4:03 a.m.
APPROVE COMMIT 21.5 RECOMMEND 21.4
David,
I took much too long to process this. I can only claim extreme busyness
with some Real Life stuff over the last several weeks. Thank you so
much for analyzing the problem and coming up with a fix. I'm going to
give credit where credit is due (although I did tweak your patch very
slightly in a couple of ways). The 21.5 patch, which I am committing,
is first, followed by the 21.4 version of the patch. Both have the same
ChangeLog entry.
See the discussion rooted here for more information:
http://list-archive.xemacs.org/xemacs-beta/200412/msg00128.html
src/ChangeLog addition:
2005-02-03 David Evers <extsw(a)appliedgenerics.com>
* process-unix.c (unix_send_process): Flush the last chunk, even
when the pipe is blocked.
xemacs-21.5 source patch:
Diff command: cvs -q diff -uN
Files affected: src/process-unix.c
Index: src/process-unix.c
===================================================================
RCS file: /pack/xemacscvs/XEmacs/xemacs/src/process-unix.c,v
retrieving revision 1.56
diff -d -u -r1.56 process-unix.c
--- src/process-unix.c 2005/01/24 23:34:05 1.56
+++ src/process-unix.c 2005/02/04 03:53:40
＠＠ -1511,30 +1511,38 ＠＠
Ibyte chunkbuf[512];
Bytecount chunklen;
- while (1)
+ do
{
int writeret;
chunklen = Lstream_read (lstream, chunkbuf, 512);
- if (chunklen <= 0)
- break; /* perhaps should ABORT() if < 0?
- This should never happen. */
old_sigpipe =
(SIGTYPE (*) (int)) EMACS_SIGNAL (SIGPIPE, send_process_trap);
- /* Lstream_write() will never successfully write less than
- the amount sent in. In the worst case, it just buffers
- the unwritten data. */
- writeret = Lstream_write (XLSTREAM (DATA_OUTSTREAM (p)), chunkbuf,
- chunklen);
- {
- int save_errno = errno;
- EMACS_SIGNAL (SIGPIPE, old_sigpipe);
- errno = save_errno;
- if (writeret < 0)
- /* This is a real error. Blocking errors are handled
- specially inside of the filedesc stream. */
- report_process_error ("writing to process", proc);
- }
+ if (chunklen > 0)
+ {
+ int save_errno;
+
+ /* Lstream_write() will never successfully write less than
+ the amount sent in. In the worst case, it just buffers
+ the unwritten data. */
+ writeret = Lstream_write (XLSTREAM (DATA_OUTSTREAM(p)), chunkbuf,
+ chunklen);
+ save_errno = errno;
+ EMACS_SIGNAL (SIGPIPE, old_sigpipe);
+ errno = save_errno;
+ if (writeret < 0)
+ /* This is a real error. Blocking errors are handled
+ specially inside of the filedesc stream. */
+ report_file_error ("writing to process", list1 (proc));
+ }
+ else
+ {
+ /* Need to make sure that everything up to and including the
+ last chunk is flushed, even when the pipe is currently
+ blocked. */
+ Lstream_flush (XLSTREAM (DATA_OUTSTREAM(p)));
+ EMACS_SIGNAL (SIGPIPE, old_sigpipe);
+ }
while (Lstream_was_blocked_p (XLSTREAM (p->pipe_outstream)))
{
/* Buffer is full. Wait, accepting input;
＠＠ -1549,7 +1557,9 ＠＠
Lstream_flush (XLSTREAM (p->pipe_outstream));
EMACS_SIGNAL (SIGPIPE, old_sigpipe);
}
+ /* Perhaps should ABORT() if < 0? This should never happen. */
}
+ while (chunklen > 0);
}
else
{ /* We got here from a longjmp() from the SIGPIPE handler */
Index: src/process-unix.c
===================================================================
RCS file: /pack/xemacscvs/XEmacs/xemacs/src/process-unix.c,v
retrieving revision 1.20.2.8
diff -d -u -r1.20.2.8 process-unix.c
--- src/process-unix.c 2005/01/31 02:55:26 1.20.2.8
+++ src/process-unix.c 2005/02/04 03:31:08
＠＠ -1293,26 +1293,38 ＠＠
Bufbyte chunkbuf[512];
Bytecount chunklen;
- while (1)
+ do
{
Lstream_data_count writeret;
chunklen = Lstream_read (lstream, chunkbuf, 512);
- if (chunklen <= 0)
- break; /* perhaps should ABORT() if < 0?
