http://invisible-island.net/xterm/
Copyright © 1997-2018,2019 by Thomas E. Dickey
Here is the latest version of this file.
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".
I've been working on xterm since early 1996 (see my changelog for details).
But the program is much older than that.
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.
There is a git repository here which gives some more of xterm's prehistory.
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.
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.
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 X.org 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.
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.
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:
ESC \.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 1999) 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.
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.
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.
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.
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.
However:
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.
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).
The most recent (and well supported) version of xterm is the one that I maintain:
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
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:
The control sequences document lists a few controls which xterm does not (completely) implement, e.g.,
None of the other terminal emulators implements those
either.
| 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.
| 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.
| 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.
| 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 (as of September 2013, it implements 546 of 556 documented controls); the other programs change far more slowly, adding only one to konsole. 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.
Not really problems, but frequently asked questions (the point of this, after all):
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.
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:
typescript file.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:
printAttributes to get more easily printed
output.printerAutoClose resource
to change xterm's behavior to close the printer pipe whenever
the terminal is told to switch the printer offline.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.
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 <root@candle.pha.pa.us> 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\
~Shift <Key>F1: string(0x1b) string("OP") \n \
~Shift <Key>F2: string(0x1b) string("OQ") \n \
~Shift <Key>F3: string(0x1b) string("OR") \n \
~Shift <Key>F4: string(0x1b) string("OS") \n \
~Shift <Key>F5: string(0x1b) string("[16~") \n \
~Shift <Key>F6: string(0x1b) string("[17~") \n \
~Shift <Key>F7: string(0x1b) string("[18~") \n \
~Shift <Key>F8: string(0x1b) string("[19~") \n \
~Shift <Key>F9: string(0x1b) string("[20~") \n \
~Shift <Key>F10: string(0x1b) string("[21~") \n \
~Shift <Key>F11: string(0x1b) string("[28~") \n \
~Shift <Key>F12: string(0x1b) string("[29~") \n \
Shift <Key>F1: string(0x1b) string("[23~") \n \
Shift <Key>F2: string(0x1b) string("[24~") \n \
Shift <Key>F3: string(0x1b) string("[25~") \n \
Shift <Key>F4: string(0x1b) string("[26~") \n \
Shift <Key>F5: string(0x1b) string("[K~") \n \
Shift <Key>F6: string(0x1b) string("[31~") \n \
Shift <Key>F7: string(0x1b) string("[31~") \n \
Shift <Key>F8: string(0x1b) string("[32~") \n \
Shift <Key>F9: string(0x1b) string("[33~") \n \
Shift <Key>F10: string(0x1b) string("[34~") \n \
Shift <Key>F11: string(0x1b) string("[28~") \n \
Shift <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 <ress@spd.dsccc.com>:
#!/bin/sh
# 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
Usage(){
echo
echo " Usage : vax -options"
echo
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
echo " If you need additional help, please call Workstation"
echo " Services at x92396."
echo
exit 1
}
# Default to a black foreground with a white background.
# Use the 9x15 and 9x15bold fonts. Connect to castor by default.
#
FG=black
BG=white
HOST=castor
FONT=9x15
BFONT=9x15bold
COLS=80
# Parse the command line arguments
#
while [ $# != 0 ];
do
case $1 in
-80) COLS=80
FONT=spc12x24c
BFONT=spc12x24b
shift
;;
-132) COLS=132
FONT=9x15
BFONT=9x15b
shift
;;
-fg) shift
FG=$1
shift;;
-bg) shift
BG=$1
shift;;
-fn) shift
FONT=$1
shift;;
-fb) shift
BFONT=$1
shift;;
-host) shift
HOST=$1
shift;;
-help) Usage;;
*) Usage;;
esac
done
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, <oahlefel@metz.une.edu.au>. 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:
#!/bin/sh
# 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
Usage(){
echo
echo " Usage : vmsterm -options"
echo
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 [crusher.saltmine.com] [earth] [192.168.7.7]"
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 "
