), I did question this, and asked the opinion of the person who was at that time developing rxvt. He had chosen a layout based on DEC's VT220 terminals, though the key labels on the typical PC keyboard did not match. At that point, neither of us knew enough to make a good case for this.

Somewhat later I could see that xterm had a number of undocumented extensions to support the VT220-style (pre-ISO 2022) character sets. I decided to complete the functionality by making xterm a VT220 emulator. This would require that it provide the same escape sequences for the editing and numeric keypads. I could not simply change the escape sequences from "PC-style" to "VT220-style", since a number of users "knew" that the keypad "ought to" send home, end, cursor keys, etc., because they had labels indicating that use. To retain compatibility (but allow easy reconfiguration to make a VT220 emulator), I added popup-menu items to switch between the modes. With minor refinements, this was the approach for about two years, culminating with the "stable" patch #88, which is essentially the version distributed with XFree86 3.3.x.

the terminfo distributed with xterm patch #88 is incorrect: the escape sequences given for home/end keys are the VT220-style, rather than the default PC-style. Too accustomed to switching modes on the fly, I overlooked a line in my .Xdefaults file:

*sunKeyboard: true

Downstream packagers (when they noticed this) accommodated the bug by modifying the VT100 translations resource which is not a good technical solution since it interferes with the users' ability to modify that resource. For example, Red Hat bug #100695 quoted a suggested patch which shows that the package had overridden the xterm behavior for shifted function keys. See this for more discussion.

But xterm continues to evolve past the stable patch #88. The keyboard support was still unsatisfactory for two reasons:

The former could be addressed by expanding the escape sequences sent by the PC-style function keys, while the latter was a VT100/VT220 design issue. I decided to redesign function-key support to separate the two styles of function keys better, but leaving the choice still controlled by the sunKeyboard resource. Partway through that, I was asked to do similar cleanup and redesign of the backspace and delete key handling, e.g., the ptyInitialErase resource. Because it is a redesign, I chose to not make the keyboard differences between the old and new xterms completely compatible. If you were to run both on the same system, one or the other would have some problems with the editing keypad or the backspace/delete keys which would be addressed by the popup-menu selections.

For example, at this time (2001/9/4):

Here is a resource file which I tested with xterm-88c, xterm-149 and xterm-158, using $TERM set to xterm-debian:

! $Id: xterm.faq.html,v 1.169 2012/02/05 11:58:56 tom Exp $
! Settings to make xterm-88c work as expected for Debian.
! Patch #88 was the basis for XFree86 3.3.1 xterm.  There were a few additions
! through patch 88c, to incorporate the ptyInitialErase resource.  Debian uses
! the VT220-style keyboard, which at #88 was the xterm-xfree86 terminfo entry,
! with one change: kbs changed from ^H to ^?.
! After patch 88, I started work on keyboard changes.  The result was that the
! xterm-xfree86 terminfo entry was set to the PC-style keyboard, and I added
! xterm-vt220, which corresponded mostly to the older (patch-88) version of the
! xterm-xfree86 terminfo entry.
! The terminfo with patch #88 assumed sunKeyboard was set (actually a bug, but
! also assumed in Debian).
! A different problem (addressed after patch #88) is that if you wanted to use
! a VT100/VT220-style numeric keypad's escape sequences, you had to have
! NumLock set.  Otherwise, in keypad application mode, the keys would transmit
! only the PC-style escape sequences corresponding to the key labels, e.g., the
! page-up string rather than the escape sequence for keypad-9.
XTerm*sunKeyboard: true
! These settings overlap to some extent (backarrowKeys says to send a 127 for
! the "backspace" key, and ptyInitialErase says to use the pty's initial sense
! of the erase character, which is reported to be the same on Linux).
XTerm*backarrowKey: false
XTerm*ptyInitialErase: true

Why can't I use the cursor keys in (whatever) shell?

VTxxx (VT100 and up) terminals may send different escape sequences for the cursor (arrow) keys depending on how they are set up. The choices are referred to as the normal and application modes. Initially, the terminal is in normal mode.

VTxxx terminals are usually set up so that full-screen applications will use the cursor application mode strings. This is good for full-screen applications, including legacy applications which may have hard-coded behavior, but bad for interactive shells (e.g., ksh, tcsh, bash) which use arrow keys to scroll through a history of command strings.

To see the difference between normal/application modes, consider this example:

Since termcaps and terminfo descriptions are written for full-screen applications, shells and similar programs often rely on built-in tables of escape sequences which they use instead. Defining keys in terms of the termcap/terminfo entry (e.g., by capturing the string sent by tputs) is apt to confuse the shell.

Depending on the terminal type, the keypad(s) on the keyboard may switch modes along with the cursor keys, or have their own independent modes. The control sequences for these are independent of the ones used for cursor-addressing, but are grouped together, e.g., as the terminfo smkx and rmkx capabilities. Terminfo entries are written assuming that the application has initialized the terminal using the smkx string before it is able to match the codes given for the cursor or keypad keys.

Alt-keys do not work in bash

See Alt-keys do not work in bash.

Colors and other graphic rendition

My terminal doesn't recognize color

First, ensure that you have set up xterm to render color. Modern xterm renders color only if you have set resources to do this; the default behavior is monochrome to maintain compatibility with older applications. The manual page describes these resources. I set them in my .Xdefaults file.

Even if you set the resources properly, there may be another application running which prevents xterm from allocating the colors you have specified. But you should see a warning message for this.

Check the terminal description, to see if it is installed properly, e.g., for ncurses, which uses terminfo.

Finally, some applications (that do not interface properly with terminfo or termcap) may need the environment variable $COLORTERM to be set.

What $TERM should I use?

XTerm provides in its sources both terminfo and termcap files. They are designed to allow scripting to override the most common choices, e.g., the backspace key.

The xterm-color value for $TERM is a bad choice for modern xterm because it is commonly used for a terminfo entry which happens to not support bce. Complicating matters, FreeBSD (after dithering for a few years on the matter) introduced a bastardized version which implies the opposite sense of bce, (because it uses SGR 39 and 49), but does not set it. After lengthy discussion, FreeBSD began using the terminal descriptions which I've written.

The most recent XFree86 version's terminal description corresponds to xterm-xfree86 (also distributed with ncurses). I have continued to make changes; the most recent version is simply named xterm-new (also distributed with ncurses).

The term "bce" stands for "back color erase". Terminals such as modern xterm and rxvt implement back color erase, others such as dtterm do not. (Roughly half of the emulators that I know about implement bce). When an application clears the screen, a terminal that implements back color erase will retain the last-set background color. A terminal that does not implement back color erase will reset the background color to the default or initial colors. Applications that paint most of the screen in a single color are more efficient on terminals that support back color erase. Inevitably, there are tradeoffs and issues with standardization of the feature as noted in the ncurses FAQ. Unsurprisingly, ncurses supports xterm's behavior.

Curses libraries that support color know about bce and do the right thing – provided that you tell them what the terminal does. That is the whole point of setting $TERM. The "xterm-color" description distributed with ncurses does not list bce, because it was applied originally to a terminal type which does not implement back color erase. It will "work" for modern xterm, though less efficient. Some other applications such as the slang library have hardcoded support for terminals that implement back color erase. Given the "xterm-color" description, those will be efficient – and fortuitously work. However, slang (through version 1.4.0) did not work properly for the terminals that xterm-color was designed for. See this page for an example of (n)curses and slang running on dtterm. That bug in slang is reported to be fixed for succeeding versions, though your application may require changes to use this fix. (The demo which comes with slang to illustrate the use of bce does not work properly, for instance).

The xterm-color value for $TERM is also (for the same reason) a bad choice for rxvt, but "works" due to the large number of hard-coded applications that override this.

Some people recommend using xtermc. That is installed on Solaris. However, it does not match any xterm in current use. (Apparently it was written for an obsolete version on Unixware). The colors work, true, but the mouse will not, nor will the function keys.

Reverse video is not reset

When running less or other programs that do highlighting, you see the highlighting not turned off properly.

This may be due to incompatible terminal descriptions for xterm. With XFree86 3.2, I modified the terminal description for XFree86 xterm to use the VT220 (aka ISO 6429) controls that allow an application to turn off highlighting (or bold, underline) without modifying the other attributes. The X Consortium xterm does not recognize these controls.

If, for example, you are running an older xterm and rlogin to a system where the newer xterm has been installed, you will have this problem, because both programs default to $TERM set to xterm. The solution for mixed systems is to install the newer terminal description as as a different name (e.g., xterm-color) and set the termName resource accordingly in the app-defaults file for the system which has the newer xterm.

However – see above.

My colors changed in vim

Some vim users may notice their colors change after updating to patch 238. Before, some text would display in a dark color using a bold font. Now, it displays in a bright color and normal font.

This is not a bug, but the result of a feature tcap-query which was added for vim in 2000. Several vim users requested that it be enabled by default in the configure script. It allows vim to ask what characters the different function keys actually send, eliminating the chance that the termcap does not match.

Vim also asks how many colors the terminal supports. Since patch 148, xterm has responded with the number of distinct colors that it can display. By default, that is 16 (8 ANSI colors with bright counterparts for displaying PC-style "bold" text).

