By Walter Alan Zintz.
Questions regarding this article should be directed to the author at firstname.lastname@example.org.
We saw how to deal with problems in a Termcap listing in the first installment of this tutorial series. Now we'll look at the analogous but more complex matter of Terminfo.
Strategic alterations to the Sun server's Termcap entry for our AT&T PC7300s solved several problems with most of the full-screen programs we use on the server. But they had no effect at all on flakey operation of Lynx--a World Wide Web browser for alphanumeric terminals--and checking everything in the Termcap entry that might have been causing this weirdness did not disclose anything that even looked suspicious. Sherlockian deduction: Lynx is not using Termcap.
[Editor's note: Lynx uses the curses library, which
can take its information from either the Termcap or Terminfo
database. Apparently, Lynx on the Sun server used by Walter's
Internet service provider uses Terminfo not Termcap, as might be
vi uses a version of curses that
employs Termcap, not Terminfo.]
Could Lynx be using an internal, hardwired terminal handler? Possibly, especially because a system administrator once told me that Lynx only works correctly on VT100 terminals. Or could Lynx be using AT&T's answer to Termcap, Terminfo? An even more likely possibility. Tests to determine which is used here would be complex and time consuming; tinkering with the internals of the Lynx browser would be even worse; so the obvious route from here is to examine the Terminfo entry for the PC7300 to see whether the problem can be isolated there.
But Terminfo entries are not viewable ASCII files, because Terminfo is a compiled database. There is an original source file, of course, and quite likely a copy is laying around on the Sun server, but the current state of the compiled version may not correspond to the source file that happens to be online. So to save a lot of grief, I'll decompile the working Terminfo entry.
That's a job for a Terminfo program called
This oddly-named program does a bewildering variety of miscellaneous
jobs, depending on the option flags used with it. In this case I'll
-L option to decompile the PC7300 entry into
comprehensible terms, not just cryptic codes. And because this
program normally flashes its output onto the standard output device,
usually the user's terminal screen, I'll have to divert it to a file.
So my actual command line is:
infocmp -L PC7300 > /tmp/verbose.pc7300.ti
And the output in the
verbose.pc7300.ti file looks like:
Terminal type PC7300 7300|unixpc|pc7300|PC7300|unix_pc|AT&T UNIX PC Model 7300 flags auto_right_margin, eat_newline_glitch, xon_xoff, numbers columns = 80, init_tabs = 8, lines = 24, strings back_tab = '\E\t', bell = '^G', carriage_return = '\r', clear_screen = '\E[2J\E[H', clr_eol = '\E[0K', clr_eos = '\E[0J', cursor_address = '\E[%i%p1%d;%p2%dH', cursor_down = '\E[B', cursor_home = '\E[H', cursor_right = '\E[C', cursor_up = '\E[A', delete_line = '\E[M', enter_alt_charset_mode = '\E[11m', enter_bold_mode = '\E[7m', enter_dim_mode = '\E[2m', enter_reverse_mode = '\E[7m', enter_standout_mode = '\E[7m', enter_underline_mode = '\E[4m', exit_alt_charset_mode = '\E[10m', exit_attribute_mode = '\E[0;10m', exit_standout_mode = '\E[0m', exit_underline_mode = '\E[0m', init_1string = '^O', insert_line = '\E[L', key_backspace = '\b', key_beg = '\Ebg', key_btab = '\E\t', key_cancel = '\Ecn', key_clear = '\Ece', key_close = '\Ecl', key_command = '\Ecm', key_copy = '\Ecp', key_create = '\Ecr', key_dc = '\Edc', key_down = '\E[B', key_end = '\Een', key_eol = '\Eci', key_eos = '\Ece', key_exit = '\Eex', key_f1 = '\EOP', key_f2 = '\EOQ', key_f3 = '\EOR', key_f4 = '\EOS', key_f5 = '\E5', key_f6 = '\E6', key_f7 = '\E7', key_f8 = '\E8', key_find = '\Efi', key_help = '\Ehl', key_home = '\Ehm', key_ic = '\Eim', key_left = '\E[D', key_mark = '\Emk', key_move = '\Emv', key_next = '\Enx', key_npage = '\Epg', key_open = '\Eop', key_options = '\Eot', key_ppage = '\EPG', key_previous = '\Epv', key_print = '\Epr', key_redo = '\Ero', key_reference = '\Ere', key_refresh = '\Erf', key_replace = '\Erp', key_restart = '\Ers', key_right = '\E[C', key_save = '\Esv', key_sbeg = '\EBG', key_scancel = '\ECN', key_scopy = '\ECP', key_screate = '\ECR', key_sdc = '\EDC', key_sdl = '\EDL', key_select = '\Esl', key_send = '\EEN', key_seol = '\ECI', key_sf = '\Erd', key_sfind = '\EFI', key_shelp = '\EHL', key_shome = '\EHM', key_sic = '\ENJ', key_sleft = '\EBW', key_smove = '\EMV', key_snext = '\ENX', key_soptions = '\EOT', key_sprevious = '\EPV', key_sr = '\Eru', key_sredo = '\ERO', key_sreplace = '\ERP', key_sright = '\EFW', key_ssave = '\ESV', key_sundo = '\EUD', key_undo = '\Eud', key_up = '\E[A', newline = '\EE', parm_delete_line = '\E[%p1%dM', parm_down_cursor = '\E[%p1%dB', parm_insert_line = '\E[%p1%dL', parm_left_cursor = '\E[%p1%dD', parm_right_cursor = '\E[%p1%dC', parm_up_cursor = '\E[%p1%dA', scroll_forward = '\n', scroll_reverse = '\EM', end of strings
A close look at these parameters shows one problem right off, in the sixth and seventh lines under the heading ``strings''. The command string to turn on bold characters and that to turn on reverse video are shown as identical. That would account for the worst problem when using the Lynx browser, so already I've confirmed that Lynx does seem to use Terminfo. I'm starting to unravel the Lynx problem, too, because a brief look back at my record of the attributes tests shows that ``enter-bold-mode'' is the one that's incorrect, and that the only correction needed is to change the ``7'' to a ``1'' there.
Before I start making corrections, though, it would be a good idea to see whether there are any other errors in this Terminfo entry. The most likely way to turn them up would be to compare the items in this entry to those in my corrected Termcap entry, which seems to be working pretty well now. I could make an eyeball comparison, but that would take time and concentration, and I still might miss the mistakes that were the easiest for the original entry's writer to make--confusing the letter ``l'' with the number ``1'', for example.
Fortunately, I can have my computer do the comparison for me.
The method may seem rather complex, but once it's understood it's
straightforward. For starters my Termcap entry has to be in a
file by itself--in this case I've named the file
pc7300.tc. Next I'll run it through a program called
which converts it to Terminfo source:
captoinfo pc7300.tc > /tmp/ck.pc7300.ti
This produces a few error messages, as
captoinfo: TERM=s4: the string termcap code 'CV' is not a valid name. captoinfo: TERM=s4: the string termcap code 'CI' is not a valid name. captoinfo: TERM=s4: the string termcap code 'KM' is not a valid name. captoinfo: TERM=s4: cap ei (info rmir) is NULL: REMOVED captoinfo: TERM=s4: cap im (info smir) is NULL: REMOVED captoinfo: obsolete 2 character name 's4' removed. synonyms are: 'PC7300|unixpc|pc7300|3b1|Safari 4'
But none of these is a real error, and none is related to the sort of difference I'm looking for. The actual converted entry looks like this:
PC7300|unixpc|pc7300|3b1|Safari 4, am, xon, cols#80, lines#24, bel=^G, blink=\E[7m\E[2m, bold=\E[1m, clear=\E[2J\E[H, cr=\r, cub1=\b, cud1=\E[B, cuf1=\E[C, cup=\E[%i%p1%2.2d;%p2%2.2dH, cuu1=\E[A, dch1=\E[1P, dim=\E[2m, dl1=\E[1M, ed=\E[0J, el=\E[0K, home=\E[H, ich1=\E[1\@, il1=\E[1L, ind=\n, kbs=\b, kcub1=\E[D, kcud1=\E[B, kcuf1=\E[C, kcuu1=\E[A, kf1=\EOc, kf2=\EOd, kf3=\EOe, kf4=\EOf, kf5=\EOg, kf6=\EOh, kf7=\EOi, kf8=\EOj, rev=\E[7m, rmso=\E[m, rmul=\E[m, sgr0=\E[0m, smso=\E[7m\E[1m, smul=\E[4m,
If you think that this format is grossly different from the
way the decompiled Terminfo entry looked above, you are so right.