echo
exit 1
}
# Default to a black foreground with a white background.
# Use the 9x15 and 9x15bold fonts. Connect to 192.168.3.3 by default.
#
FG=black
BG=white
HOST=192.168.3.3
FONT=9x15
BFONT=9x15bold
COLS=80
# Parse the command line arguments
#
while [ $# != 0 ];
do
case $1 in
-80) COLS=80
FONT=spc12x24c
BFONT=spc12x24b
shift
;;
-132) COLS=132
FONT=9x15
BFONT=9x15b
shift
;;
-fg) shift
FG=$1
shift;;
-bg) shift
BG=$1
shift;;
-fn) shift
FONT=$1
shift;;
-fb) shift
BFONT=$1
shift;;
-host) shift
HOST=$1
shift;;
-help) Usage;;
*) Usage;;
esac
done
xterm -title "VMSTERM" -sb -sl 1000 -geo ${COLS}x24 -fg ${FG} -bg ${BG} \
-cr blue -fn ${FONT} -fb ${BFONT} -xrm \
"XTerm*vt100.translations: #override \n \
~Shift <Key>F1: string(0x1b) string("OP") \n \
~Shift <Key>F2: string(0x1b) string("OQ") \n \
~Shift <Key>F3: string(0x1b) string("OR") \n \
~Shift <Key>F4: string(0x1b) string("OS") \n \
~Shift <Key>F5: string("Break") \n \
~Shift <Key>F6: string(0x1b) string("[17~") \n \
~Shift <Key>F7: string(0x1b) string("[18~") \n \
~Shift <Key>F8: string(0x1b) string("[19~") \n \
~Shift <Key>F9: string(0x1b) string("[20~") \n \
~Shift <Key>F10: string(0x1b) string("[21~") \n \
~Shift <Key>F11: string(0x1b) string("[23~") \n \
~Shift <Key>F12: string(0x1b) string("[24~") \n \
Shift <Key>F1: string(0x1b) string("[23~") \n \
Shift <Key>F2: string(0x1b) string("[24~") \n \
Shift <Key>F3: string(0x1b) string("[25~") \n \
Shift <Key>F4: string(0x1b) string("[26~") \n \
Shift <Key>F5: string(0x1b) string("[28~") \n \
Shift <Key>F6: string(0x1b) string("[29~") \n \
Shift <Key>F7: string(0x1b) string("[31~") \n \
Shift <Key>F8: string(0x1b) string("[32~") \n \
Shift <Key>F9: string(0x1b) string("[33~") \n \
Shift <Key>F10: string(0x1b) string("[34~") \n \
Shift <Key>F11: string(0x1b) string("[28~") \n \
Shift <Key>F12: string(0x1b) string("[29~") \n \
<Key>Print: string(0x1b) string("[28~") \n \
<Key>Cancel: string(0x1b) string("[29~") \n \
<Key>Pause: string(0x1b) string("Om") \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("OQ") \n \
<Key>KP_Multiply: string(0x1b) string("OR") \n \
<Key>KP_Subtract: string(0x1b) string("OS") \n \
<Key>KP_Add: string(0x1b) string("Ol") \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 \
~Shift <Key>Up: string(0x1b) string("[A") \n \
Shift <Key>Up: scroll-back(1,lines) \n \
~Shift <Key>Down: string(0x1b) string("[B") \n \
Shift <Key>Down: scroll-forw(1,lines) \n \
<Key>Right: string(0x1b) string("[C") \n \
<Key>Left: string(0x1b) string("[D")" \
-e telnet $HOST
The control sequences for doing this are documented in ctlseqs.ms.
The usual context for this question is setting the title according to the current working directory. People post answers to this periodically on the newsgroups. Here is one that I have seen, from Roy Wright <nobody@roystoy.dseg.ti.com>. In your /etc/profile after:
if [ "$SHELL" = "/bin/pdksh" -o "$SHELL" = "/bin/ksh" ]; then
PS1="! $ "
elif [ "$SHELL" = "/bin/zsh" ]; then
PS1="%m:%~%# "
elif [ "$SHELL" = "/bin/ash" ]; then
PS1="$ "
else
PS1='\u@\h:\w\$ '
fi
add:
if [ "$TERM" = "xterm" ]; then
PS1="\033]2;\u@\h:\w\007bash$ "
fi
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 <mi@rtfm.ziplink.net>: Make sure that the locale (LANG and/or LOCALE environment variable) is known to X Window System. Check ${X11ROOT}/lib/X11/locale.* for it. If it is not listed in either one of the files, find the nearest match and add an alias to it. Restart X if you have made changes.
On a related note, some people want to know how to read the title from an xterm. This works for modern xterm and dtterm, but not for other variations:
#!/bin/ksh
# Echo the current X term title bar to standard output.
# Written by Icarus Sparry <icarus@bath.ac.uk> 11 Apr 1997
#
exec </dev/tty
old=$(stty -g)
stty raw -echo min 0 time ${1-10}
print "\033[21t\c" > /dev/tty
IFS='' read -r a
stty $old
b=${a#???}
print -R "${b%??}"
But it is possible to avoid escape sequences altogether (from Hemant Shah <shah@typhoon.xnet.com>):
$ 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).
Your copy of xterm may not have enough permissions to use existing pty's:
# ls -l /dev/ptmx crw-rw---- 1 root tty 5, 2 Aug 21 20:19 /dev/ptmx
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.
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.
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.
XTerm has two useful options for controlling the shell that is run:
.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).
Well, it may be set, but not correctly. You may notice these symptoms:
stty -a
shows the rows and/or columns values as 0, or some other value (such as 65) which has nothing to do with the actual window size.
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.
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
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.
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.
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.
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.
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.
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.
You must have the eightBitInput resource set to
do this.
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.
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.
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
or
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(0x7f)
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.
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.