The interpretation of this depends on the application: termcaps do not tell how to display more than 8 colors. But vim understands how to tell xterm to display using 16 colors. It makes a difference when displaying bright colors. Vim has a table of 16 color names ("dos-colors"), which one can use to define parts of the color scheme. If the terminal supports only 8 colors (colors 0-7), vim uses the bold attribute to simulate colors 8-15.

Changing the color scheme to use bold where it is wanted will make the colors work as before – and work consistently with other terminals.

Aren't bright colors the same as bold?


Actually, "bold" happens to be whatever the terminal shows when it is sent the control-string that says "show bold".

The standard (ANSI aka ISO-6429 or ECMA-48) says no more than that. ANSI specified eight (8) colors. In fact, ANSI did not specify the appearance. That is an implementation detail.

XTerm can be configured to use colors 8-15 for displaying bold text. Or it can be configured to use those colors as part of a 16-color scheme (a feature of aixterm). They use different control strings. When xterm is configured to use the 16-color scheme, it displays bold text by relying on the font to show "bold" (usually thicker characters).

By default, colors 8-15 are brighter versions of colors 0-7 (with some special handling for blue). But again, xterm is configurable and you can use anything that you like for the numbered colors.

Can I set a color by its number?

Well, yes: you can set a color in several ways:

That last (an index) is what some people think of as the color number. The short answer is that you can find on the web tables of colors and match them up to the “color number”. But the number itself has no meaning.

In my reply to tput setaf color table? How to determine color codes?, I noted

You may find this question/answer helpful as well: RGB values of the colors in the Ansi extended colors index (17-255)

although both question and answer raise additional questions. This FAQ is the logical place to answer those questions.

Presumably you are reading this to better understand how xterm works. But you may be interested in the way in which other terminals emulate xterm. If so, this explanation may help as well.

The long answer is that the correct mapping depends on the terminal — other terminals do not necessarily match xterm.

From a shell script, you might use tput with a parameter to an escape sequence referred to as setaf in the terminal description. tput attaches no particular meaning to the number. That actually depends upon the particular terminal emulator.

A while back, ANSI defined codes for 8 colors, and there were two schemes for numbering those. The two are seen in some terminal descriptions as the pairs setf/setb or setaf/setab. Since the latter has the connotation of "ANSI colors", you will see that used more often. The former (setf/setb) switched the order for red/blue as noted in Why are red/blue interchanged?, but in either case, the scheme was established for just numbering the colors. There is no predefined relationship between those numbers and RGB content.

For specific terminal emulators, there are predefined color palettes which can be enumerated easily enough — and can be programmed using these escape sequences. There are no relevant standards, and you will see differences between terminal emulators, as noted in I don't like that shade of blue.

However, convention is often confused with standards. Because xterm has been around a while, it is regarded as a standard by some.

XTerm had color support before I began working on it at the end of 1995. Some of this was mentioned in XFree86's changelog:

XFree86 3.1.2Be (10 January 1996)
203. Major xterm cleanup (including prototyping), and fixes to the colour
     code (Thomas E. Dickey).
XFree86 3.1.2a (23 September 1995)
 14. Colour support for xterm (David Wexelblat).
 13. Fix usage of $LINES and $COLUMNS by xterm on SVR4 (David Wexelblat).

and some was not:

Thus, from the start there were two types of color support in xterm. ANSI colors treats the available colors as an array (its palette) which can be programmed, while dynamic colors applies a single color to a feature.

There have been some changes since the color_xterm in 1992:

Resource 1992 1995 2016
color0 Black black black
color1 Red red3 red3
color2 Green green3 green3
color3 Yellow yellow3 yellow3
color4 Blue blue3 blue2
color5 Magenta magenta3 magenta3
color6 Cyan cyan3 cyan3
color7 White gray90 gray90
color8   gray30 gray50
color9   red red
color10   green green
color11   yellow yellow
color12   blue rgb:5c/5c/ff
color13   magenta magenta
color14   cyan cyan
color15   white white
colorUL   yellow foreground
colorBD   white foreground
colorRV     foreground
colorIT     foreground

In development of xterm over the past 20 years, we

Much of that has been adopted by other developers for different terminal emulators. That is summarized in Why not make "xterm" equated to "xterm-256color"?.

As hinted by the table, the 16-color extension was partly implemented in xterm by late 1995, using the scheme of Linux console: bold fonts are shown as brighter equivalents of the ANSI 8 colors. Unlike the Linux console, xterm can use bold fonts and (aside from providing similar appearance to the Linux console for programs such as dialog) there was no reason to pretend that bold and bright were synonymous.

The colorUL and colorBD features are part of this discussion because I incorporated those into the indexing scheme for colors. More on that later.

First, deal with the 256- and 88-color extensions.

The reason for 256 colors is that the index would fit in a byte. Larason's scheme was simple enough:

The xterm source-code includes scripts for demonstrating the colors, e.g., using the same escape sequences that tput would use:

I added the scripts in patch #94 because of some user comments that there were scripts of that sort available, that there were some deficiencies in those, and and it would be nice to have some good examples in xterm's source. Coincidentally, that gave Todd Larason and Stephen P Wall a starting point for the changes to support 256- and 88-colors.

The 256-color extension came first. 88-colors (using the same control sequence) came next, to reduce the amount of memory needed. XTerm stores both foreground and background color indexes for each cell on the screen. That is two bytes, which doubled the amount of memory used by xterm for the scrollback. More important, however, was the number of entries in the colormap. With 256 colors, 65536 entries might be used, but 88 colors use at most 7744 entries. In the late 1990s, inexpensive displays were far less capable, requiring workarounds to get acceptable performance.

The 256- and 88-color schemes (a 16-color table of ANSI (or aixterm) colors, followed by a cube and then a grayscale “ramp”) are similar. An intermediate 157-color scheme could have been provided,

88 157 256 Usage
16 16 16 ANSI/AIX
64 125 216 cube
8 16 24 grayscale "ramp"

but the savings in the colormap would be less compelling:

Colors Size Bits
88 7744 13
157 24649 15
256 65536 16

Like the aixterm 16-color extension, these colors are stored in an array. Unlike aixterm (whose developers invented a new set of escape sequences not found in ANSI or ECMA-48), we used sequences found in ECMA-48: SGR codes 38 and 48. However, the feature evolved:

As others incorporated the xterm 256-color feature, the ability to set the palette was usually not done before announcing that a program had the 256-color feature. Others acquired the ability to set the palette after a lapse of years. As an exception, Geoff Wing (rxvt developer) implemented the complete feature in August 2002 (release 2.7.9). Any xterm-compatible implementation with support for 256-colors automatically supports 88-colors, since the palette is modifiable, which makes comments such as this at best badly informed.

A few non-xterm applications may support the feature, e.g.,

though the results may not be satisfactory. Here are screenshots for 88- and 256-colors which I made in February 2014 while discussing deficiencies of the FreeBSD console emulator:

teken with 88-colors

teken with 256-colors

XTerm stores the colors for colorUL, etc., at the end of the color array used for ANSI, 16-, 88- and 256-colors. An application can modify the colors using OSC 4, which does not reduce the range available for the SGR 38/48 index used for selecting colors (underline, bold, reverse — and italics — all have their place in the video attribute fields). Like dynamic colors, this was a feature found in XFree86 but not in X11R5 or X11R6. According to David Dawes, some people liked the feature. Mark J Olesen incorporated the same into rxvt mid-1996, and I added the other two attributes. However, it was mainly popular with Red Hat users who wanted to color their manpages. After Werner Lemberg changed groff behavior in 2001 to color manpages, this feature is not that well known.

Finally, there are the default foreground and background colors set using SGR 39/49.

If one wants to enumerate the colors which can be set by index in xterm, there are multiple indices that are needed:

The sample scripts in xterm's sources demonstrate these features. Some are written in POSIX shell, the remainder are in Perl.

I don't like that shade of blue

Nobody does. But there are no universal solutions.

If your terminal (or the application running in it has a dark background, then darker blues are hard to see. With a light background, yellows are hard to see.

The available standards do not help: there are no standards for terminal colors. Here is an illustration which I made in reply to a bug report, contrasting different choices for blue, against some of the other terminals which (were said to) provide "standard vt100 colors":

Contrasting blue in terminal emulators

Of course, anyone developing a terminal emulator already knew that vt100's never did do colors.

Ultimately it is up to the application running in a terminal to enforce the colors it needs. XTerm merely provides the best compromise on default visibility that I and my users have found.

Why doesn't xterm support italics?

Well, actually it does and it doesn't.

You can display "any" font using xterm (though proportional fonts may be disappointing).

But xterm has specific types of graphic rendition that it will do. If you want italics, then xterm has an option (italicULMode) to use that rendition instead of underlining. That is the usual typographic alternative, though of course some people want both at the same time.

However, standard curses does not support italics. Few terminals do this reliably, so it was disregarded long ago, never was supported except for low-level applications (in terminfo). No bit was reserved in the curses header for adding italics for high-level applications. (As a special case, ncurses was modified to provide partial support, but programs using this feature will not work with other implementations).

