teacl/src/problems.txt

293 lines
15 KiB
Plaintext

This file contains informal notes about TECO-C. It includes bugs
and ideas for improvements.
[Unfortunately, it isn't as up to date as possible, some problems
in here are fixed, some aren't strictly "problems"...Manfred]
. Mark made ~ and F~ commands which aren't documented.
. Mark aded Curses support, which works for split screen scrolling but
doesn't support vtedit. It also needs to completely enter or leave
split screen mode when the 7:W command is executed, because I/O is
lousy to the screen if curses is active, even when all it should do
is simple output to the screen.
. n,m:w commands should return a value, but don't
. Steve Freyder's bug: <-^N-1;a> on a large file loops infinitely under
MS-DOS and/or Unix?
. if you set eightbit (4096,0et) with TECOC and exit, it stays set. If you
do it with TECO32, it doesn't. Why? Changing the screen height/width
works as expected. (VMS)
. do a 0,1:w to TECO32, it just returns. Do it to TECOC, it dies with
bad parameter value. TECOC should be the same.
. I do a getdvi on the output channel and then do a sensemode in
open_output_terminal. They're redundant. The sensemode should be it.
. EW causes a backup file to be created.
Problems reported by Richard Graham:
. EI$ doesn't seem to work. EIfilename$ does. (unsure about this)
. :]p= returns -1 10 times, then 0 if executed when TECO first starts
up on a PC.
. Try ":stext$ .ue :s$" and get ISA error, and then (on PC) the ?
command doesn't work. Generalized problem: TECO-C doesn't
"overwrite" a numeric value that's on the stack with a new one.
. How to search for a control-E? ED flag # 1, ^Eq and ^Eu don't
seem to let Richard do it. He says TECO-11 will take FS^E$^E$.
Problems running under Unix:
. ^C doesn't work.
. look at "man 3 time" and "man localtime" to implement date/time
. Is there a way to sensibly set ET_EIGHTBIT under Unix?
. ZUNIX.C's ZChIn doesn't handle the NoWait argument
Problems running on PC:
1. editing .EXE files doesn't work. should it?
2. ZChIn doesn't handle PC-extended characters from the keyboard
3. After the work to make ^C's work better, doing a ^C in the
middle of a macro like detab.tec now works; but, ".=" displays
0. shouldn't . be in the middle of the file wherever the
^C happened?
There's a test case in TSTQR.TEC that causes TECO-11 to give the
UTC error. TECOC does too, but from then on it's screwed up. Why?
To support eight-bit characters, there are several places where casts
to "unsigned char" are used. The better solution is to go through the
code and declare everything to be unsigned char, then delete the
casts. Some of the casts are in CHMACS.H.
If you search for a string using a match control construct and the
search fails, the "?SRH Search failure "xxx"" error message shows the
match control construct with a <> surrounding it. TECO-11/32 does
not.
BldStr is used by commands other than search commands, yet it reports
the "illegal search string" error sometimes. It looks like the error
messages from TECO-11/32 aren't really correct. Need an error message
like "illegal string build construct" or something.
The error message "illegal ^E command in search string" would be
better if it were "illegal ^E construct in search string"
In ERR.C, the routines ErrMsg, ErrStr and ErrPSt all share code. Fix
it. make verbose messages for system errors. flesh out error
handling in z files Make a test case that checks the cases:
1. call to ErrMsg
2. call to ErrChr
3. call to ErrStr
4. call to ErrPSt
5. call to ZVrbos on VAX and non-VAX machines
re-indent the code (automatically) to use 4-space indenting. This
will solve the problem that some lines are now too long. use clean
indenting scheme
A command like n+^T$$ won't get parsed right. Is this a general
problem?
merge zfrsrc and baksrc
convert pushex to pshnbr and pshopr
terminal modes aren't right on Sun
use standard Isdigit, etc. macros/functions?
The non-VMS ZPrscl's concatenate argv's and spaces to re-construct the
command line. There are better ways that are operating-system
dependent.
The construct: <0a"ad>' is not well-formed in TECOC, but it is in
TECO-11/32 The loop exits when the "end loop" command is skipped. In
TECOC, this means the loop nesting control variable isn't decremented,
so we "think" we're still in a loop even though we're not. This isn't
a problem in TECO-11/32. A more common programming construct is:
<0a"ad|1;'> which works without problems in TECOC.
The @-modified form of the ! "command" is supported in TECO-11/32, not
in TECOC.
In BOTH tecos, output of a long string to SYS$OUTPUT doesn't work if
the string is long enough. TECOC gets a signal, TECO-11/32 gets a
"OPCODE reserved to Digital" error. To fix, must make the "terminal
output buffer" dynamic.
All system calls must correctly handle being interrupted by control-C.
Page 228 of "Unix Programmer's Manual, Volume 1" talks about this a
little. Each system call should check for errors, but should not
treat control-Cs as an error.
If type ^C while it's executing a macro, get "Operation completed
under control-C" and a dollar sign!