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:
/usr/lib/X11/app-defaults/XTerm /usr/openwin/lib/app-defaults
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 am 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
/usr/lpp/X11/lib/X11/app-defaultsThere 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 \
@Num_Lock<Key>KP_0: string(0)\n\
@Num_Lock<Key>KP_1: string(1)\n\
@Num_Lock<Key>KP_2: string(2)\n\
@Num_Lock<Key>KP_3: string(3)\n\
@Num_Lock<Key>KP_4: string(4)\n\
@Num_Lock<Key>KP_5: string(5)\n\
@Num_Lock<Key>KP_6: string(6)\n\
@Num_Lock<Key>KP_7: string(7)\n\
@Num_Lock<Key>KP_8: string(8)\n\
@Num_Lock<Key>KP_9: string(9)\n\
@Num_Lock<Key>KP_Add: string(+)\n\
@Num_Lock<Key>KP_Decimal: string(.)\n\
@Num_Lock<Key>KP_Divide: string(/)\n\
@Num_Lock<Key>KP_Enter: string(\015)\n\
@Num_Lock<Key>KP_Equal: string(=)\n\
@Num_Lock<Key>KP_Multiply: string(*)\n\
@Num_Lock<Key>KP_Subtract: string(-)\n\
<Key>Prior:scroll-back(1,page)\n\
<Key>Next:scroll-forw(1,page)\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\
~Shift<Key>Home: string(\033[1~)\n\
~Shift<Key>End: string(\033[4~)\n\
~Shift<Key>Prior: string(\033[5~)\n\
~Shift<Key>Next: string(\033[6~)\n\
Shift<Key>Prior: scroll-back(1,page) \n\
Shift<Key>Next: scroll-forw(1,page) \n\
Shift<Key>Home: scroll-back(100,page) \n\
Shift<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.
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.,
ESC [ H ESC [ F
for normal mode, and
ESC O H ESC O F
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.
*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:
xkb with X11R6
interfered with xterm's use of the NumLock key for the numeric
keypad.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):
kbs from ^H to
^?. As noted, the terminfo I wrote for XFree86
3.3.x has an error. Setting
*sunKeyboard: true
in the app-defaults file fixes the problem with xterm-88, which was that I documented in the terminfo the behavior with that resource set. Similarly, setting
*backarrowKey: false
is one way to address Debian's change to
kbs.
*backarrowKeyIsErase: true
which would not affect the home/end keys. (The color resources are redundant, so that is not a problem either).
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
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.
See Alt-keys do not work in bash.
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.
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.
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.
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.
No.
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.
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:
The “dynamic colors” feature came from a patch written by Erik Fortune (at SGI). Someone applied this to the XFree86 sources (probably early 1995).
Since X11R4, xterm had colors for foreground and background in the VT100 and Tek4014 widgets, as well as cursor- and mouse-colors which could be set via resources. But those were static. The dynamic colors feature allowed those colors to be set via escape sequences.
“Colour support” was a set of changes for ANSI
color. It might have been based on a patch (said to be of
unknown authorship) for X11R5 xterm incorporated into a
program called color_xterm. Raymond's comment in
terminfo.src implies that this program was distributed
earlier; however the copy of color_xterm-alpha4
which I have at hand has file modification dates starting in
December 1995. Wexelblat's commit is an earlier
non-patch use of the feature for xterm.
Both were probably due to Tom Weinstein (also at SGI) in
1992, which you can find in the
historic Linux archive. The README.color
file in this earlier
color_xterm says
2) Added ISO 6429 support for color text. You can set the foreground
and background color for text using SGR. For example, to make the
foreground red, you do: "^[[31m". The values from 30 to 37 set
foreground, those from 40 to 47 set background. The default colors
are:
0) black 1) red 2) green 3) yellow 4) blue 5) magenta
6) cyan 7) white
These are settable with the resources "color0" to "color1"
Aside from README.color, there was no
documentation. The terminal description was unmodified.
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 color0Blackblackblackcolor1Redred3red3color2Greengreen3green3color3Yellowyellow3yellow3color4Blueblue3blue2color5Magentamagenta3magenta3color6Cyancyan3cyan3color7Whitegray90gray90color8gray30gray50color9redredcolor10greengreencolor11yellowyellowcolor12bluergb:5c/5c/ffcolor13magentamagentacolor14cyancyancolor15whitewhitecolorULyellowforegroundcolorBDwhiteforegroundcolorRVforegroundcolorITforeground
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:
8colors.sh, 16colors.sh256colors.pl and
256colors2.pl88colors.pl and
88colors2.plI 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. Reducing that to a single byte allowed a similar scheme using a 4x4x4 cube and a proportionately shorter grayscale ramp.
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:
The default color palette for xterm uses header-files generated using scripts similar to the ones provided for demonstrations (patch #112).
The first 16 colors (except for blue) use names in the X
rgb.txt.
The X libraries cannot handle enough resources to specify all of the 256 colors as well as other features in xterm.
Starting with patch #129, I made the resource settings for colors past the first 16 a compile-time option. If you prefer to have the colors as X resource values, you lose UTF-8. Since xterm accepted escape sequences for setting the palette, this was not a problem.
We used semicolon (like other SGR parameters) for separating the R/G/B values in the escape sequence, since a copy of ITU T.416 (ISO-8613-6) which presumably clarified the use of colon for this feature was costly.