XTerm stores each cell of the display in fixed-size structures. One byte stores the graphic rendition. XTerm is using all of the bits in this byte for its VT220 emulation:

Mnemonic Bit Description
INVERSE 0 show cell reverse-video
UNDERLINE 1 show cell underlined
BOLD 2 show cell as bold
BLINK 3 show cell as blinking
BG_COLOR 4 use background color
FG_COLOR 5 use foreground color
PROTECTED 6 character cannot be erased
CHARDRAWN 7 character has been drawn here on the screen

While additional bytes could be added to each cell, the cost to the typical user has so far not been in line with the usefulness of the feature.

For those who are not constrained by cost, since patch #305 xterm provides an experimental compile-time option to support italics. The main reason for implementing this is to be able to test the italics feature added in ncurses (patch 5.9.20130831):

The increase in size is not entirely wasted. The SGR attributes for dim, strike-out, and double-underscore also are implemented. However, the last two are not in the portable terminfo definition (from X/Open), and are not supported in the higher-level curses interface (there is no A_STRIKE for that reason).

Here are screenshots showing the ncurses test-program displaying video attributes (including italics). The first uses bitmap fonts:

ncurses – video attributes with bitmap-font

and the second uses a (same size) TrueType font:

ncurses – video attributes with TrueType font

"grep --color" does not show the right output

GNU grep (version 2.5) introduced a --color option.

It does this for each highlighted match:

  1. it writes the text up to (not including the match)
  2. it writes an ANSI color control control sequence
  3. it writes the matched text
  4. it writes a control sequence to clear to the end of the line
  5. it writes an ANSI control sequence to reset graphic rendition.
  6. repeat this process until the entire line is written.

One problem is in the second and fourth steps. If the preceding text brought us up to the last column, then xterm (and any VT100-compatible terminal) is waiting for graphic text to wrap to the next line. Any controls would take effect on the current column position. Newlines are ignored while in this state.

However, if xterm gets a control sequence while waiting to wrap to the next line, it will update the screen according to that control. Then it is ready to accept more data. But at this point, it is no longer waiting to wrap; the special case is for newline versus graphic characters. For instance, backspacing clears the state (vttest illustrates this). So the data starts to write at the current column (the last one on the line), rather than at the beginning of the next line. In that case, grep's output will not look right.

Here are some relevant bug reports:

That description of wrapping is odd, say more?

This is one of the aspects of the so-called "vt100 glitch", as mentioned in the terminfo manpage:

Terminals which ignore a line-feed immediately after an am wrap, such as the Concept and vt100, should indicate xenl.

When the terminal reaches the right margin, it is in a special state where it ignores tab characters and other formatting controls (carriage return and newline), and in effect is expecting only printable characters to wrap to the next line.

Without it, it is misleading to refer to a terminal as a vt100 emulator. After all, it is a well-known feature named for the VT100. The applicable standards (ISO-6429, ECMA-48) do not go into enough detail to address this sort of behavior, so the other terminal emulators can be referred to most accurately as ANSI terminals (if they obey the other guidelines).

In 2004, I added a test-screen to vttest to demonstrate this. It was in response to someone who insisted that xterm was wrong and one of those other terminal emulators was "right". I investigated, found that the behavior had not changed in xterm at least since the early 1990s, and that it matched the description of behavior from the DEC manuals. One of my users verified the correctness of the test on a VT520.

Reviewing the results with xterm-alikes or less ambitious "vt100 emulators" in mid-2013:

In the vttest page, I have provided screenshots to illustrate these points.

Since 2013, Mattias Engdegård created a test program to explore this area, citing DEC's internal standard document for terminals (DEC STD 070 Video Systems Reference Manual). That document refers to this as the last column flag.

Revisiting this in 2019, no improvement has been observed in the problematic programs from 2013. A VTE user gave this example

echo -e "\e[2J"
echo -e "\e[1;79Hx\b\vx\b\vx\n"
echo -e "\e[5;80Hx\b\vx\b\vx\n"

In this case, DEC's documentation for the last column flag mentions that cursor-positioning resets the flag. Because xterm takes that into account (while the others from 2013 do not), the example will show different results. Markus Schmidt provided a screenshot which demonstrates that DEC's documentation is correct and that some terminal emulators (e.g., xterm, zoc, MacOS Terminal, st) implement this detail in the same way that the hardware terminal did:

screenshot of last column flag with cursor positioning

That color scheme is odd, say more?

Occasionally someone questions the behavior of the bce (background color erase) feature in xterm, and mentions that some DEC terminal did not behave that way with ANSI colors.

First off:

The design used for xterm imitated Linux console, which itself came about from different people (see this page for some background).

The VT525 programmer's reference manual is vague on the details (ANSI color is mentioned in a fraction of one percent of the manual), but the DEC standard for terminals is clear that it would not implement bce: any erase command will reset the video attributes. It documents ANSI color in the section on video attributes without mentioning a special case. Color would be reset as well.

Odd behavior

Why can't I select/paste in 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 select/paste by pressing the shift key while clicking with the mouse.

This is all done using the translations resource (see the Default Key Bindings section in the manual page).

Why can't I select/paste to/from other programs?

Whether you select text in xterm and paste into another window, or the reverse, the X client in which you have selected text may provide the data in different formats and different containers:


Originally (and by default) xterm made the selected data available with ISO-8859-1 encoding (Latin-1). Since patch #101 (1999), it has provided it also in UTF-8.

Regarding the type of data:

  • X11R4's ICCM documented "string" selection data with ISO-8859-1, while
  • X11R6 documented "compound text" (another name for multibyte encoding, without specifying what encoding).
  • Selection data using UTF-8 was an extension by XFree86.

The client holding the selection advertises the formats that it can provide, and other client(s) ask for it using one of those formats.

Xterm can ask for UTF-8 even if it is not configured to use UTF-8. In that case, it converts (a small number of) useful characters to their ASCII or VT100 line-graphics equivalents, and uses a "#" character for those which cannot be converted.


By default, xterm follows the Inter-Client Communication Conventions Manual (ICCM). That dates back to X11R4 in 1989, with minor updates in 1996 for X11R6. The copyright for ICCM 1.0 is 1988/1989, making it slightly older than Microsoft Windows.

The ICCM specifies "selection atoms" which are maintained by the X server. According to the ICCM:

The selection named by the atom PRIMARY is used for all commands that take only a single argument and is the principal means of communication between clients that use the selection mechanism.


The selection named by the atom CLIPBOARD is used to hold data that is being transferred between clients, that is, data that usually is being cut or copied, and then pasted.

xterm uses PRIMARY by default. The default translations also update something called CUT_BUFFER0 (also part of the ICCM).

Unlike the PRIMARY selection, a cut buffer can hold only "type STRING and format 8" (which happens to be ISO-8859-1). That sounds like a drawback, but on the other hand, cut buffers are persistent, while the PRIMARY selection is not. An X client can provide data using the PRIMARY selection only as long as it holds the selection.

If xterm does not own the selection, it cannot supply the data (and you cannot select/paste). Initially, xterm held the PRIMARY selection only as long as the text was highlighted. Another application could assert the selection, but generally losing the PRIMARY selection in xterm was the same as losing highlighting. That has been improved, e.g., using the keepSelection resource in patch #230 (2007), as well as refinements to retain highlighting when it updates other parts of the window.

A more likely reason for failing to select/paste is that the other application may not use the same selection atom (container). In the mid-1990s, Netscape set out to compete with Internet Explorer. Part of that involved copying many aspects of the way Internet Explorer worked, including the way it worked with the Microsoft Windows clipboard. Netscape on non-Windows platforms, "of course" assumed the clipboard was the way to do things, and used the X11 clipboard rather following the ICCM. (The way it used the X11 clipboard was also not in line with the ICCM, but it was "close").

Not all applications followed Netscape and its descendents, making it a nuisance if one wanted to select/paste text to/from the web browser.

Since patch #209 (2006), xterm has provided a workaround: a menu entry (and resource selectToClipboard) which changes xterm's behavior for a special token SELECT in its default translations. If the resource is true (or the menu item enabled), xterm provides its selection to the CLIPBOARD. A menu item is provided, of course, since many applications follow the ICCM. In the default translations, these lines use SELECT:

Shift <KeyPress> Select:select-cursor-start() \
                           select-cursor-end(SELECT, CUT_BUFFER0) \n\
Shift <KeyPress> Insert:insert-selection(SELECT, CUT_BUFFER0) \n\

Why can't I select tabs in xterm?

This issue was noted early on, here in 1997.

XTerm is copying from the screen, which stores only printable characters. That includes spaces and line-drawing characters. But tabs are special; they are used for more than one purpose.

If the screen is cleared in some part, that stores nulls. Cursor addressing does not fill in nulls as it jumps around, though xterm does supply blanks for the most useful cases, especially when getting data for a selection.

Full-screen programs such as text-editors tend to write in random fashion, and generally do not print nulls to the screen. Curses on the other hand, may supply tabs where you thought there were none. Also, the terminal driver can expand tabs (and often is set to do this by default).

So the whole thing is unreliable: unless you make special arrangements for each of the programs running inside xterm, you would often get a tab when you expect, and vice versa.

