2000-11-10 08:30:00 MET

I have checked older versions of rpm. The problem seem to have been introduced
between versions
3.0.3 and 3.0.4 of the distribution.

Further investigation showed me that in the gzdFdopen function of rpmio.c, the
statement
   fdSetFdno(fd, -1);
stores -1 in the structure pointed to by the fdi variable of the main function,
the file descriptor which
was 4 before this is later pushed on the following element of the fps table.
However, there is a test later on in main :
  if (gzdi == NULL || Ferror(gzdi)) {
   fprintf(stderr, _("cannot re-open payload: %s\n"), Fstrerror(gzdi));
    	exit(EXIT_FAILURE);
    }

The Ferror (gzdi) tests both the element where fd = 4 and the previous ones
where fd = -1. This
triggers the error handling.  The Fstrerror function then returns the errcookie
pointer, which is NULL
since its creation by XfdNew. This leads to a segmentation violation in the
fprintf call.

I think there are two separate problems:
  - an inconsistency between the resetting of the fd to -1 which seems to means
this entry is not
    valid anymore and the the Ferror test which fails when it detects this -1

  - an initialization problem of the errcookie pointer to NULL.


                                                                               
hope this helps
                                                                                     
Luc

I believe this is now fixed in rpm CVS. The "fix" I am remembering is in
rpm2cpio.c to
change the check for the return code to

    gzdi = Fdopen(fdi, rpmio_flags);    /* XXX gzdi == fdi */
    if (gzdi == NULL) {
        fprintf(stderr, _("cannot re-open payload: %s\n"), Fstrerror(gzdi));
        exit(EXIT_FAILURE);
    }

Please reopen if I'm misremembering.