- This should never happen. */
old_sigpipe =
(SIGTYPE (*) (int)) signal (SIGPIPE, send_process_trap);
- /* Lstream_write() will never successfully write less than
- the amount sent in. In the worst case, it just buffers
- the unwritten data. */
- writeret = Lstream_write (XLSTREAM (DATA_OUTSTREAM(p)), chunkbuf,
- chunklen);
- signal (SIGPIPE, old_sigpipe);
- if (writeret < 0)
- /* This is a real error. Blocking errors are handled
- specially inside of the filedesc stream. */
- report_file_error ("writing to process", list1 (proc));
+ if (chunklen > 0)
+ {
+ int save_errno;
+
+ /* Lstream_write() will never successfully write less than
+ the amount sent in. In the worst case, it just buffers
+ the unwritten data. */
+ writeret = Lstream_write (XLSTREAM (DATA_OUTSTREAM(p)), chunkbuf,
+ chunklen);
+ save_errno = errno;
+ signal (SIGPIPE, old_sigpipe);
+ errno = save_errno;
+ if (writeret < 0)
+ /* This is a real error. Blocking errors are handled
+ specially inside of the filedesc stream. */
+ report_file_error ("writing to process", list1 (proc));
+ }
+ else
+ {
+ /* Need to make sure that everything up to and including the
+ last chunk is flushed, even when the pipe is currently
+ blocked. */
+ Lstream_flush (XLSTREAM (DATA_OUTSTREAM(p)));
+ signal (SIGPIPE, old_sigpipe);
+ }
while (Lstream_was_blocked_p (XLSTREAM (p->pipe_outstream)))
{
/* Buffer is full. Wait, accepting input;
＠＠ -1327,7 +1339,9 ＠＠
Lstream_flush (XLSTREAM (p->pipe_outstream));
signal (SIGPIPE, old_sigpipe);
}
+ /* Perhaps should abort() if < 0? This should never happen. */
}
+ while (chunklen > 0);
}
else
{ /* We got here from a longjmp() from the SIGPIPE handler */
--
Jerry James
http://www.ittc.ku.edu/~james/
12:53 p.m.
New subject: [Q21.4] Email truncation bug begone!
QUERY
Jerry James <james(a)xemacs.org> writes:
APPROVE COMMIT 21.5 RECOMMEND 21.4
David,
I took much too long to process this. I can only claim extreme busyness
with some Real Life stuff over the last several weeks. Thank you so
much for analyzing the problem and coming up with a fix. I'm going to
give credit where credit is due (although I did tweak your patch very
slightly in a couple of ways). The 21.5 patch, which I am committing,
is first, followed by the 21.4 version of the patch. Both have the same
ChangeLog entry.
See the discussion rooted here for more information:
http://list-archive.xemacs.org/xemacs-beta/200412/msg00128.html
src/ChangeLog addition:
2005-02-03 David Evers <extsw(a)appliedgenerics.com>
* process-unix.c (unix_send_process): Flush the last chunk, even
when the pipe is blocked.
Jerry,
Can you please define 'RECOMMEND 21.4' in this case? Please bear in
mind that 21.4.17 (which fixes a killer bug in the
soon-to-be-deprecated 21.4.16) will be released on Sunday.
Thanks,
Vin
4:50 p.m.
New subject: [Q21.4] Email truncation bug begone!
Vin Shelton <acs(a)xemacs.org> wrote:
Can you please define 'RECOMMEND 21.4' in this case? Please
bear in
mind that 21.4.17 (which fixes a killer bug in the
soon-to-be-deprecated 21.4.16) will be released on Sunday.
I mean that this patch fixes a real bug in the 21.4 series. David Evers
described how the bug manifested for him while using JDE:
http://list-archive.xemacs.org/xemacs-beta/200412/msg00128.html
It has been manifesting as the dreaded email truncation bug for some of
us 21.5 users. In general, this bug can cause subprocesses on Unix
systems to not receive all of the data that XEmacs was supposed to send
them. I have tested the 21.5 version of the patch as thoroughly as I
know how. It solves the problems I was seeing and, so far at least, has
not introduced any new ones. The 21.4 code is nearly identical
(the differences are mostly cosmetic), David reports that it solves his
JDE problem ... I think it should go into 21.4.17.
Just to be on the safe side, I'll build a 21.4 with this patch included
and run with it for the rest of today. I'll exercise it as well as I
can. If I don't pop up and say, "Hold the presses!", assume that I
didn't run into any trouble. (I'm going with negative instead of
positive acknowledgement because I'm leaving town tonight, and I may not
have time to send an email saying "No problem", although I will
certainly take the time if I do run into a problem.)
Regards
--
Jerry James
http://www.ittc.ku.edu/~james/
8:33 p.m.
