Re: [AC21.4R21.5] Email truncation bug begone!

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&gt; * 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 */