For the special case where your expectations would match the available data, it is solvable. There are basically two ways it could be done:

As of 2010, a few other terminals did implement this feature. But the reason that it's been low-priority is that it's of very limited usefulness when copying between terminal sessions (and for that matter, from other clients).

Can bracketed-paste solve my problems?

That depends. Bracketed-paste only gives an application a clue that the input from the keyboard is being pasted rather than typed. It was intended to help with the autoindent feature in text editors. But not all programs recognize the feature (see this page for more information).

FVWM does weird things when I try to resize xterm

I have an old (3.1.2G) bug report for xterm which may be related to the second (3.9s) problem:

I have not observed the first, but have reproduced the second.

Why doesn't the screen clear when running vi?

This refers to the "alternate screen" feature, which has been used in its termcap file since 1988. On various systems, this feature may have been removed, although it has always been in the xterm sources.

The feature is controllable (it can be enabled or disabled). However, as it was originally conceived, that ability to control it applies only to programs using termcap.

Under SunOS 4.x, the termcap description for xterm embeds in the ti and te capabilities a command to switch to xterm's alternate screen (e.g., while running vi), and return to the normal screen on exit. This has the effect of clearing the screen. The corresponding terminfo symbols for ti and te are smcup and rmcup, respectively.

Beginning with Solaris 2.x, the terminfo description did not use the alternate screen (it is a matter of preference after all), so that the text from vi remains on the screen after exit. Sun patched the X11R5 terminfo description to omit the smcup and rmcup capabilities. However, Sun began distributing modern xterm on the freeware companion (a CDROM) beginning with Solaris 8. In Solaris 10 for instance, the ncurses 5.6 package provided a usable terminal description for xterm which uses the alternate screen. Solaris 11 distributes modern xterm (though perhaps oddly) using an old—unpatched—terminal description.

Because it is in the terminal description, the feature is configurable...

For example (from Bjorn Helgaas <>) this procedure adds these capabilities to the "xterm" terminfo definition on HP-UX 10.20:

cp /usr/lib/terminfo/x/xterm /usr/lib/terminfo/x/xterm.orig
untic xterm > /tmp/xterm.src
echo " smcup=\E7\E[?47h, rmcup=\E[2J\E[?47l\E8," >> /tmp/xterm.src
tic /tmp/xterm.src

In this example, the terminfo strings are a series of operations:

\E7 saves the cursor's position
\E[?47h switches to the alternate screen
\E[2J clears the screen (assumed to be the alternate screen)
\E[?47l switches back to the normal screen
\E8 restores the cursor's position.

However, xterms that are linked with termcap are more flexible in this area than those linked with terminfo libraries. The xterm program supports a resource titeInhibit which manipulates the $TERMCAP variable to accomplish this. It sets the $TERMCAP variable for the client with the ti and te capabilities suppressed. Systems that use terminfo cannot do this. If you are running terminfo with the alternate screen controls in the terminal description, then you can suppress the switching to the alternate screen by the titeInhibit, but not the associated cursor save/restore and clear-screen operations.

XFree86 3.9s xterm implemented a different set of controls (private setmodes 1047, 1048 and 1049) which address this (in addition to the older set of controls, for compatibility). The new set of controls implements the entire ti sequence (save cursor, switch to alternate screen, clear screen) and te (switch to normal screen, restore cursor) as two control sequences that can be disabled by titeInhibit.

The 1049 code is a refinement of 1047 and 1048, clearing the alternate screen before switching to it rather than after switching back to the normal screen. Since patch #90 in 1998 xterm allows you (with a popup menu entry designed to exploit this behavior) to switch the display back to the alternate screen to select text from it, to paste into the normal screen. You can also set or clear the titeInhibit resource using another popup menu entry (Enable Alternate Screen Switching).

Most other terminal emulators implement only half of the feature. They recognize the control sequence, but do not provide the ability to change it at runtime, e.g., using a menu entry. Like any other half-done implementation, that is a bug which should be reported to the developers of those programs.

Why doesn't the screen clear when I type control/L?

Control/L is ASCII form-feed. Printers do something with form-feed. Terminals do not, as a rule (though I agree it would be nice, e.g., this).

Interpreting form-feed is normally done by your shell, not by the terminal emulator. In a quick check:

VT100s did not respond to form-feed. A few terminal emulators interpret form-feed (PuTTY and SunOS console), but neither matches VT100 behavior.

Because most people do not see the difference between a form-feed which they type (and is presumably echoed as a form-feed) versus a form-feed which is sent from an application to the terminal, this leads to confusion. Several years ago, I pointed this out as one of the errors in the C FAQ (notwithstanding Summit's comment, he did not update the FAQ).

Why is the cursor misplaced after running vi?

Vi and other full-screen applications use the termcap ti/te (terminfo smcup/rmcup) strings to initiate and end cursor addressing mode. As mentioned in the discussion of titeInhibit, full-screen applications can expect the initialization string to save the cursor's position, and the end-string to restore it.

A few applications (reportedly IRIX 5.x and 6.x vi incorrectly move the cursor before initializing cursor-addressing. This will cause the end-string to restore the cursor to its position when it was saved by the initialization string (typically at the upper left corner of the screen).

The usual reason is due to the cursor save/restore controls in the ti/te strings. If your application runs a subprocess which in turn runs another full-screen application (or when reinitializing the screen after the shell process), it will save the cursor position again, so the position which is restored when finally exiting your program is the last one saved, not the first. Modern xterm (from late 1998, patch 90) changes the behavior of the cursor save/restore operations so they apply only to the current screen. That makes it less likely to misplace your cursor.

Why doesn't the scrollbar work?

Originally xterm was built using imake rather than a configure script. One feature of imake that is not possible to guess within the configure script is the wide-prototype compile-time definition NARROWPROTO. When this is not set properly, the Athena widget scrollbars do not work properly. xterm's configure script has a fallback case which allows disabling imake. However, this is moot with the Xorg "modular" build, whose compiler options are unrelated to imake or older versions of any libraries that it may distribute. In this case, the configure script needs some help. Use this option to enable or disable NARROW proto (and disable imake with the --disable-imake option) to match the whims of Xorg hackers.

For instance

configure --disable-imake --disable-narrowproto

Can I improve the scrollbars?

Is that a problem with the appearance, or the way they work?

The appearance can be modified (though few do this) by linking with one of the variants of the Athena widget set (Xaw).

To illustrate, here are a few screenshots:

Xaw (default)

xterm – default scrollbar with Xaw


xterm – scrollbar with XawPlus


xterm – scrollbar with Xaw3d


xterm – scrollbar with neXtaw

Those variants use the same calling interface, so supporting them is simple. Adapting to other toolkits would be much more difficult. For instance (see the discussion of mxterm), replacing the scrollbars may require replacing other parts from the library to get consistent initialization and operation. In the case of Motif, it had nothing like the Athena widget set's popup menus.

Can I improve the scrolling speed?

Several years ago (before 2010) there was a webpage which gave its author's notion of what constituted a “good” terminal emulator: cat'ing (sending) a large file to the terminal would complete in minimal time. Apparently that was the sole interest. Interestingly, its author stated that xterm was the slowest although the presented data do not show this. Also, although the page says “Linux” some of the data are for programs running on Windows. The page spawned a few imitators (with no better methodology), none was systematic, none did any analysis.

Of course, developers do not do that in practice. The terminal is useful for interactive tasks. Compiling is best done by redirecting the build messages to a log file or using a batch process. End users have a different outlook.

There is more than one factor involved in scrolling speed. Here are a few:

Scrolling speed is only one aspect of terminal speed, but it is easy to measure. Other aspects (such as the speed with which an application can change color, move the cursor around the screen, write text in various places) can also be measured. But comparing terminals based on that speed can be misleading. When the terminal drops updates to keep up with an application's speed, the result may be unnoticeable (if the application is fast enough), or it may not.

For example, running the dots program from the ncurses-examples shows some interesting misbehavior with gnome-terminal and konsole: both “choke” at times for a few seconds. The dots program prints colored cells randomly around the screen, pausing briefly 1% of the time. However when dots is terminated, it prints the program's notion of the output rate. In spite of the pauses, the program saw a fairly good rate of output. Some terminal emulators cannot keep up with dots; one possible explanation for the discrepancy is that the terminal emulator discards output (as in the special case of scrolling).