When I decompiled that entry I went for maximum readability, but
normal Terminfo source, the only kind the Terminfo compiler
tic understands, has a
terseness similar to that of Termcap. To make a comparison with
the source code generated from the Termcap entry, I'll have to
decompile the Terminfo entry again, into the terser form,
infocmp -I PC7300 > /tmp/pc7300.ti
Next I'll convert each of the entries to be compared into a
form that the Unix
comm utility can
handle, with each capability on a separate line. A fairly simple
script for the
ex editor will do this, and write the
results to a new file called by the old filename with a plus sign
f %+ 1d %s/^ * %s/, /,^m/g %! sort wq
Note that the
^m in the fourth line is a
control-M or newline. To get the
vi editor to put that
into the script, I had to type a control-V preceding it.
Finally, I use the Unix
comm utility to find any
differences between the two versions, as follows (from the
comm -3 ck.pc7300.ti+ pc7300.ti+ > comm.out
This produces a listing of just those lines that are in one file but not the other. In this case the lines that are only in the file generated from the Termcap entry will be flush against the left margin; those only in the file decompiled from the Terminfo entry will be indented. The actual output:
blink=\E[7m\E[2m, bold=\E[1m, bold=\E[7m, cbt=\E\t, cub1=\b, cub=\E[%p1%dD, cud=\E[%p1%dB, cuf=\E[%p1%dC, cup=\E[%i%p1%2.2d;%p2%2.2dH, cup=\E[%i%p1%d;%p2%dH, cuu=\E[%p1%dA, dch1=\E[1P, dl1=\E[1M, dl1=\E[M, dl=\E[%p1%dM, ich1=\E[1@, il1=\E[1L, il1=\E[L, il=\E[%p1%dL, is1=^O, it#8, kBEG=\EBG, kCAN=\ECN, kCPY=\ECP, kCRT=\ECR, kDC=\EDC, kDL=\EDL, kEND=\EEN, kEOL=\ECI, kFND=\EFI, kHLP=\EHL, kHOM=\EHM, kIC=\ENJ, kLFT=\EBW, kMOV=\EMV, kNXT=\ENX, kOPT=\EOT, kPRV=\EPV, kRDO=\ERO, kRIT=\EFW, kRPL=\ERP, kSAV=\ESV, kUND=\EUD, kbeg=\Ebg, kcan=\Ecn, kcbt=\E\t, kclo=\Ecl, kclr=\Ece, kcmd=\Ecm, kcpy=\Ecp, kcrt=\Ecr, kdch1=\Edc, ked=\Ece, kel=\Eci, kend=\Een, kext=\Eex, kf1=\EOP, kf1=\EOc, kf2=\EOQ, kf2=\EOd, kf3=\EOR, kf3=\EOe, kf4=\EOS, kf4=\EOf, kf5=\E5, kf5=\EOg, kf6=\E6, kf6=\EOh, kf7=\E7, kf7=\EOi, kf8=\E8, kf8=\EOj, kfnd=\Efi, khlp=\Ehl, khome=\Ehm, kich1=\Eim, kind=\Erd, kmov=\Emv, kmrk=\Emk, knp=\Epg, knxt=\Enx, kopn=\Eop, kopt=\Eot, kpp=\EPG, kprt=\Epr, kprv=\Epv, krdo=\Ero, kref=\Ere, krfr=\Erf, kri=\Eru, krpl=\Erp, krst=\Ers, ksav=\Esv, kslt=\Esl, kund=\Eud, nel=\EE, ri=\EM, rmacs=\E[10m, rmso=\E[0m, rmso=\E[m, rmul=\E[0m, rmul=\E[m, sgr0=\E[0;10m, sgr0=\E[0m, smacs=\E[11m, smso=\E[7m, smso=\E[7m\E[1m, xenl,
At over a hundred lines, that's a pretty formidable list of
discrepancies. I'm cheered up, though, by a brief inspection
that shows many of these lines are not real problems. The very
first line, the string to turn on blinking characters, is just my
own attribute combination that I invented to replace the PC7300's
lack of a blink attribute. Of course, it doesn't appear in the
standard Terminfo entry. The next two lines tell me about
something I've already discovered: the mistake in the Terminfo
bold-characters item. All those items beginning with lower-case
``k'' that are in the standard Terminfo listing but not the one
converted from Termcap? Most of them are for function keys that
are defined in a separate keymap file that's incorporated by
reference in the Termcap entry. The
program is not smart enough to follow this reference, and even if
it were, the items in the keymap file do not use standard Termcap
terminology. So half of the trouble has vanished under the light
of knowledgeable inspection--although I should add those items
from the keymap file to the basic Termcap entry, using standard
