[Still seems like there is more to do in terms of taking a hard look
and really figuring out just what is going on.

-kingdon]

Date: Mon, 07 Sep 1998 06:38:45 +1000
From: mbishop@acnielsen.com.au (Murray Bishop)
To: bug-cvs@gnu.org
Subject: log-11 fails on Windows-NT
Content-Type: multipart/mixed; boundary="------------79F651AE3DF8"

This is a multi-part message in MIME format.

--------------79F651AE3DF8
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit

Using CVS-1.10.1 built on Windows-NT with MS Visual-C++ 4.2
the sanity.sh test log-11 fails.

The problem is that trailing spaces aren't being removed
from the first two lines of the multi line log message set
in test log-3.

Here's the unexpected log-11 output with quotes around each
line to show the extra spaces.

"revision 1.1"
"date: 1998/09/06 03:44:14;  author: mbishop;  state: Exp;"
"line 1     "
"     "
"line 2"
"----------------------------"

Here's the file that's passed to CVS after -F in test log-3.

C:\u\104029~1.1-V\ccvs\WinDebug>od -c comment.tmp
0000000   l   i   n   e       1                      \r  \n
0000020          \r  \n   l   i   n   e       2  \t  \r  \n  \t  \r  \n
0000040          \t          \r  \n
0000047

Interestingly enough, whitespace following "line 2" and two
lines of whitespace after that are removed.

In src/commit.c(commit), the comment.tmp file is opened 
in binary mode. 
On windows and OS/2, that means that the \r\n sequences 
marking endlines are not translated to \n.

It looks like the trailing whitespace trimming happens in
src/subr.c(make_message_rcslegal).
It's called from src/rcs.c(RCS_checkin).

make_message_rcslegal has one loop removing ' ' and '\t'
preceding '\n' (which regards \r as non-white),
and another loop removing characters matching isspace()
from the end of the string. (I guess to remove trailing
newlines left by the first loop).
MS Visual C++ treats 0x09-0x0d and 0x20 as space, so the
\r are removed from end string.

After either or both of the attached patches, log-11 passes.

Could the patch to commit be checked in?

--------------79F651AE3DF8
Content-Type: text/plain; charset=us-ascii; name="COMMIT.DIF"
Content-Transfer-Encoding: 7bit
Content-Disposition: inline; filename="COMMIT.DIF"

*** commit.c-10.1	Sun Sep 06 10:05:47 1998
--- commit.c	Sun Sep 06 14:17:59 1998
***************
*** 422,429 ****
  	if (saved_message)
  	    error (1, 0, "cannot specify both a message and a log file");
  
! 	/* FIXME: Why is this binary?  Needs more investigation.  */
! 	if ((logfd = CVS_OPEN (logfile, O_RDONLY | OPEN_BINARY)) < 0)
  	    error (1, errno, "cannot open log file %s", logfile);
  
  	if (fstat(logfd, &statbuf) < 0)
--- 422,432 ----
  	if (saved_message)
  	    error (1, 0, "cannot specify both a message and a log file");
  
! 	/* What we have here is the name of file whose contents are
! 	 * the log message. I guess it's better to open text mode for
! 	 * those OS where it makes a difference.
! 	 */
! 	if ((logfd = CVS_OPEN (logfile, O_RDONLY)) < 0)
  	    error (1, errno, "cannot open log file %s", logfile);
  
  	if (fstat(logfd, &statbuf) < 0)

--------------79F651AE3DF8
Content-Type: text/plain; charset=us-ascii; name="SUBR.DIF"
Content-Transfer-Encoding: 7bit
Content-Disposition: inline; filename="SUBR.DIF"

*** subr.c-10.1	Fri May 29 10:22:04 1998
--- subr.c	Mon Sep 07 06:21:40 1998
***************
*** 527,533 ****
  	if (*mp == '\n')
  	{
  	    /* At end-of-line; backtrack to last non-space. */
! 	    while (dp > dst && (dp[-1] == ' ' || dp[-1] == '\t'))
  		--dp;
  	}
  	*dp++ = *mp;
--- 527,533 ----
  	if (*mp == '\n')
  	{
  	    /* At end-of-line; backtrack to last non-space. */
! 	    while (dp > dst && isspace (dp[-1]) && (dp[-1] != '\n'))
  		--dp;
  	}
  	*dp++ = *mp;

--------------79F651AE3DF8--