Seeing that raised the question of what variation to expect from different terminal emulators, to point out which might discard output to achieve fast scrolling speeds. A simple script showing the elapsed time to send ncurses's terminfo.src (1.1Mb) a given number of times to the terminal was used. Here is a table illustrating the differences, using the available terminal emulators for Fedora 26 and Ubuntu 17 in November 2017:

Mode Terminal Fedora Ubuntu
1 10 99 1 10 99
Remote gnome-terminal (1) (1) (1) (1) (1) (1)
konsole 0.12 2.10 23.2 0.26 2.65 25.7
mlterm (2) (2) (2) 0.30 3.07 30.4
pterm / putty 0.15 1.42 14.6 0.55 5.66 56.2
rxvt (3) 0.25 2.97 29.5 0.23 3.03 29.5
st / stterm (4) 0.07 0.50 4.40 0.15 1.42 14.4
terminology 0.10 1.00 10.1 0.19 2.01 19.0
urxvt 0.05 0.38 3.24 0.17 1.60 15.7
xterm 0.31 3.50 34.8 0.47 4.41 44.1
xterm + fastScroll 0.09 0.82 8.36 0.39 2.43 22.9
Local gnome-terminal 0.12 1.16 11.4 0.29 3.14 30.6
konsole 0.11 0.82 7.97 0.22 2.17 20.1
mlterm (2) (2) (2) 1.01 7.59 105.
pterm / putty 0.17 1.52 14.6 (5) (5) (5)
rxvt 1.23 11.9 118. 1.75 16.9 166.
st / stterm (4) 0.08 0.61 5.10 0.21 1.63 15.9
terminology 0.09 1.03 10.1 0.43 1.64 16.0
urxvt 0.07 0.53 4.52 0.26 2.41 23.7
xterm 1.77 18.5 178. 2.70 26.5 259.
xterm + fastScroll 0.12 0.96 9.92 0.25 2.36 22.9


  1. On both systems, gnome-terminal failed to connect remotely.

  2. Fedora does not have mlterm.

  3. On Ubuntu, the urxvt package hijacks the name “rxvt”, so the “rxvt” actually tested was rxvt-xpm from the rxvt 2.7.10 package.

  4. Fedora has st 0.70, while Ubuntu has version 0.60, which is a couple of years older.

  5. Running locally on Ubuntu, pterm 0.70-1 dumped core.

Regarding the selection of terminal emulators:

Interestingly, performance is better running remotely. In the test, the machines are not identical:

Possibly displaying on the virtual machines does not perform as well as via XQuartz. But that is a lot of difference to explain. More likely, the local X server is performing badly on some calls.

Using the fastScroll feature made xterm performance comparable to the “desktop” applications. But as usual, with performance data, your mileage may vary.

Why can't my program read the window title?

The longstanding control sequence for reading the window title is something that can be abused in special conditions. For novice (unknowledgable) users, this can be a problem.

XTerm provides resource-settings and menu entries to allow this and related features to be enabled or disabled. See for example allowWindowOps The default resource settings in xterm can be overridden by a packager. However, a knowledgable user can override those default settings.

It is also possible that an overzealous packager may have crippled xterm by removing the functionality altogether. (That should be reported as a bug, to me).

For instance, one of those sent me a "security fix" some years ago, which deleted most of the control sequences which return data to the host. It broke the resize program, and selection, among other uses considered to be benign. In contrast, the same features used in other terminal emulators are tolerated by the same people, so rather than being a misguided attempt at fixing security issues, patches such as that appear to be an attempt at harassment.

Why can't my program set the window size?

Some overzealous packagers, perhaps influenced by the demonstration I provided, are protecting you against the possibility of your xterm becoming inaccessible. (That's unlikely...).

You should be able to override it, as noted above via resource settings or menu entry ("Allow Window Ops").

Why is the text in the wrong place?

Are you using Ubuntu? This was a frequently-reported problem for Ubuntu users. With other systems, it could occur (as of September 2012), but is less frequent. But it was an issue with Ubuntu since 2008.

There ware several related symptoms, e.g.,

Here are some of the corresponding bug reports:

Since the problem was not in xterm, all I could do is to help forward those bug-reports to whatever package owns compiz. What these had in common is that someone wrote code which was tested against only a small subset of the X protocol.

Looking for solutions (since compiz was not being fixed), it was possible to disable compiz. The means for doing this varied with time. Aside from pointing to the root cause of the problem, there was little advice that was useful.

Ubuntu dropped compiz in 2017. Its replacement (GNOME shell) has fewer bugs.

Sample .Xdefaults Color-Settings for XTerm

This example dates from March 1997:

XTerm*internalBorder:  10
XTerm*highlightSelection:  true
XTerm*VT100.colorBDMode:  on
XTerm*VT100.colorBD:  blue
XTerm*VT100.colorULMode:  on
XTerm*VT100.colorUL:  magenta
XTerm*VT100.eightBitInput:  true
XTerm*VT100.eightBitOutput:  true
XTerm*scrollBar:  true
XTerm*VT100.titeInhibit:  true
XTerm*VT100.colorMode:  on
XTerm*VT100.dynamicColors:  on
! Uncomment this to use color for underline attribute
XTerm*VT100.colorULMode:  on
XTerm*VT100.underLine:  off
! Uncomment this to use color for the bold attribute
XTerm*VT100.colorBDMode: on
XTerm*VT100.color0: black
XTerm*VT100.color1: red3
XTerm*VT100.color2: green3
XTerm*VT100.color3: yellow3
XTerm*VT100.color4: blue3
XTerm*VT100.color5: magenta3
XTerm*VT100.color6: cyan3
XTerm*VT100.color7: gray90
XTerm*VT100.color8: gray30
XTerm*VT100.color9: red
XTerm*VT100.color10: green
XTerm*VT100.color11: yellow
XTerm*VT100.color12: blue
XTerm*VT100.color13: magenta
XTerm*VT100.color14: cyan
XTerm*VT100.color15: white
XTerm*VT100.colorUL: yellow
XTerm*VT100.colorBD: white
XTerm*VT100.cursorColor: lime green

XTerm comes with two copies of each resource file, one with color only (, which is installed as XTerm-color), and the regular one (, installed as XTerm). To use the XTerm-color file in conjunction with a separate XTerm app-defaults file which does not contain color, add the following line to your .Xdefaults file:

*customization: -color

Since 1997, the resource files grew in size and number. Besides XTerm and XTerm-color, there are also resource files for xterm using different class values, together with the -color flavors of these. Because the -color flavors differ only by an #include statement, the makefile generates these from Here are the others:

Program Resource
xterm XTerm
uxterm UXTerm
koi8rxterm KOI8RXTerm

Besides just adding files, I continued testing more resource combinations. Originally (in the 1990s for instance), developers could reasonably expect their users to configure resources for themselves, rather than use a single prepackaged flavor. That was a while ago. After 2000, I developed nicer resource files. Rather than modify the installed app-defaults file, I use this feature from X:

Directories named by the environment variable XUSERFILESEARCHPATH or the environment variable XAPPLRESDIR (which names a single directory and should end with a ‘/’ on POSIX systems), plus directories in a standard place (usually under /usr/share/X11/, but this can be overridden with the XFILESEARCHPATH environment variable) are searched for for application-specific resources. For example, application default resources are usually kept in /usr/share/X11/app-defaults/. See the X Toolkit Intrinsics – C Language Interface manual for details.

That is, if you set the XAPPLRESDIR environment variable to point to a directory, you can put application resource files there, and X will find those before the system app-defaults files. That allows more flexibility and better control over the various applications than putting everything into a single .Xdefaults file.

On the opposite extreme, some people advise using xrdb. Not everyone. Back around 1990 I had an informative conversation with one of the developers at the Software Productivity Consortium. He was a member of a team developing a set of X widgets. The gist of our conversation was that

What is this warning message?

xterm: Error 11, errno 22: permission denied
Actually, any message like this denotes a failure which requires studying the xterm source to determine the exact problem.

You have either found a bug in xterm, or there is something wrong with your computer's configuration, e.g., not enough pty's, incorrect permissions, etc.

The first number is an internal code (defined in error.h in xterm's source), and the second is the system error number (defined in /usr/include/sys/errno.h). The system error number is easier to lookup, but the internal error code tells you where to look in the source.

input method doesn't support my preedit type
Ignore this if you do not know what input method is. Input methods are used to enter composite characters (e.g., umlauts, other types of punctuated characters, East Asian characters, etc). Your computer's libraries support this, but are missing configuration tables, and xterm is warning you.

If the message bothers you (e.g., if you aren't starting xterm from a window manager menu), you can suppress it by setting a resource:


Warning: Actions not found: ignore, "xxx"
The action "xxx" (for example "scroll-back") is specified in a resource file whose translations match widgets that do not support them. For example, this

XTerm*translations:     #override\n\
<Leave>, ~Ctrl ~Meta <Btn2Up>: ignore()\n\
<Key>KP_8: scroll-back(1,line)\n\
<Key>KP_2: scroll-forw(1,line)\n\
<Key>KP_8: scroll-back(1,halfpage)\n\
<Key>KP_2: scroll-forw(1,halfpage)

will produce warnings such as

Warning: Actions not found: ignore, scroll-back, scroll-forw
Warning: Actions not found: ignore, scroll-back, scroll-forw
Warning: Actions not found: ignore, scroll-back, scroll-forw

This is a correct form, assigning the actions to the "VT100" widget.

XTerm*VT100.translations:     #override\n\
<Leave>, ~Ctrl ~Meta <Btn2Up>: ignore()\n\
<Key>KP_8: scroll-back(1,line)\n\
<Key>KP_2: scroll-forw(1,line)\n\
<Key>KP_8: scroll-back(1,halfpage)\n\
<Key>KP_2: scroll-forw(1,halfpage)

Warning: Cannot allocate colormap entry for "xxx"
This comes from the X library. Modern xterm uses the default color map. What this means is that if your X server has insufficient space to store color information for more than one color map, other applications which could use other color maps may conflict with xterm. In practice, that is 256 unique colors on the screen at a time—not enough for a fancy background or an application such as Netscape.