Jerry James <james(a)xemacs.org> writes:
APPROVE COMMIT 21.5 RECOMMEND 21.4
Jerry, just for the record:
The subject line should say:
Re: [AC21.5R21.4] Email truncation bug begone!
instead of
Re: [AC21.4R21.5] Email truncation bug begone!
This is what you actually did, right?
Best regards,
Adrian
David,
I took much too long to process this. I can only claim extreme busyness
with some Real Life stuff over the last several weeks. Thank you so
much for analyzing the problem and coming up with a fix. I'm going to
give credit where credit is due (although I did tweak your patch very
slightly in a couple of ways). The 21.5 patch, which I am committing,
is first, followed by the 21.4 version of the patch. Both have the same
ChangeLog entry.
See the discussion rooted here for more information:
http://list-archive.xemacs.org/xemacs-beta/200412/msg00128.html
src/ChangeLog addition:
2005-02-03 David Evers <extsw(a)appliedgenerics.com>
* process-unix.c (unix_send_process): Flush the last chunk, even
when the pipe is blocked.
xemacs-21.5 source patch:
Diff command: cvs -q diff -uN
Files affected: src/process-unix.c
Index: src/process-unix.c
===================================================================
RCS file: /pack/xemacscvs/XEmacs/xemacs/src/process-unix.c,v
retrieving revision 1.56
diff -d -u -r1.56 process-unix.c
--- src/process-unix.c 2005/01/24 23:34:05 1.56
+++ src/process-unix.c 2005/02/04 03:53:40
＠＠ -1511,30 +1511,38 ＠＠
Ibyte chunkbuf[512];
Bytecount chunklen;
- while (1)
+ do
{
int writeret;
chunklen = Lstream_read (lstream, chunkbuf, 512);
- if (chunklen <= 0)
- break; /* perhaps should ABORT() if < 0?
- This should never happen. */
old_sigpipe =
(SIGTYPE (*) (int)) EMACS_SIGNAL (SIGPIPE, send_process_trap);
- /* Lstream_write() will never successfully write less than
- the amount sent in. In the worst case, it just buffers
- the unwritten data. */
- writeret = Lstream_write (XLSTREAM (DATA_OUTSTREAM (p)), chunkbuf,
- chunklen);
- {
- int save_errno = errno;
- EMACS_SIGNAL (SIGPIPE, old_sigpipe);
- errno = save_errno;
- if (writeret < 0)
- /* This is a real error. Blocking errors are handled
- specially inside of the filedesc stream. */
- report_process_error ("writing to process", proc);
- }
+ if (chunklen > 0)
+ {
+ int save_errno;
+
+ /* Lstream_write() will never successfully write less than
+ the amount sent in. In the worst case, it just buffers
+ the unwritten data. */
+ writeret = Lstream_write (XLSTREAM (DATA_OUTSTREAM(p)), chunkbuf,
+ chunklen);
+ save_errno = errno;
+ EMACS_SIGNAL (SIGPIPE, old_sigpipe);
+ errno = save_errno;
+ if (writeret < 0)
+ /* This is a real error. Blocking errors are handled
+ specially inside of the filedesc stream. */
+ report_file_error ("writing to process", list1 (proc));
+ }
+ else
+ {
+ /* Need to make sure that everything up to and including the
+ last chunk is flushed, even when the pipe is currently
+ blocked. */
+ Lstream_flush (XLSTREAM (DATA_OUTSTREAM(p)));
+ EMACS_SIGNAL (SIGPIPE, old_sigpipe);
+ }
while (Lstream_was_blocked_p (XLSTREAM (p->pipe_outstream)))
{
/* Buffer is full. Wait, accepting input;
＠＠ -1549,7 +1557,9 ＠＠
Lstream_flush (XLSTREAM (p->pipe_outstream));
EMACS_SIGNAL (SIGPIPE, old_sigpipe);
}
+ /* Perhaps should ABORT() if < 0? This should never happen. */
}
+ while (chunklen > 0);
}
else
{ /* We got here from a longjmp() from the SIGPIPE handler */
Index: src/process-unix.c
===================================================================
RCS file: /pack/xemacscvs/XEmacs/xemacs/src/process-unix.c,v
retrieving revision 1.20.2.8
diff -d -u -r1.20.2.8 process-unix.c
--- src/process-unix.c 2005/01/31 02:55:26 1.20.2.8
+++ src/process-unix.c 2005/02/04 03:31:08
＠＠ -1293,26 +1293,38 ＠＠
Bufbyte chunkbuf[512];
Bytecount chunklen;
- while (1)
+ do
{
Lstream_data_count writeret;
chunklen = Lstream_read (lstream, chunkbuf, 512);
- if (chunklen <= 0)
- break; /* perhaps should ABORT() if < 0?