Using semicolon was incorrect because some applications could expect their parameters to be order-independent. As used for the R/G/B values, that was order-dependent. The relevant information, by the way, is part of ECMA-48 (not ITU T.416, as mentioned in Why only 16 (or 256) colors?). Quoting from section 5.4.2 of ECMA-48, page 12, and adding emphasis (not in the standard):
Each parameter sub-string consists of one or more bit combinations from 03/00 to 03/10; the bit combinations from 03/00 to 03/09 represent the digits ZERO to NINE; bit combination 03/10 may be used as a separator in a parameter sub-string, for example, to separate the fractional part of a decimal number from the integer part of that number.
and later on page 78, in 8.3.117 SGR – SELECT GRAPHIC RENDITION, the description of SGR 38:
(reserved for future standardization; intended for setting character foreground colour as specified in ISO 8613-6 [CCITT Recommendation T.416])
Of course you will immediately recognize that 03/10 is ASCII colon, and that ISO 8613-6 necessarily refers to the encoding in a parameter sub-string. Or perhaps you will not.
It took several years for this to become an issue. The developers of other terminal emulators were not the ones who first complained about it. In fact, though the order-dependence was mentioned, no one pointed to a specific program which was affected. Still, it was a known problem.
Finally, in 2012 (patch #282), I extended the parser to accommodate the "correct" syntax. The original remains, simply because of its widespread use. As before, it took a few years for other terminal developers to notice and start incorporating the improvement. As of March 2016, not all had finished noticing.
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.
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.
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":
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.
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:
and the second uses a (same size) TrueType font:
GNU grep (version 2.5) introduced a --color
option.
It does this for each highlighted match:
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:
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:
xterm, kterm, mlterm, some operating system consoles are consistent with the VT100 behavior.
rxvt, screen, putty (pterm), konsole, vte (gnome-terminal, xfce4-terminal) are not consistent with VT100 (and behave differently compared to each other).
I included screen here because it claims to be a vt100 emulator, and putty since it claims to be an xterm emulator. I did not include tmux, because it does not make either claim.
In the vttest page, I have provided screenshots to illustrate these points.
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:
Aside from the VT525, DEC terminals had no support for ANSI colors.
Likely, they were thinking of a terminal emulator which supported colors. A while back, there was more than one which said they were a “VT340” and the misconceptions began. Not all of those behaved the same.
Some developers were aware of this, others were not. The comp.os.vms newsgroup thread How to setting color in code for a VT terminal shows both.
From the outset, modern xterm was a VT100 or VT220 with ANSI colors. No technical manual was available for a VT525 at the time. Lacking a technical manual, information about a VT525 was no more reliable than the statements about a VT340.
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.
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).
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:
- formats
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.
- containers
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.
and
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\
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 do 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).
I have an old (3.1.2G) bug report for xterm which may be related to the second (3.9s) problem:
When I change font size often I will get the double-refresh, and when that happens the text program gets 2 resize events.. Running a quick test, I got this: Going to a bigger font, it got a 53x20 resize, then a 80x24 resize. Going to a smaller font, it got a 120x27 resize, then a 80x24 resize.
Earlier I made a mention of changing font size in rxvt (And xterm does it to) causing 2 resize events. Well I just happened to do it in fvwm (Instead of fvwm 95) and found it seems to be a 'feature' of fvwm95, not XFree86 as I'd initially assumed.
Using the XFree86 xterm-53 with the active icon feature on, I get some problems resizing where the xterm window shrinks as small as possible and won't stay at whatever size you set it thereafter.
Comment out the PixmapPath and IconPath from your .fvwmrc file to disable the fvwm icons and restart the WM. Start an xterm. Iconify xterm and maximize it again. Use resize button or corners to resize the xterm.
The xterm now shrinks to a tiny size and attempts to resize it result in it shrinking again.
I've tried this with fvwm 1.23 and fvwm 2.0.46 with the same results. Olvm, olvwm and twm all behave correctly so it may be a fvwm problem.
I have not observed the first, but have reproduced the second.
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 <helgaas@dhc.net>) 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:
smcup\E7saves the cursor's position\E[?47hswitches to the alternate screenrmcup\E[2Jclears the screen (assumed to be the alternate screen)\E[?47lswitches back to the normal screen\E8restores 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.
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).
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.
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
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)
- XawPlus
- Xaw3d
- 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.
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:
When xterm was first written, machines had less memory, and scrolling back a thousand lines seemed good enough for users. Internally, xterm stored the current screen and saved-lines in a large array. It scrolled the array by shifting the entire array by a given number of rows. For a thousand lines saved-lines (the scrollback region), that works well enough.
But the saveLines
resource allows a full integer, and during the
mid/late-1990s, a few users found that setting the resource
to a million lines made xterm very slow.
Still, the graphics display was fast enough. By the way, xterm uses the XCopyArea function, and normally (attempts to) display all of the updates to the screen.