During resource initialization, xterm attempts to allocate an entry from the color map for each color which it might use. If there are not enough free slots in the color map, you will see a "Cannot allocate" message for each color that xterm failed to allocate. Those colors will be rendered in the foreground color, making full-screen color applications such as dialog unreadable.

This problem is alleviated with patch 129, which modified xterm to delay the most color allocation until the colors are first needed. If a color is never needed (xterm allocates 20 colors in this manner), that reduces the number of slots in the color map that are needed. Even with this improvement, xterm must still allocate 4 colors during initialization to determine how to display the cursor. If none of those colors can be allocated, xterm reverts to monochrome.

Known Bugs in XTerm and Look–alikes

These are the known bugs (or limitations) in modern xterm. They are also present in the other versions based on the X Consortium sources (color_xterm, ansi_xterm, kterm).

Note that of the emulators that support color, some do not support bce (back color erase). The bce capability is also called the "new color model", though it has been implemented in the IBM PC for quite a while. Technically, not implementing bce (or allowing the choice between it and its complement) is not a bug, since few hardware terminals (with good reason) implemented this feature.

X11R6.3 XTerm

The X Consortium version of xterm (and versions based on it) has additional bugs not in modern xterm:

(These bugs are also present in the X11R5 version).

Update 2004/04/08:
Complicating this discussion is the "X.Org" xterm (from 2004). That is the XFree86 xterm from XFree86 CVS with all visible "xfree86" strings changed to "X.Org" or "xorg", depending on the use. For example the "xterm-xfree86" terminfo entry becomes "xterm-xorg". The change history for the related CVS for X.Org shows this. Similarly, the release notes for X11R6.7 included my notes for XFree86 4.4.

As of 2009, it was apparent that "X.Org" xterm had died a natural death, since none of the people who created it had any likelihood of maintaining it. Instead, X.Org defers to my version of xterm.

Reviewing in 2014, the major vendors have been using modern xterm (different patch levels) for some time. However, there are documentation problems with AIX, beyond what is noted here:

The other vendors provide documentation which is more up-to-date.

COLOR_XTERM download

This is based on the X Consortium X11R5 source, with the same bugs.

Not exactly a bug, but it does not build on Linux with X11R6.3

ANSI_XTERM download

This is based on the X Consortium source, with the same bugs.

CXTERM download

CXterm stands for "Chinese Xterm". This is based on the X Consortium source.


This is distributed with CDE. It implements more of the DEC VT220 than the X Consortium xterm, and also adds controls to manipulate the window and icon.

EMU 1.3 download

This is not based on the X Consortium source. The authors state that it implements VT220 emulation. It is in need of maintenance, since it builds with some problems to produce an executable that (on Linux and SunOS) does not handle the carriage return and newline translations properly. So I am unable to run vttest on this emulator.

ETERM link

Eterm was based on rxvt, though the appearance differs. The terminal emulation capabilities appear similar, though I am not able to run the full suite of tests in vttest with this emulator (the core dump noted for rxvt, as well as hanging while awaiting response from one or more control sequences). Oddly, it appears that neither Eterm nor rxvt implement CPR (cursor position report). Finally, it reserves F1 (function-key) for a popup menu. This applies to versions of Eterm through 0.9.


Unless specifically mentioned, GNOME Terminal and VTE's issues generally accumulate, with occasional veering off with skin-deep "rewrites". Each sighting provides a new episode.

Starting in 1999 —

GNOME Terminal is developed separately from both xterm and rxvt, and was originally based on the zvt (zterm) widget. Like kvt), it appears to have been developed imitating other terminal emulators (Linux console and xterm) rather than strictly emulating a VT102. The documentation is fragmentary (with a comment suggesting that the author does not know where to find relevant information), and the program fares badly with vttest. Beginning with late 1999, reports indicate that it does not properly parse ANSI control sequences: the vim editor is using xterm's vt220-style "Send Device Attributes" (Secondary DA) control sequence to obtain the terminal emulator's version. That is, it sends


expecting a response such as


for vt100. The bug report indicates that the "c" sent by vim is echoed rather than interpreted by the emulator.

But it suffices for vi.

Moving on to 2001 —

A more recent GNOME Terminal uses the VTE widget. I observed version in late 2001, which mentioned it in the credits (although VTE 0.1's ChangeLog mentions no date before February 2002). It does not implement a complete vt102: it was missing several features which can be demonstrated in vttest). Most of the bugs in the Device Attributes responses remain, but it works a little better with vim. However, there are problems with the alternate screen that show up with vim. Again, these can be demonstrated with vttest (menu 11.6.3 in the 20011130 snapshot).

Moving on to 2002 —

Rather than evolving from zvt, VTE is largely a new work. It does credit zvt in one place. However, its source code uses xterm's source code as a resource, accounting for odd (often incomplete) chunks. Reviewing 0.9.0 (September 2002):

Jumping to 2010 —

Later versions of VTE incorporate more features (and comments, symbol names, etc), from xterm's source. In some instances, the copied features were disabled by Red Hat's package for xterm. Here is a related bug report, for key bindings.

The documentation for GNOME Terminal asserts:

GNOME Terminal emulates the xterm application developed by the X Consortium. In turn, the xterm application emulates the DEC VT102 terminal and also supports the DEC VT220 escape sequences. An escape sequence is a series of characters that starts with the Esc character. GNOME Terminal accepts all of the escape sequences that the VT102 and VT220 terminals use for functions such as to position the cursor and to clear the screen.

That sounds fine, except that it is both inaccurate and misleading:

combining the "X Consortium" and "DEC VT220", for example, since that was done after the demise of said organization.

It emulates a subset of VT100, lacks support for most of the VT220 control sequences (including some used for positioning the cursor) that are not recognized by a VT100.

Even in the subset which it emulates, GNOME Terminal has bugs. Many of these are easy to demonstrate with vttest.

as noted in Xterm Control Sequences, xterm (mostly after "X Consortium") supports control sequences which are not VT100/VT220. GNOME Terminal implements many of these, but not all.

Perhaps that was unintentional – GNOME developers did not appear to document what their program does outside of that remark. However, an inspection of the changelog for libvte (VTE) does show that most of the borrowing from xterm is cited in an oblique manner – not once mentioning XFree86 for example, leaving the impression (as indicated by "X Consortium") that all of the work on xterm was done before development of GNOME Terminal commenced.

Most of this observation was documented between 2000 and 2007. Other than maintenance, development of GNOME Terminal appeared to have paused in 2005. As of 2009, its maintainer was (of the development team), the least knowledgeable about terminal emulation. So there was no progress on the large number of bug reports related to xterm-compatibility.

Revisiting in 2018 —

Regarding documentation, the situation was not as good as reported earlier. The problematic documentation was not even part of the "official" GNOME Terminal, but was an add-on by a Debian developer, adapted from GNOME Terminal's online help. The developer's relationship was mentioned in a Debian bug report:

#127622: ncurses-term: terminfo entry for gnome-terminal swaps Backspace/Delete:

>> "TD" == Thomas Dickey <> writes:

> On Sun, Jan 06, 2002 at 06:13:30PM +0100, Christian Marillat wrote:


>> The upstream author should consider ours Debian changes has official
>> changes ?

> sure - get gnome-terminal's author to make the changes.  (I generally
> don't add customizations to ncurses' terminfo unless I see them incorporated
> intact by more than one other source).

Sorry to say that, but upstream don't care about my patches. I've
forwarded patches since one year, and these patches has never been
included by upstream. Upstream sayd "commited", but I never seen any


The manual page provided as an attachment to GNOME #311565 identifies the author. However, four years later there is still no manpage for GNOME Terminal. GNOME #701691 mentions this in conjunction with GNOME Terminal's incompatible behavior versus other terminals for the “-e” option:

 Christian Persch 2013-06-06 11:10:43 UTC

This works as designed. Note that both -x and -e are deprecated; the only supported way to pass the arguments is after -- like this:

$ gnome-terminal -- emacs file


 Christian Persch 2013-06-06 16:02:54 UTC

There are no docs for the gnome-terminal command line options.

Admittedly, GNOME Terminal has some documentation, in its online help pages. As mentioned, the misleading comments about X Consortium came from that material, which remained for more than ten years before being revised early in 2013. Here are a few links for that process:

The document editors removed this statement

Run any application that is designed to run on VT102, VT220, and xterm terminals

as well as the extended comment about the X Consortium, and replaced it with a less specific statement.

Terminal is a terminal emulator application for accessing a UNIX shell environment which can be used to run programs available on your system.

Terminal supports escape sequences that control cursor position and colors.

The assertion about GNOME-Terminal's support for “any” persisted in its package description in Debian as of 2018:

Whether the revised manual is improved or even helpful is debatable. For instance, it tells the reader how to turn the scrollbar on and off (using a dialog, of course). But for command-line options, it can print only about 45 lines of option names and short (less than 10 words) descriptions for each if one types

gnome-terminal --help-all

Other programs do the equivalent. In a quick check using Debian 8:

Bug #311565 was (writing in 2018) more than four years ago, but still there is no manual page, for either the command-line options or the control sequences which it supports.

