Bug 67552

NFS? Local files?

No, these are local files that I'm seeing this behavior on.  Have a machine here
at the desk if you would like to poke at it.

Take a look at:

http://bugs.guug.de/db/12/1254.html

(page linked from http://bugs.guug.de/db/pa/lmutt.html)

Making this a public bug so that people can comment on it.

Jakub, this seems related to the glibc version in use?

Right before this bug went public, I submitted it as bug number 68272.  Here is 
what I had to say:

The problem seems to be in the file mbox.c.  Around line mbox.c:908, several
checks are made to ensure the mailbox is sane:

[...]
if (fseek (ctx->fp, offset, SEEK_SET) != 0 ||  /* seek the append location 
*/      
      /* do a sanity check to make sure the mailbox looks ok */
      fgets (buf, sizeof (buf), ctx->fp) == NULL ||
      (ctx->magic == M_MBOX && mutt_strncmp ("From ", buf, 5) != 0) ||
      (ctx->magic == M_MMDF && mutt_strcmp (MMDF_SEP, buf) != 0) ||
      (ctx->magic == M_KENDRA && mutt_strcmp (KENDRA_SEP, buf) != 0))
[...]

The fgets in this check fails on my system.  If I stick a perror in the code
following the fgets, it prints "No such file or directory."  Because this check
fails, i is set to -1, causing the error message to be printed around line
mbox.c:962.

This "No such file or directory" error from fgets is strange.  None of the other
operations on ctx->fp fail, including the open operations.  Could there be a bug
in glibc?  Probably not.

I am having more trouble finding the source of this problem than I expected.  I
will update this bug report as I discover more.

And later:

I still have these problems if I downgrade the mutt rpm on my system from the
limbo version (which has never worked) to the 7.3 version (which used to work
with 7.3).  This seems to support my theory that the problem is in a library
(glibc?), not mutt.

This looks like glibc mmap stdio problem, debugging...

Oh yes, freopen ("...", "r+", f) on stdio mmap f is not doing the right thing,
will have a patch soonish.

http://sources.redhat.com/ml/libc-hacker/2002-07/msg00012.html

*** Bug 68272 has been marked as a duplicate of this bug. ***

*** Bug 68915 has been marked as a duplicate of this bug. ***

Fixed in glibc-2.2.90-15.

Fix confirmed with Milan-re0724.nightly tree.

I have a similar problem with mutt and glibc-2.2.90-15 did not fix it.
Basically, when mutt checks for new mail, it sometimes fails and prints "Mailbox
was externally modified.  Flags may be wrong.". It happens about 10% of the time
(but the problem is deterministic).

In mutt-1.4i the problem occurs at:
mbox.c:
	  /*
	   * check to see if the content-length looks valid.  we expect to
	   * to see a valid message separator at this point in the stream
	   */
	  if (fseek (ctx->fp, tmploc, SEEK_SET) != 0 ||
	      fgets (buf, sizeof (buf), ctx->fp) == NULL ||
	      mutt_strncmp ("From ", buf, 5) != 0)
	  {

In this boolean expression, fseeks works, but fgets fails with errno==2 (no such
file or directory), which makes me thinks there is still a bug in glibc.

I have tried several versions of mutt, and all of them reproduces the bug.

The problem is indeed fixed with glibc-2.2.5-37.
Should I open a new bug report?

If you are seeing behavior different than described above, then yes, please open
a separate bug and post details there.

Ok. So, see bug 70076 for this new problem.