If you type a control-C while help is being displayed, you get two
prompts
Control-C works fine at command splits. Checks need to be inserted
everywhere commands do loops on command arguments. For instance, 23p
needs a check in the loop that calls SinglP.
Does ExeCtT do the right thing when it's read with no wait and the
user types control-C?
The following weird thing happens (I forget which TECO):
1. compile a file/DEBUG/NOOPT in one terminal session
2. open the file with TECO in another session
3. debug the application in the first session. Can't find file.
4. do ^Z^Z to leave TECO in the second session
5. file is deleted!!!
The ^EUq string build construct behaves differently than the
TECO-11/32 algorithm for a special case: if the character in the
q-register is a ^E. The command "5uas^EUA$" will search for ^E in
TECO-11/32 and will fail in TECO-C.
ZFrSrc exists only because I was playing with SCANC on a VAX. Delete
the VAX-specific code and then merge what's left with BakSrc, using
SIncrm in the loops. This is much cleaner.
Local q-registers need to be tested further.
CBf, SBf, TBf, ZBf do not expand?
Finish creation of TSTPW.TEC, a test case that demonstrates that
ZWRITE mimics TECO-11/32 for all cases of the edit buffer/gap and all
cases of line termination: FF, VT, CR/LF, $/CR/LF, and GapBeg and
GapEnd.
Open SOFDEV.RNO for reading, do _glossary$v$$ and see junk. Also, do
rt -r notes.rno and then y$$y$$ and see an extra blank line at the
bottom of the buffer.
Use TRNLNM instead of TRNLOG in ZINIT and ZEXEEI. Make a
general-purpose logical name translation procedure to be used by both?
File COMMANDS.TXT has several "not implemented" lines that aren't
right
Add help text for E%q and EQq to hlp file.
In TECO-11/32, the EG command uses the BldStr-like routine to build
its string. Teco-c should call BldStr from ZExeEG.
The macro system needs to be carefully checked.
If do 8192ET, then a null command string with just two escapes won't
"work". Need to hit three, and then get an error.
Scan Appendix C of TECO-11/32 manual for goodies.
In ZExeEI:
1. status != RMS$_NORMAL in ZRedEI does not error
gracefully.
2. If find $$ in the middle of the read buffer, can't
continue after executing the string.
3. EI buffer won't expand.
4. File closure on error on ^C ?
ExeCSt was changed to do CBfPtr < CStEnd instead of CBfPtr <= CStEnd.
Track and check.
Why clear q-register stack in MemIni if it's been statically
initialized in TECOC?
Page 5 of TECO.LST shows ET and ED encodings. Compare/contrast
In ExeFGt, if we're not in a loop, should behave as if we're
<ESC><ESC>. We're not doing that now. Put in
If (MstTop)
{
CBfPtr = CStEnd;
return(SUCCESS);
}
into the code before the first return(FAILURE).
On the VAX, deal with disk quota overflow. Larry Kilgallen described
the handling of this: when you get a disk quota exceeded, just issue
the identical $PUT call again to exploit overdraft. This doesn't work
for creating files.
Date and time (^B and ^H) commands work under VAX/VMS. They will work
differently under other systems. Make a standard?
Say jz10v instead of zj10v and the two TECO's behave differently
Normal TECO handles two numbers in a row differently than TECO-C does.
TECO-C dies with a debugging message!!!
use register variables
use declaration of variables within blocks
see if type casts are superfluous in places. C already does it?
The September 1984 issue of Scientific American has an article by
Niklaus Wirth that describes a fast searching algorithm. Use it?
The entire ", ', <, >, O system should be checked.
CmdMod should be checked.
Is MovGap right?
The W command is stubbed out now. For debugging, ExeW exists, SkpE
skips over EG, EM, EN and EZ as if they were valid commands. A
problem with this is that the command to turn on SEEALL mode
"-1,3:W^[" returns ERR_NCA error, "negative argument to comma".
The code for INCCBP and the other places where ++CBfPtr happens needs
the MStTop test to be changed. Now, if a UTC should be reported, it's
not. It returns SUCCESS!
ExeO needs it's handling of CBfPtr to be fixed. There aren't checks
against CStEnd, and the last "--CBfPtr" may be wrong.
Shouldn't CBfPtr be renamed CStPtr? It would relate better to CStBeg,
CStEnd, etc.
help subsystem works on VAX, won't on another machine
Get rid of ??? which marks unfinished or questionable stuff. Get rid
of calls to ExeNYI.
DoEvEs doesn't implement the ES flag exactly like the manual says
DoEvEs has a system-dependent structure (UDummy)
the free function is an int on the VAX, void on the SAGE
Standard input and standard output are not used to communicate with
the terminal. UNIX people may expect that they are. This rules out
using TECO-C in a pipe-oriented manner.
Use DbgDCh instead of ZDspCh in ErrMsg for passed string displaying
difference between TypChr, EchoIt, CharOut?
COLON sounds like a character identifier. So does ATSIGN.