It's less encouraging that all eight of the numbered function
keys items, ``kf1'' through ``kf8,'' show gross discrepancies:
the strings from the two entries don't even resemble each other.
The strings for numbered function keys generally follow a series
fron one key to another. The entry converted from Termcap shows
this pattern, but the native Terminfo entry doesn't, so the
latter is probably incorrect. It's easy to test the string that
any key sends: enter the
vi editor's insert mode and
press the key in question, then see what's entered into the file.
(All the alleged strings begin with the ASCII escape character,
so to quote this control character into the file I must press
control-V just before I press the key I'm checking.)
Testing two of the numbered function keys this way shows that in both cases the Termcap-derived entry is the one that's correct. A horrible thought seeps into my brain: perhaps the native Terminfo descriptions of the myriad other function keys' strings are also incorrect? Checking a randomly-chosen five of these keys shows that, in every case, the native Terminfo description bears no resemblance to what the key actually sends. This is perplexing enough that I have to investigate the whole matter before I start making changes.
How can Lynx run at all with garbled key-send items? Well, none of us uses any function keys except the four cursor arrows when running Lynx. A quick check as above showed that the cursor-arrow descriptions are not defective, which is why the problem did not show up earlier. Second, while a few typographical errors are par for the course when a new entry is written, how could this much utter nonsense appear in a single entry? And why didn't any of the myriad Sun and PC7300 users out there discover and correct this long ago?
Those latter questions may never be answered. My first
thought was that the key-send items may have come from someone
captoinfo program on the wrong Termcap
listing. This sounded just probable enough that I picked out
four of the known defective items, translated them to Termcap
format by hand, and searched the
to see whether they all came from another terminal's entry. No
entry had more than one of these key-send descriptions, and two
of the four descriptions were not in any of the entries. Could
the strings simply be associated with other keys, so that only
the key-name to string-sent links were wrong? An eyeball check
showed no sign of this. Most perplexing of all, the strings that
native Terminfo alleged were sent by these keys were usually
mnemonics for the key names--the `Move' key's alleged string was
\Emv, the `Copy' key's was
\Ecp, and so
on. Because these key-strings are strictly for keyboard to CPU
communications, and machines have no use for mnemonics, this
aspect is utterly unexplainable.
But this problem was as easy to fix as it is difficult to
explain. I did not need to know exactly which key-send items
were wrong; didn't even need to know the correct strings! I just
went into the Terminfo source file with the
editor, went to the start of each incorrect or merely doubtful
key-send string, typed a ``change'' command, and typed as the new
string a control-V followed by pressing the function key whose
string I wanted to insert. Pressing the function key sent the
actual key-send string to
vi, just as if I'd typed
it character by character, and the control-V quoted in the escape
character that begins every string sent by a function key.
Finally, I did a global substitution to replace each escape
character with the ``\E'' string that represents it in a Terminfo
or Termcap listing, and the job was done.
