Re: Problem with recent tar on NetBSD/sparc-1.5

[Paul Eggert]

> Date: Sun, 9 Sep 2001 17:25:28 +0200 (CEST)
> From: Adrian Bunk <address@hidden>
> 
> Yes, I called ktrace with the "-d" option.

Odd.  I thought that meant that ktrace would trace all descendants of
tar, including gzip.  But never mind, I think we're on the track of
the trouble now.

>   http://www.fs.tum.de/~bunk/tar-out.bz2

Here's the key part of that trace:

  1488 tar      CALL  utimes(0x46080,0xeffff360)
  1488 tar      NAMI  "ImageMagick-5.3.8/PerlMagick/t/zlib/write.t"
  1488 tar      RET   utimes 0

At this point, 'tar' should go on to drain the pipe (file descriptor
3), but instead it immediately closes it:

  1488 tar      CALL  close(0x3)
  1488 tar      RET   close 0

The relevant function in the tar source code is buffer.c's
'close_archive'.  It contains the following code:

  if (access_mode == ACCESS_READ
      && ! _isrmt (archive)
      && S_ISFIFO (archive_stat.st_mode))
    while (rmtread (archive, record_start->buffer, record_size) > 0)
      continue;

Can you please see why that while loop is not being executed?  You can
do this either with a debugger or by inserting printf statements.

It seems to me that access_mode must be ACCESS_READ, and also _isrmt
(archive) must be zero, so the likely problem is
S_ISFIFO (archive_stat.st_mode).  So I would focus on the value of the
S_ISFIFO expression.  POSIX 1003.1-200x d7 says that S_ISFIFO must
yield nonzero on a pipe, but perhaps it doesn't work that way on your
host.

Also, does config.h define STAT_MACROS_BROKEN on your host?

And does <sys/stat.h> define S_ISFIFO in your compilation environment?