Later, rxvt came along. It limited the number of saved-lines to a signed 16-bit integer, i.e., 32767 (and some packagers limited it to only a few thousand lines), and moved just the pointers to the line data when scrolling rather than shifting all of the text. It also uses XCopyArea, noting in its features
/* * Define to remove support for XCopyArea() support. XCopyArea() is useful * for scrolling on non-local X displays */ /* #define NO_SLOW_LINK_SUPPORT */
Unlike xterm, rxvt did not attempt to display all updates. If it fell behind, it would discard some of the updates, to catch up. Doing that had a greater effect on the apparent scrolling speed than its internal memory organization, since it was useful for any number of saved-lines. One drawback was that ASCII animations were somewhat erratic.
A few other terminal emulators, such as konsole copied the rxvt feature. Others copied, in turn, from whatever source. As a result, one cannot compare the speed of different terminal emulators, since they do not follow the same rules.
The issue with xterm shifting a large array was a problem which was addressed by changing all of the pointers to its line data into a circular array in 2009 (patch #244).
Even after improving the memory performance of scrolling, rxvt and its imitators still appeared to scroll faster.
The fastScroll
resource added in patch #244 provides a simple implementation
of the rxvt (mis?)feature for xterm.
As implemented, it is rather crude (sometimes xterm — like konsole — appears to stop, since it is waiting for a new set of screen updates after having discarded some).
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 | |
Notes:
On both systems, gnome-terminal failed to connect remotely.
Fedora does not have mlterm.
On Ubuntu, the urxvt package hijacks the name “rxvt”, so the “rxvt” actually tested was rxvt-xpm from the rxvt 2.7.10 package.
Fedora has st 0.70, while Ubuntu has version 0.60, which is a couple of years older.
Running locally on Ubuntu, pterm 0.70-1 dumped core.
Regarding the selection of terminal emulators:
Keeping mind that this is an xterm FAQ, the Linux console (and Windows console, and PuTTY running on Windows) are off-topic.
The table mentions programs which at one time or another
have set TERM=xterm.
The actual test does not rely upon the terminal description, nor in fact on any terminal description. The distinction was made for their relevance to this FAQ.
Given that, rxvt 2.7.10 is listed, as well as its descendent urxvt (rxvt-unicode).
Other variations of rxvt (such as aterm and mrxvt) were considered, but since much of the related code is identical, not very interesting.
Both systems have several variants of the skins for the VTE library, but for both systems, the developers have a heavy bias in favor of the GNOME desktop. Comparing the performance of the various skins would be pointless, since not all are equally supported (due to the GNOME developers' practice of making incompatible changes), and would make an unbalanced comparison in any case.
The Unix port of PuTTY, pterm is listed. It uses GDK.
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.
For a while, XCopyArea was a problem, where the Xorg hackers had degraded its performance radically. While that might still be the underlying issue, st and urxvt do use that function.
Another possibility is mentioned in Why is the text in the wrong place? where the apparent root cause was a server feature which only implemented parts of the X protocol.
Using the fastScroll feature made xterm performance comparable to the “desktop” applications. But as usual, with performance data, your mileage may vary.
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.
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").
Are you using Ubuntu? This is a frequently-reported problem for Ubuntu users. With other systems, it can occur (as of September 2012), but is less frequent. But it has been an issue with Ubuntu since 2008.
There are several related symptoms, e.g.,
Here are some of the corresponding bug reports:
Since the problem is not in xterm, all I can do is to help
forward those bug-reports to whatever package owns
compiz. What these have in common is that someone
has written code which is tested against only a small subset of
the X protocol.
Looking for solutions (since compiz is not being fixed), it is possible to disable compiz. The means for doing this vary with time. Aside from pointing to the root cause of the problem, there is little advice that is useful.
borderWidth
resource to zero.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 (XTerm-col.ad, which is
installed as XTerm-color), and the regular one
(XTerm.ad, 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
XTerm-col.ad. 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
XUSERFILESEARCHPATHor the environment variableXAPPLRESDIR(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 theXFILESEARCHPATHenvironment 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
xrdb prevented you from changing things
dynamically.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.
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:
XTerm*openIm:false
XTerm*translations: #override\n\
<Leave>, ~Ctrl ~Meta <Btn2Up>: ignore()\n\
~Shift <Key>KP_8: scroll-back(1,line)\n\
~Shift <Key>KP_2: scroll-forw(1,line)\n\
Shift <Key>KP_8: scroll-back(1,halfpage)\n\
Shift <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\
~Shift <Key>KP_8: scroll-back(1,line)\n\
~Shift <Key>KP_2: scroll-forw(1,line)\n\
Shift <Key>KP_8: scroll-back(1,halfpage)\n\
Shift <Key>KP_2: scroll-forw(1,halfpage)
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.
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.
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.
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
This is based on the X Consortium source, with the same bugs.
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.
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 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
\E[>c
expecting a response such as
\E[>0;138;0c
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 1.4.0.4 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):
src/vte.c —a 14,125 line
file. For example, the chunks related to DEC VT220 keyboard
queries and DEC private modes contain comments copied from
xterm's source code.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:
- inaccurate
- 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.