ChrFuncs requires DefChars
memory overflow
skipping commands is NOT finished.
In help, for errors, trace down the possible ways the code can print
each error message. Some of them are unobvious. Give an explanation
of all the ways an error can be generated to help a user figure out
his error.
Page 92 of the manual lists two informational messages. The 16 bit of
the ET mode control flag suppresses informational messages. These
messages are neither generated nor suppressed by TECO-C.
Doing an "ERfilename$" puts the expanded filespec in the filespec
buffer. Under TECO-11/32, doing a subsequent "ER$" to change I/O
streams does not wipe out the filespec buffer. It does in TECO-C.
For all the xxxBeg, xxxEnd, xxxPtr things, change all the comments to
xxxBeg; /* first char of xxx buffer */
xxxEnd; /* last char of xxx buffer */
xxxPtr; /* last char of xxx, plus 1 */
check the MOBY MUNGER. It mentions lots of bugs in other TECOs. Some
of the bugs are esoteric and quite possibly exist in TECO-C
nulls are not stripped from input.
VAX TPU is able to read a file that is currently locked by another
user, which is useful for looking at BATCH log files while a BATCH job
is running. TECO-C can't. I looked at the microfiche code to see how
TPU does it. It seems to try to open the file, and if it gets
RMS$_FLK, then it tries to open the file again after setting the UPI
bit in FAB$SHR. This means that it is supposed to do block I/O to the
file. I couldn't find the place where actual file reading is done
(it's in $fileio, which is in TPU$CAL_MODULE, which I couldn't find.)
Does TPU do block I/O (section mapping?)
The following commands are not yet implemented in TECO-C
W video scope control
^Y equivalent to ".+^S,."
EL open log file
^F return value in console switch register
nEC control memory management
The following commands are not implemented in TECO-C, and probably
never will be, as they require operating system features that cannot
be assumed by a portable TECO.
EZ zero output tape
EM position magnetic tape
In addition, most of the commands defined in Appendix C of
the manual (Incompatible, Obsolete and System-Specific Commands)
are not implemented.
The ^R command (set or return radix) supports the 8, 10 and 16
radices. TECO-C parses command-line numeric arguments using the
current radix. This can make command execution confusing for radix
16. Consider the command string "14A=$$", which returns the ascii
value of the 14th character after the character pointer. If the radix
is 16, the "A" is parsed as part of the number, and the ascii value of
a completely different character will be returned.
This is a trade-off. TECO-11/32 always parses command-line arguments
using radix 10. This destroys TECO's utility as a calculator for
hexadecimal. I believe access to a base 16 calculator is more
important than possible user confusion. Thus, be careful if you want
to change TECO's radix. Remember to change it back. Also note that
if you use TECO-C as a hexadecimal calculator, you must precede
hexadecimal numbers which start with one of the letters A-F with a 0
in order to make TECO-C parse the number correctly.
You won't get "%Superseding file" messages.
Differences between TECO-C and TECO-11 or TECO32 under VMS:
. video only under Unix, commands somewhat different, see VIDEO.TXT
. Command line syntax different under Unix.
. EO returns 100+
. 2048 bit of ET flag reverses meaning of DELETE and BACKSPACE keys
. -1EJ command returns additional operating system types
. 0:W returns additional terminal types
. s^n$ gives "invalid search argument", TECO-11/32 does nothing
. on ^Z^Z^Z, third ^Z is echoed
. default directory is TECO$MACROS for EI command
. TECO-C won't allow HP or HPW when edit buffer is empty
. TECO-C supports q-register I/O commands E% and EQ, TECO-11
doesn't
. TECO-C supports ^EQ* and ^EQ_ as string build contructs, TECO-11
doesn't
. TECO-C does case sensitivity correctly, not with bit checking,
so TECO-C does not consider "`","{","|","}" and "~" to be
equivalent to "@","[","\","]" and "^", respectively.
. The -P, -Y, and -N commands, which are implemented in TECO-11
version 39, are not yet implemented in TECO-C.
. TECO-11 supports switches on file name arguments to commands like
ER, EB, etc. TECO-C does not.
. TECO-11/32 supports files with nonstandard record formats (like
FORTRAN carriage control). TECO-C does not.
. TECO-11/32 does funny stuff with lines ending in <ESC><CR><LF>.
TECO-C doesn't (it can't under MS-DOS, which uses <CR><LF> as
line terminators in files.)
. Escape-sequence handling not in TECOC yet.
. <DELIM> can be anything in TECO-11/32, not in TECOC
. :EGcmd args' is in TECO-11/32
. ^T on/off handling in TECO-11/32 is better than in TECOC
. values returned by initialization macro are used by TECO-11/32
. ^E can be set by the user in TECO-11/32
. :^T command exists in TECO-11/32
. nOtag0,tag1,tag2$ with an "n" that's out-of-range is supposed to
. cause no branch at all. Under TECO-32, a negative "n" causes a
branch to tag0. TECO-C correctly does no branch.