And that was the end of the easy improvements. From this
point on it was all grinding detail work, and without any manuals
for our PC7300s I had to answer every question the hard way.
Sometimes that meant creating a file containing a string I needed
to test, then using the Unix
cat command to send
that string to the terminal. More often I could use Terminfo
itself to simplify the testing. I had three Terminfo entries in
source form: the native one, the one I had converted from
Termcap, and a third that a helpful user had sent me, containing
his and his friends' fixes for these problems. On any item where
these differed, and that included most items, I could run Lynx
alternating between the three Terminfo entries to see which
worked the best. Of course, that meant I first had to set up a
way to switch Lynx between the three Terminfo entries.
(At this point it's reasonable to ask why I should bother with
all this work when I have a Terminfo version of the Termcap entry
that has been working pretty well. It's tempting to stick with
that version, but there are reasons against it:
captoinfocan't convert, or can't convert correctly. That keymap file is a good example.
captoinfosometimes adds information that's not in the Termcap file, basically by guessing. For instance, if the Termcap entry doesn't specify a string to sound the terminal bell,
captoinfowill assume that the ASCII
belcharacter, control-G, is what does it.
All in all, a little more work with the Terminfo entry seems likely to save future trouble, and may fix some current problems we just didn't know we had.)
As with Termcap, there's a standard place for Terminfo-driven
programs to look for the compiled data on terminals, and as with
Termcap this can be overridden by an environmental variable.
With Terminfo each terminal's description lives in its own file,
so the environmental variable must be set to the full pathname of
the top directory above the individual files. The native
Terminfo entry was already set up in the standard place, so my
first task was to use the
mkdir command to set
up directories for the from-Termcap and from-a-benefactor
versions of Terminfo. I called them
/u3/walter/terminfo3/, respectively. (Because I
expected to be working on the Terminfo corrections off and on for
some days, I did not want to put these entries in such a volatile
The Terminfo compiler--
tic--also takes care of
placing the compiled entries in the proper place. It checks for
an environmental variable that gives a special place for programs
to seek compiled Terminfo entries, and if it finds one it puts
the compiled output in that place. If none is found, it puts its
output in the standard Terminfo location, ordinarily
/usr/lib/terminfo/. Because I need
to put the Terminfo entry I've derived from termcap into
/u3/walter/terminfo2/, I need to set that
environmental variable to point there before I compile the source
code into a working entry--that will automatically put the output
where I want it, as well as keeping this variant entry from
overwriting the native entry. I set this variable the same way I
TERMCAP environment variable in the first
part of this tutorial. Because I'm using the C shell on this
system, I type:
setenv TERMINFO /u3/walter/terminfo2
Now all I need to finish the task is to type:
Compiling the donated entry follows the same pattern.
Now I can use either of the new Terminfo entries, simply by
TERMINFO variable to the correct value,
and I can go back to using the native entry by unsetting that
variable. So now I'm ready to run practical tests on any
One absolutely fundamental terminal capability is being able to
send the cursor from wherever it is now to a given row and a given
column. In Terminfo this is denoted by the
it's classified as a string item, but actually it's a way to generate
a string that carries the correct information for the particular row
and column the program wants to move to. My three Terminfo entries
have different formulas for this item:
Before I started analyzing the code, I checked to see how well
each of these items worked in practice. To do this I set the
TERMINFO variable to each of the three locations in
turn, each time pointing Lynx to a Web site where the buttons are
scattered around the screen and seeing how well the cursor moved
when I jumped from one button to another. Both the native
Terminfo entry and the one I had converted from Termcap seemed to
work exactly as they should--I couldn't tell them apart just by
watching the cursor make its moves. With the donated Terminfo
entry, cursor placement was always accurate, but motion was very
slow and rather jerky.
Now down to analysis. The three items agree that the string
to be sent to the terminal for an absolute cursor move begins
with the ASCII escape character followed by a left square bracket
[), has a semicolon (
;) in the middle,
and ends with a capital ``H''. Nothing else in any of the items
is sent verbatim to the terminal; rather, it all tells the
program how to calculate what should be sent to indicate the row
and column on which the cursor should land.