- misleading
- 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 is 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 <dickey@herndon4.his.com> 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 changes. Christian
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
and
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
xtermterminals
as well as the extended comment about the X Consortium, and replaced it with a less specific statement.
Terminalis a terminal emulator application for accessing a UNIX shell environment which can be used to run programs available on your system.
Terminalsupports escape sequences that control cursor position and colors.
The assertion about GNOME-Terminal's support for “any” persists in its package description in Debian as of 2018:
- Access a UNIX shell in the GNOME environment.
- Run any application that is designed to run on VT102, VT220, and xterm terminals.
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.
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:
#600659 – Home/End generate wrong control sequences.
Escape sequences of the "default" kbd mode should be okay now. Removal of non-default kbd modes is continued in bug 730137.
#169295 – builtin termcap parser not needed.
all: Use terminfo instead of termcap
#730137 – Drop (or fix) nondefault fkey modes.
Since addressing bug 600659, VTE's default mode pretty accurately matches XTerm. VT220 is okay-ish, Legacy is so-so, HP and Sun are quite broken.
#728900 – use terminfo instead of termcap.
Do we want to rely in term{cap,info} *at all*? So far vte has used hardwired sequences most of the time. Even if we clean up everything and drop all fkey modes except the default (request: bug 600659 comment 73) we'd need to keep hardwired sequences for numerous reasons. E.g. Home/End should generate ^[[H/^[[F which are not present in terminfo. Application cursor keys and application keypad mode alter some sequences, with AFAICT no terminfo support whatsoever. F1..F4 completely change their sequences when a modifier is pressed, again probably no support for it in terminfo. Terminfo is just able to encode the complexity (app keypad mode, app cursor mode, modifiers, numlock) we need. Wouldn't life be much simpler with just hardwired xterm-compatible sequences? (With probably a way to override them from config file or dconf for experts.)
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 Terminal.app) 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:
In a recent rewrite (early 2018), they introduced a skeleton of code from Paul Williams' sample parser. However, that is only a skeleton. For the flesh – the usual approach (see above).
Two thirds of the functions listed in the skeleton are no-ops (not implemented). Some of those listed as implemented do not work (see vttest screenshot).
Oddly enough, the developers decided to make the program claim that it is a VT525, though the skeleton demonstrates that it lacks almost all of features provided by the corresponding hardware terminal. In October 2018, the listing shows mostly a subset of VT100, with some (longstanding) features adapted from xterm, and a few inspired by ECMA-48.
The latter requires some comment: xterm's control-sequences document has mentioned ISO-6429 and ECMA-48 since its earliest version. VTE's developers used that as a reference (see bug reports) along with xterm's source-code rather than ECMA-48 or DEC's manuals until around the end of 2016.
They did (eventually) read documentation referred to by xterm's documentation. But that took a while, and they began copying from those sources before understanding the tradeoffs in those. Late in 2016, they started copying features mentioned in ECMA-48. The motivation for that appears to be (unsurprisingly) copying from yet another source, e.g., kitty (yet another copyist: see alacritty, but also see this).
Introducing those ECMA-48 features caused additional failures for VTE versus vttest, which the VTE developers solved by making a copy of vttest, to “fix” the bug. The problematic feature dealt with clearing tab-stops (selection 2).
ECMA-48 specifies the active line in selection 2. For an explanation of the term, see ECMA-48 section 6.1.5 Relationship between active data position and active presentation position. This is distinct from the feature which DEC's terminals supported.
The VT520 manual is explicit in this case (only selections 0 and 3 are supported). Per Lindberg's 1985 test demonstrated that the unsupported selection 2 was ignored in a VT100.
The VT520 did not support the other selection (2) because that would have been used to support bi-directional text. There is no point in having two differently-worded descriptions of the same identical feature.
At the time the VTE developers copied the feature, they had not begun to develop support bi-directional text (see for example #321490 (vtebidi) and #767529 (vteemoji).
At this writing (two years later), the developers are talking about working on that. The tab-clear operation is still identical in VTE for both selections 2 and 3, and does not implement any of the features from ECMA-48 not in a DEC terminal.
Other problems with VTE —
These are a few of the interesting bugs found in VTE (or GNOME Terminal) during 2017:
system crash due to running out of memory
while running ncurses sample programs dots and
picsmap.
failure to open a remote connection, noticed
while gathering data for a discussion
of scrolling performance.
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:
Its home page refers to "at least two versions". I recall seeing an older version which was apparently not based on VTE. There did not appear to be any relevant page (as of 2009) for that version.
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 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 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: larry.mcrcim.mcgill.edu (132.206.1.1) in /X/mterm.src/mterm.ball-o-wax.
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:
The original patch changes only the scrollbars to Motif, leaving the popup menus in Athena widgets. That was not what I wanted. My motivation for using Motif is not for performance or esthetics, of course, but to make it simpler to build on hosts that have no Athena widgets installed.