Notes on VTE

VTE is used by developers who provide a facade (also referred to as a “skin”) for simple desktop-oriented terminal emulation. This page gives a number of examples with sizes for the skins. The actual program size is far larger in each case, making the size of the skin irrelevant.

For more than ten years, VTE's README file asserted

VTE supports Unicode and character set conversion, as well as emulating any terminal known to the system's terminfo database.

The latter part of that ("emulating any terminal") was incorrect. It did have the ability to work with the standard function-key definitions which can be defined in a terminfo description. That feature was discarded in 2014.

Notes from 2010 —

Some of the function-key logic was adapted from xterm; generally refactoring the xterm source-code to make it appear different. In places however (naming conventions and comments), there was some verbatim copying. The same observation can be made of "character set conversion". None of that is reflected in VTE's git-log.

As an aside, the credits in GNOME Terminal's "About" box also are inaccurate. For several years (according to its change-log), most of the work on VTE (the principal part of the program) was done by Nalin Dahyabhai.

xterm on the other hand, can be told with the tcapFunctionKeys resource setting to use a more complete subset, based on the ncurses extended terminal descriptions. However, terminal descriptions describe only one particular configuration of a terminal. Even xterm's terminfo/termcap descriptions do not cover the (literally) thousands of keyboard combinations which are available via its resource settings.

Outside of function-keys, VTE provided no ability to emulate “any terminal”. A casual glance at its source code revealed the following:

For instance, VTE cannot emulate dtterm, because of differences in color behavior. In fact, VTE does not use any of the termcap data to support its interpretation of color control sequences.

After 2014 —

Until 2014, VTE used a termcap file, with its own reader, presumably under the impression that could be used to describe “any terminal” (although it was fairly well known that terminals could support escape sequences not found in any terminal description). As a separate file, the termcap was a nuisance, whether it was bundled with VTE (and inaccessible to users) or not. The developers tried it both ways.

One recurring problem was that VTE's termcap did not match xterm's function-keys. Even when VTE's developers modified the termcap to match as well as the termcap could, the match was still incomplete. None of the modified keys were correct, since none of those are described by termcap. That meant that a control modifier with a cursor-key or function key was likely to be misread by programs running in VTE.

Finally in 2014, the VTE developers decided to change it. First, one decided to adapt a chunk of source-code from ncurses, perhaps thinking that was the way to get a better reader. That did not work well, and finally they discarded the whole feature, hardcoding the behavior to match xterm's default configuration.

Here are bug reports which give the story:

As a minor improvement, those changes removed some of the code whose origin was cut/paste from xterm. But that does not mean that the VTE developers stopped that practice. For instance, a change in late 2017 here reminded me to check what VTE does when saving/restoring the cursor position. It turns out that it does something similar, because (see #731205 and #741193) the developer studied xterm's source-code and imitated it (see source changes and followup fixes). There's a quirk in the resulting program (it pays attention to send/receive and insert modes, which are unrelated, while also missing the handling of wrap state), but if the developer had read the documentation (DEC's manuals), that detail would be missing. In reviewing the documentation, I noticed a different aspect which might be used to improve xterm, and ultimately appear in VTE (or perhaps not, since it is in an area poorly supported by VTE, i.e., the bug which was reported).

VTE developers do more than copy from xterm, of course. There are other programs (such as Konsole and which get similar treatment. Because they tend to copy from others rather than doing their own solutions, they have not acquired the experience to see why features were added or modified (or removed), just that it is there. For instance:

Other problems with VTE —

These are a few of the interesting bugs found in VTE (or GNOME Terminal) during 2017:

Other uses of VTE —

Because of GNOME Terminal's reputation for excessive code bloat, developers of every other program based on VTE advertise their version as reduced memory usage, faster startup, etc. Here are a few of the available ones:


Of particular note, MGT 1.4.0 announcement claims that it works properly for all of vttest)'s tests. On the positive side, it does do VT52 emulation, but (reading the source code did not help) it apparently does not really do VT220 from vttest's perspective.

HANTERM download

HanTerm stands for "Hangul term" (Korean). This is based on the XFree86 source.


More than just a rewrite of kvt into C++. But there are several incompatibilities between konsole (noted with version 1.0.2 in late 2001) and xterm:

The problems with setmode 1049 were fixed after some time; other issues linger on.

Like GNOME Terminal, konsole's documentation is incomplete and inaccurate. This gem from its handbook illustrates the problem:

After a decade, Konsole is the first rewrite from the ground up. While xterm has definitely been hacked to death (its README begins with the words Abandon All Hope, Ye Who Enter Here), Konsole offers a fresh start using contemporary technologies and understanding of X.

The problem:

KTERM download

KTerm stands for "Kanji term" (Japanese). This is based on the X Consortium source, with the same bugs (though the list of original authors has been removed; the modifications that comprise kterm is relatively small).

There is a variation of xvt (ancestor of rxvt) originally known as kvt bundled with KDE which may be referred to as "kterm", but I do not find it interesting, other than to comment that it was a poor choice of name.


Mlterm is not based on xterm or rxvt source, though it implements many of their features. It does fairly well with vttest, except for some odd misbehavior in operations that save/restore the cursor position.


There are a few variants of this: the xterm bundled with some Motif clients is more common. More interesting, however is one (not Motif), attributed to "Der Mouse".

(mouse@Lightning.McRCIM.McGill.EDU) Available: ( in

I saw only an incomplete version of this while it was advertised in the mid-90's. It is available by email from <mouse@Rodents.Montreal.QC.CA>. or via ftp. This is not a patched version of xterm, though it was apparently written, like rxvt, to emulate vt100's. While it does have some interesting features (such as blinking characters), overall it does not do as well with vttest as the more widely known emulators.


There are several variants on this: xterm adapted for Motif libraries. I have seen none that work properly:


Distributed with Redhat 5.2, it is a repackaging of xterm-sb_right-ansi, to use the Xaw3d widget set. This is based on the X Consortium X11R6 source, with the same bugs.

Starting with Redhat 6.0, nxterm is the XFree86 3.3.6 xterm. Unfortunately Redhat neglected to update their termcap for nxterm to match the program.

RXVT link

Rxvt's manual page states the following unqualified comment:

rxvt, version 2.6.2, is a colour vt102 terminal emulator intended as an xterm(1) replacement for users who do not require features such as Tektronix 4014 emulation and toolkit-style configurability. As a result, rxvt uses much less swap space -- a significant advantage on a machine serving many X sessions.

How much is much less? Perhaps not as much as one would think from reading that. The Tektronix emulation in xterm (which has been optional since late 1997) accounts for about 25kb of the code.

The toolkit-style configurability glibly referenced is the ability to redefine keys on the keyboard without recompiling the program, i.e., the translations resource. It also is the way mouse events and other actions are passed to xterm.

The toolkit-style configurability accounts for about 300kb, which does add up if you happen to be running 50 xterm processes (i.e., about 10Mb).

This comment was topical in December 2001:

Compared with something like GNOME Terminal, which takes 2-3 times, or KDE konsole, which takes 15-20 times as much memory to run, xterm and rxvt memory requirements are indistinguishable to the normal user.

In June 2010, the numbers had changed somewhat. Here is a table showing the total application and library sizes needed for each of the terminal emulators on my development machine. All sizes are in kb (1024 bytes).

program base size total size libraries
aterm 127 10763 45
color_xterm 142 3647 13
Eterm 1 5126 19
fbiterm 6 2424 8
gnome-terminal 292 14587 51
hpterm 146 14386 31
konsole 2 39815 71
kterm 226 4194 17
mlterm 316 6606 27
mrxvt 298 4515 19
multi-aterm 144 2821 7
pterm 405 12817 42
rxvt 2.6.4 108 2725 6
rxvt 2.7.10 152 2829 7
rxvt-unicode 1259 13641 49 211 15274 29
wterm 110 2922 11
xfce4-terminal 148 14059 48
xgterm 953 4602 14
xhpterm 130 2748 6
xiterm 12 3762 16
xterm (everything) 346 5484 24
xterm (minimal) 186 4123 15
xterm-r5 135 4164 11
xterm-r6 140 4169 11

Counting the libraries is appropriate, since some programs such as xiterm and the VTE-based programs are implemented in libraries.

These comments apply to versions of rxvt through 2.21:

A newer version (upgraded to an beta as of 2.6.PRE3, however, since it no longer dumps core in vttest) is reported to fix the ech bug. However, it is less VT100-compatible than the earlier versions such as 2.21b because it does not render reverse video (DECSCNM) properly. All versions do not update the screen frequently enough, making animation ineffective. See vttest, tests 1 and 2.

One longstanding issue with rxvt impacts use of xterm. While rxvt does not use the X Toolkit (and corresponding X resource matching), it does read your .Xdefaults and app-defaults files to extract resource settings. That in itself would not be a problem. However, since rxvt also looks for resources in the XTerm class (a parasitic relationship like setting $TERM to "xterm" based on the presumption that it is a nuisance to install its configuration files), there have been several occasions on which xterm's app-defaults files have been modified to accommodate rxvt's variant usage.