- This should never happen. */
old_sigpipe =
(SIGTYPE (*) (int)) signal (SIGPIPE, send_process_trap);
- /* Lstream_write() will never successfully write less than
- the amount sent in. In the worst case, it just buffers
- the unwritten data. */
- writeret = Lstream_write (XLSTREAM (DATA_OUTSTREAM(p)), chunkbuf,
- chunklen);
- signal (SIGPIPE, old_sigpipe);
- if (writeret < 0)
- /* This is a real error. Blocking errors are handled
- specially inside of the filedesc stream. */
- report_file_error ("writing to process", list1 (proc));
+ if (chunklen > 0)
+ {
+ int save_errno;
+
+ /* Lstream_write() will never successfully write less than
+ the amount sent in. In the worst case, it just buffers
+ the unwritten data. */
+ writeret = Lstream_write (XLSTREAM (DATA_OUTSTREAM(p)), chunkbuf,
+ chunklen);
+ save_errno = errno;
+ signal (SIGPIPE, old_sigpipe);
+ errno = save_errno;
+ if (writeret < 0)
+ /* This is a real error. Blocking errors are handled
+ specially inside of the filedesc stream. */
+ report_file_error ("writing to process", list1 (proc));
+ }
+ else
+ {
+ /* Need to make sure that everything up to and including the
+ last chunk is flushed, even when the pipe is currently
+ blocked. */
+ Lstream_flush (XLSTREAM (DATA_OUTSTREAM(p)));
+ signal (SIGPIPE, old_sigpipe);
+ }
while (Lstream_was_blocked_p (XLSTREAM (p->pipe_outstream)))
{
/* Buffer is full. Wait, accepting input;
＠＠ -1327,7 +1339,9 ＠＠
Lstream_flush (XLSTREAM (p->pipe_outstream));
signal (SIGPIPE, old_sigpipe);
}
+ /* Perhaps should abort() if < 0? This should never happen. */
}
+ while (chunklen > 0);
}
else
{ /* We got here from a longjmp() from the SIGPIPE handler */
--
Adrian Aichner
mailto:adrian＠xemacs.org
http://www.xemacs.org/
9:11 p.m.
"Adrian Aichner" <Adrian.Aichner(a)t-online.de> wrote:
Jerry James <james(a)xemacs.org> writes:
> APPROVE COMMIT 21.5 RECOMMEND 21.4
Jerry, just for the record:
The subject line should say:
Re: [AC21.5R21.4] Email truncation bug begone!
instead of
Re: [AC21.4R21.5] Email truncation bug begone!
This is what you actually did, right?
Right. Computer, do what I meant, not what I said!
Thanks for catching that, Adrian. Blushingly yours,
--
Jerry James
http://www.ittc.ku.edu/~james/
2:51 a.m.
New subject: [AC21.4] Email truncation bug begone!
APPROVE COMMIT 21.4
Jerry James <james(a)xemacs.org> writes:
APPROVE COMMIT 21.5 RECOMMEND 21.4
David,
I took much too long to process this. I can only claim extreme busyness
with some Real Life stuff over the last several weeks. Thank you so
much for analyzing the problem and coming up with a fix. I'm going to
give credit where credit is due (although I did tweak your patch very
slightly in a couple of ways). The 21.5 patch, which I am committing,
is first, followed by the 21.4 version of the patch. Both have the same
ChangeLog entry.
See the discussion rooted here for more information:
http://list-archive.xemacs.org/xemacs-beta/200412/msg00128.html
src/ChangeLog addition:
2005-02-03 David Evers <extsw(a)appliedgenerics.com>
* process-unix.c (unix_send_process): Flush the last chunk, even
when the pipe is blocked.
Thanks for the patch, David. This patch will appear in 21.4.18.
- Vin
PS. This would be an interesting bug to add to the test suite. Any
takers?
3:15 p.m.
New subject: [AC21.4] Email truncation bug begone!
Vin Shelton wrote:
APPROVE COMMIT 21.4
....
Thanks for the patch, David. This patch will appear in 21.4.18.
When is 21.4.18 expected to be released?
It would be nice to have an official stable XEmacs that doesn't
suffer the problems described in
http://list-archive.xemacs.org/xemacs-beta/200412/msg00128.html
Cheers,
---- Dave
7023
days inactive
7260
days old
6 comments
4 participants
participants (4)
-
Adrian Aichner -
David Evers -
Jerry James -
Vin Shelton