I'll start with the string
appears after the capital ``H'' in the item from the donated
entry. A dollar sign followed by a number within angle brackets
is the Terminfo signal that the host computer should insert that
many milliseconds of padding at that point. The
*) says that if the move involves a jump
of more than one line, there should be that many milliseconds of
padding for each line moved. The slash mark (
indicates that this padding is necessary even when terminal
communication is via the XON/XOFF protocol. (This last is an
example of a Terminfo exclusive that has no equivalent within
A check of the source listing for this entry shows that quite a few other items have associated padding, including items such as turning certain character attributes off and on. All this padding easily accounts for the slow, stuttering progress when moving from button to button within Lynx while running at slow line speeds (we're using the PC7300 internal modems, which max out at 1200 baud). There is another item in this entry to tell the program that padding is not needed when running at less than 9600 baud, but not all systems think this applies generally, and Lynx appears to be one of those exceptions. In sum, this padding is admittedly not needed when running only 1200-baud modems, and its appearance in the entry is causing trouble, so out it goes.
Getting back to the main sections of the three items, the
percent sign (
%) that appears so frequently in them
is the signal that what immediately follows is an instruction.
(To put an actual percent sign in a Terminfo or Termcap item,
type it twice.) Terminfo is equipped with a virtual calculator:
it has a stack, uses reverse Polish notation, and has arithmetic,
boolean and bitwise operations. These percent-sign notations
all serve as instructions to the virtual calculator.
The three items all have a
%i, which means that
the PC7300 considers the leftmost column and the topmost row as
being numbered one rather than zero. In each item the semicolon
separates the method of indicating the row (on the left) from the
method of indicating the column. And the
%pn tells the program to push the
nth parameter onto the stack, while the
%d tells it to pop the most recent value from the
stack, as a signed decimal number, and put it in the string at
this point. (Does that seem like an awfully longwinded way to
say that to move the cursor to row 7 and column 23 you simply
(escape)[7;23H down the line? It is, but most
terminals use much more complex encoding.)
The only point on which the items disagree is how those
numerical values are to be formatted. The characters between a
percent sign and the following ``d'' are formatting instructions,
which generally use the
printf() notation. So the
native Terminfo item calls for no special formatting, the item
converted from Termcap says that both output numbers must be padded to two digits, using
leading zeroes if necessary, and the donated item asks for two
digits in the row number but three in the column number.
But as I discovered in my preliminary tests, all three versions put the cursor in the correct row and column. So do I go with the native Terminfo item for the sake of simplicity, or with the donated item on the grounds that circumstances may crop up some day that will require this formatting? (The from-Termcap item is not likely to be correct because the PC7300 has modes in which there are more than 99 columns.) This is where personal judgement comes into play. Whatever the decision, it's a very good idea to go back and change the working Termcap entry to be the same, just for consistency.
The rest of the debugging work is pretty much more of the same. Most of it won't be as complex as the example we've just gone through, because absolute cursor movement is one of the most complicated functions a terminal performs. But all the work will involve comparing items to each other, analyzing them with the help of a Terminfo reference, and testing at the start and (often) at various points thereafter.
There's a lot of work to be done here, but I can get more out of it all than just correct Termcap and Terminfo entries. Here are a few examples of uses for terminal-operation data that do not involve Termcap or Terminfo:
catwhenever this happens that will reset all the terminal modes to normal.
.loginfile on the Sun server and modify my prompt visually. At the start of my prompt string I'll put the characters that turn on, say, the combination of bold and dim characters; very distinctive visually. At the end of my prompt string I'll put the characters that turn off fancy video modes. Then my prompt on the Sun server will appear in unmistakeable form, while everything else appears normally.
The only special point to remember is that, because these uses
do not involve Termcap or Terminfo, the metacharacters they use
will not work here. That is, instead of using the
\E sequence to represent the ASCII escape character,
it will be necessary to put the escape character itself in the
Last Modified: Thursday, 07-Sep-95 12:55:39 PDT