That comment applies mainly to the resource patterns. However, even when the pattern is reasonably unambiguous, but overbroad, the results can be conflicting. For example, some versions of rxvt may accept a font resource which does not match the XLFD pattern. It accepts a prefix of "xft:". This feature (apparently introduced by konsole) tells rxvt to interpret the remainder of the string as a TrueType (Xft) font rather than a bitmap font. xterm uses the faceName resource for these values.

st link

Rxvt revisited, this program originally depended only on the X11 library. Since then, it has grown a lot, though the project page does not mention it. As of January 2013, it was in heavy development, and (according to comments on its developer's list) growing steadily as the developers implemented useful features adapted from xterm.

For instance, in 2013, the size counting libraries for st 0.3 on my Debian testing machine was on a par with rxvt (and half that of xterm, which uses the X Toolkit library). Revisiting it late in 2015, it had left rxvt behind and was nearly as large as xterm. Here are the sizes which I found in Fedora 22:

LDD-Size Program
3452986 rxvt
6060960 st
6771039 mrxvt
7785780 xterm
15060195 urxvt
20934874 lxterminal
21089908 lilyterm
21358156 xfce4-terminal
25738679 roxterm
31195794 gnome-terminal
32780414 terminology
87813125 konsole

By the way, the project page quotes the README file from xterm's sources, omitting my editorial comment at the top noting that the paraphrase of the opening from Dante's Inferno dated from 1991, and pointing to this FAQ to provide better context.


It has some features which are also in color_xterm:(non-bce ANSI color, colorBD and colorUL resources, cursor warping, etc. The main feature is its Tektronix graphics emulation, which is the main reason for this particular program. Neither program has a change-log, so it is not easy to say which influenced the other.

That is from reading the source code. However testing under Debian, something is wrong with the resource processing (neither popup menus nor colors work).

As of March 2022, Debian's xgterm package does not install the terminal description which the IRAF developer provided. That was just an alias for xterm-r5, which still needs some work. If the “xgterm” terminal description is missing, xgterm falls back to “xterm” which is not a close match. To address this problem, ncurses provides a workable xgterm terminal description.


This appears to be rxvt 2.20, lightly reformatted, with a few ifdef's changed.

That is, it was. The name was later appropriated by a different program, which also uses the name iterm. Like gnome-terminal, iterm aims to be an xterm-emulator rather than a VT102- or VT220-emulator.

An earlier attempt by the same author (the "CSI-xterm") incorporated in 2002 some of the changes I made for XFree86 xterm via cut and paste (but does not mention this in its README). The "borrowed" changes comprised about 10% of the patch provided for X11R6.5.1, summarized here:

 Imakefile     |   25 +
 RELNOTES-I18N |  104 ++++++      |    1
 button.c      |  155 ++++++++-
 charproc.c    |  979 ++++++++++++++++++++++++++++++++++++++++++++++++++++------
 data.c        |    6
 data.h        |    4
 error.h       |    8
 fontutils.c   |   78 ++++
 fontutils.h   |    8
 input.c       |   11
 main.c        |   40 +-
 main.h        |    1
 misc.c        |   46 ++
 ptyx.h        |  156 ++++++++-
 screen.c      |  513 +++++++++++++++++++++++++++---
 scrollbar.c   |   36 +-
 util.c        |  218 +++++++++++-
 18 files changed, 2183 insertions(+), 206 deletions(-)

This patch was said to be the basis for Solaris 10 xterm, and was briefly referred to as the Solaris "color xterm". It did not use the bce color model however, and Sun provided no terminal description for it.

Back to iterm: the author's README in the patch used the same terminology as in the later work, demonstrating their relationship:

 This is a patch for the xterm of X11 release 6.5.1 to fix its
internationalization defects. This patch enables xterm to handle
whatever the character set encodings and scripts support underlining
operating system supports via the technology called CSI(Code Set
Independence) and XOM(X Output Method). Traditionally, several
X terminal emulators which are hard-wired to specific languages and
encodings were introduced to support local language requirements, such
as kterm, hanterm, cxterm, UTF-8 xterm and so on. This  truly
internationalized terminal emulator supersedes the needs of those
multiple locale specific terminalemulators.

The key to understanding the "code set independence" is that the author intended to treat existing character encodings on an equal basis with Unicode and UTF-8. Some of that is reflected in the Solaris International Language Environments Guide, but in explaining how this is done, the documentation is weak, lacking detail.

Either version of iterm has similar problems running vttest.

How do I build XTerm?

Building a copy of xterm is simple, provided that you have a development configuration for X11:

If you have a working xmkmf script (or correctly configured imake utility), all you need to do is type


I have written a configure script for xterm which can use imake (or xmkmf) to generate a Makefile from the Or it can do without imake entirely. I have restructured xterm to eliminate most hardcoded #ifdef's, replacing them with definitions that can be derived with the configuration script. The configure script is more flexible than xmkmf, since it allows you to enable or disable a variety of features. Type

configure --help

to get a list of options.

Though I have replaced most hardcoded ifdef's with autoconfigured values, it will still continue to build properly with the imake environment.

However, I usually build xterm using the configure script. By default, it looks for imake and will use it to help with a few places where a reliable configure check cannot be created. One of these (see Why doesn't the scrollbar work?) can be a problem.

As with all of my projects, I routinely check for strict compiler warnings. For gcc, that is done with the "gcc-stricter" script which you can find here. The X libraries have a longstanding issue had been ignored (as of mid-2012). To work around this (and get useful warnings), I applied this patch:

--- Intrinsic.h.orig    2009-08-25 13:22:15.000000000 -0400
+++ Intrinsic.h 2009-12-06 09:48:39.000000000 -0500
@@ -66,7 +66,11 @@

 #define XtSpecificationRelease 6

+typedef const char *String;
 typedef char *String;

 /* We do this in order to get "const" declarations to work right.  We
  * use _XtString instead of String so that C++ applications can
--- Xresource.h.orig    2009-07-19 14:43:21.000000000 -0400
+++ Xresource.h 2009-12-06 10:11:19.000000000 -0500
@@ -338,8 +338,8 @@
 } XrmOptionKind;

 typedef struct {
-    char           *option;        /* Option abbreviation in argv          */
-    char           *specifier;     /* Resource specifier                   */
+    _Xconst char    *option;       /* Option abbreviation in argv          */
+    _Xconst char    *specifier;            /* Resource specifier                   */
     XrmOptionKind   argKind;       /* Which style of option it is          */
     XPointer       value;          /* Value to provide if XrmoptionNoArg   */
 } XrmOptionDescRec, *XrmOptionDescList;

I made note of it on the Xorg mailing list, but as you can see, there was no response.

Finally, I spent the requisitve time to integrate the change—and complete the process of transforming the X Toolkit documentation from nroff to docbook/xml.

How do I report bugs?

You should report bugs to me. I also respond to bug reports in a number of bug-tracking systems, though some are less open to searches than others. See also:

Additional Information

There appears to be no comprehensive source of information on xterm better than the documentation which comes with the source code

The XTerm Manual

The command-line options, X resources and similar configurable options of xterm are documented in the manual page.

Here are copies of the file in various forms: html, pdf, ps and text.

Xterm Control Sequences

Control sequences, i.e., programming information are in the file which I bundle with the program source. (It used to be in the same directory in the X distribution, but was moved to a different part of the tree long ago). Note that you must format this file with different options than a manpage, e.g.,

tbl | nroff -ms >ctlseqs.txt
tbl | groff -ms >

As a PostScript or PDF file, the individual letters of the control sequences are all boxed, for emphasis, but I find the text file equally readable.

Here are copies of the file in various forms: html, pdf, ps and text.

resize – set TERMCAP and terminal settings to current xterm window size

resize is useful by itself, but is maintained for historical reasons as part of xterm. html, pdf, ps and text.

uxterm – a UTF-8 wrapper for xterm

XTerm does not automatically set your locale. It can be told to use your locale settings. This is a shell script which sets xterm's resources to use UTF-8 encoding, and use UTF-8 fonts. There is a similar lxterm script, but it relies upon non-portable applications, unlike uxterm.

Here are copies of uxterm's documentation: html, pdf, ps and text.

Incidentally, there was a different program named "uxterm" before the shell script was added to xterm in mid-2000. Roman Czyborra commented in 1998 that it was based on the original X11 xterm source (very likely, since "strings" run on the executable shows the xterm actions, resources and even the Tek4014 support). There are few references to it to provide details: the first appearance was in 1994, and the last was Czyborra's page in 1998. For the curious, there is a copy on (no Linux executables, no source, however).

koi8rxterm – a KOI8-R wrapper for xterm

As a special case, this wrapper is packaged with xterm to provide KOI8-R encoding.

Here are copies of koi8rxterm's documentation: html, pdf, ps and text.

luit – Locale and ISO 2022 support for Unicode terminals

luit also is maintained as part of xterm, since its upstream maintainer is inactive, and the ostensible maintainers have more than once delivered unusable versions, causing many bug reports to be issued against xterm.

Other Sites

Interesting but misleading:

Ongoing/future work