I set those changes aside, having found (the hard way) that the Motif library has hardcoded behavior regarding the control right-mouse button. According to the O'Reilly book on Motif programming (volume 6), it does a server grab when processing menus. Making the menus behave just as in the Athena widgets can cause the X server to hang. (I was able to do this with both Lesstif and Motif libraries). Given that, I decided to restructure the menus entirely, making a toolbar which could support at compile-time either widget set.
Distributed with Redhat Linux 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'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 have 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 |
| terminal.app | 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:
ech (erase characters)
does not follow DEC VT220 (also ISO 6429), causing applications
using this function to misbehave.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.
Rxvt revisited, this program depends only on the X11 library. As of January 2013, it is in heavy development, and (according to comments on its developer's list) growing steadily as the developers implement useful features adapted from xterm.
For instance, the size counting libraries for st 0.3 on my Debian testing machine is on a par with rxvt (and half that of xterm, which uses the X Toolkit library).
By the way, the 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 Linux, something is wrong with the resource processing (neither popup menus nor colors work).
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:
xterm-6.5.1-i18n-0.7.patch.gz Imakefile | 25 + RELNOTES-I18N | 104 ++++++ XTerm.ad | 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.
Building a copy of xterm is simple, provided that you have a development configuration for X11:
Xaw library for popup menus.xmkmf. These utilities produce a
Makefile from the Imakefile. They are not essential, but
useful, particularly on systems with unusual
configurations.If you have a working xmkmf script (or correctly
configured imake utility), all you need to do is type
xmkmf make
I have written a configure script for xterm which can
use imake (or xmkmf) to generate a
Makefile from the Makefile.in. 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 which has been ignored so far (as of mid-2012). To work around this (and get useful warnings), I apply 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
+#ifdef _CONST_X_STRING
+typedef const char *String;
+#else
typedef char *String;
+#endif
/* 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.
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:
There appears to be no comprehensive source of information on xterm better than the documentation which comes with the source code
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.
Control sequences, i.e., programming information are in the
ctlseqs.ms 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 ctlseqs.ms | nroff -ms >ctlseqs.txt tbl ctlseqs.ms | groff -ms >ctlseqs.ps
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 is useful by itself, but is maintained for historical reasons as part of xterm. html, pdf, ps and text.
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 ibiblio.org (no Linux executables, no source, however).
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 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.
I have found Richard Shuford's archive to be invaluable for notes on the DEC VT220 and related terminals. This was a webpage but was last seen via ftp. I have a snapshot of the ftp site, here:
Though not available at the time that I was collecting most of my notes, VT100.net is also a good source of primary information.
As part of my Git exports work in 2016, I made a repository of the major X release copies of xterm source. Alan Coopersmith has a more extensive repository of the X source, which is also useful although it does not cover the full timespan. In my repository, I combined the controls-sequences document which was in specs with the program:
man directory, and program in the
xterm directory.doc tree,
others went along with the source-code, and moving the
programs such as xterm under a clients
directory.I use RCS as a starting point for creating Git repositories because it is the simplest way to capture each file's timestamp. Since the X developers switched source repositories between X10 and X11, and the RCS identifiers were no longer in sequence, I used a script to change those identifiers to a form that would not interfere with checking the sources into RCS.
The dates shown are, of course, for the xterm source-code:
| Date | Release | Mine | Alan's | |
|---|---|---|---|---|
| old-xterm | Program | Specs | ||
| 2005-12-14 | X11R6.9.0 | link | n/a | n/a |
| 2005-01-12 | X11R6.8.2 | link | n/a | n/a |
| 2004-08-20 | X11R6.8.0 | link | n/a | n/a |
| 2004-04-02 | X11R6.7 | link | n/a | n/a |
| 2001-02-09 | X11R6.6 | link | n/a | n/a |
| 2000-08-21 | X11R6.5.1 | link | n/a | n/a |
| 1998-02-09 | X11R6.4 | link | n/a | n/a |
| 1996-12-09 | X11R6.3 | link | n/a | n/a |
| 1996-02-02 | X11R6.1 | link | n/a | n/a |
| 1995-01-30 | X11R6 | link | link | link |
| 1993-11-11 | X11R5 | link | link | link |
| 1989-12-23 | X11R4 | link | link | link |
| 1988-10-27 | X11R3 | link | link | n/a |
| 1988-03-01 | X11R2 | link | link | n/a |
| 1987-09-15 | X11R1 | link | n/a | n/a |
| 1986-12-25 | X10R4 | link | n/a | n/a |
| 1986-05-17 | X10R3 | link | n/a | n/a |
The ncurses FAQ How do I get color with VT100?
discusses a widely cited bit of misinformation.
For instance, this
web search gives 3,000 hits in March 2015.
Also widely cited, Consistent BackSpace and Delete Configuration gives advice regarding backspace and delete keys which is heavily biased toward Linux. For instance:
the console referred to is the Linux console, which initially had as a goal VT220 emulation. Linux never came close to meeting that goal, which was abandoned in the late 1990s when UTF-8 became more important.
As part of that, Linux's keyboard was (actually
modelled on xterm) said to be VT220, and its
coding for the backspace key sent DEL. In
contrast, ncurses' terminal database says
kbs for the vt220 sends
^H (BS).
the guideline uses "newer", "right" and "correct" in
the part which describes DEL, versus "dirty"
and "break", "broken" in that addressing
BS.
In addition to bias, the technical remedies are unsuitable
for generic advice. In particular, the comments about
terminfo, xmodmap and xterm's translations
resource are suitable only for special cases because the
proposed solutions create problems of their own.
The page itself was written in 1997, with only minor fixes since then. Thus, it does not reflect any of the improvements made to xterm. Its lack of relevance does not prevent people from citing it. For instance, this page's accepted answer recommends that (although neither gives a useful answer to the question). Here are a few clues:
Debian's current package for ncurses uses the
--with-xterm-kbs configure option which I
added in 2012.
Debian also applies patches to many of the terminal
descriptions, including adding a patched copy of xterm's
terminfo file to ncurses's terminfo file.
The patched copy is redundant and a source of problems (since the two overlap, with slightly different goals regarding PC- and VT220-style keyboards). My intent in adding the configure option to ncurses was to wean them away from the patch. That has not happened yet.
kbs value. Rather
(referring to the source for 1.9a), it uses the termios
setting:
/*
* Check for backspace key using termios VERASE - the terminfo
* kbs entry is extremely unreliable, so cannot be safely
* used. termios should have a better idea.
*/
bspace = tty->tio.c_cc[VERASE];
if (bspace != _POSIX_VDISABLE && key == bspace)
key = KEYC_BSPACE;
At the same time, tmux sets
$TERM to "screen", by default. Debian
patches that terminal description, too. Applications
running inside tmux use that terminal
description. If instead tmux translated the
backspace key to match the value from erasechar (for
the given $TERM), its clients would receive
consistent information.
Thus, rather than blaming the user (for a "badly
configured" xterm), the actual problem is a design flaw
in tmux which should have been sent to its
developers in a bug report.
Noted here, someone pointed out an NCDware document describing its terminal control sequences.
Disregarding the title Using VT320 Terminal Emulator Escape Sequences, it described some variant of xterm rather than a DEC VT320. VT320s for example had no "alternate screen". Nor did it have a feature for the "curses (1) fix".
The NCD documentation (dated December 12, 1997) does not mention xterm. A related manual does mention xterm, but only in other sections. There are other issues with the manual. For example, aixterm (16-color) control sequences are documented as "NCD-specific values". NCD did add escape sequences for status line (kterm did this as well, according to the 6.2.0 sources dated July 1996), as well as VT220 national replacement characters (which I added early in 1998).
Terminal Function Key Escape Codes has good
intentions, but falls astray in several respects.
For a different treatment of the same material, see my
notes:
Table of function-keys for XTerm and other Terminal Emulators
First off, it (like the the documentation for GNOME Terminal) misattributes work which I did, crediting the X Consortium:
This brings me to xterm. xterm has a long history, and the function key definitions have changed over time. The original xterm from the X Consortium (even before they were absorbed by The Open Group) used escape codes based on the VT220, but extended to cover the range from F1 to F48. F1 through F12 generated, respectively, codes
^[[11~to^[[15~,^[[17~to^[[21~,^[[23~, and^[[24~.Shift-F1throughShift-F12were used for F13 through F24, and generated codes from^[[11;2~to^[[24;2~. SimilarlyCtrl-F1throughCtrl-F12were used for F25 through F36 and generated codes^[[11;5~to^[[24;5~, andCtrl-Shift-F1throughCtrl-Shift-F12were used for F37 through F48 and generated codes^[[11;6~to^[[24;6~. None of the base xterm$TERMtypes on my system correspond to this series of escape codes, though you can still get xterm to exhibit the old behavior by setting theOldXtermFKeysresource to 'true'.
Not only that, but the comment (and much of the page) is inaccurate. For instance:
Insert and
Delete on the keypad.Shift to get F13 through F24.
Control) in patch #51
(1997/9/15) as the sunKeyboard
resource.Shift for the
VT220 UDK (user-defined keys).Shift; the key combination was
popular.Control to get F25 through F36.
XK_F35.sunKeyboard resource to do this.sunKeyboard to tell
xterm to look at other keyboards to simulate the
keypad.OldXtermFKeys is the resource
class; the actual resource is
oldXtermFKeys.By the way, although The Open Group made changes, none of those have been incorporated in this version of xterm. That was intentional (see discussion). Consequently, the xterm copyright makes no mention of The Open Group.
Done, except for the corresponding support in the VT52 emulation. It would be nice to have a dialog to control this.
My configure script currently provides tests for the variations of Athena widgets (Xaw3D, neXtaw). I intend to make additional changes to support Motif scrollbars and menus. Motif requires a different style of interface for the menus: binding a popup menu to control right mouse may cause the server to hang. As an intermediate step, I implemented a toolbar for the Athena widgets. In turn, that works well enough except with XFree86 4.x: the Xaw library geometry management is broken. (Other implementations of the Athena widgets work well enough).