[Bug target/92361] New: [8/9 Regression] failing fortran libcgns test on powerpc64le-linux-gnu

[doko at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92361

            Bug ID: 92361
           Summary: [8/9 Regression] failing fortran libcgns test on
                    powerpc64le-linux-gnu
           Product: gcc
           Version: 8.3.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: doko at debian dot org
  Target Milestone: ---

[forwarded from https://bugs.debian.org/944127]

A recently uploaded libcgns 8.4.0-1~exp1 in Debian ftbfs on ppc64el, but
succeeds to build on all other architectures.

The ppc64el failure [0] occurs in the test suite (test cgread_f03), when a C
va_arg function (src/cg_ftoc.c:cg_goto_f()) is called from Fortran code
(src/tests/cgread_f03.F90:421). This function is called several times. Is is
successful at first, then it fails badly because the corresponding hidden
string length parameters[1] are equal to 0. This shouldn't happen because the
related strings are literals: 'Zone_t', 'GridCoordinates_t', 'end'.

[0]
https://buildd.debian.org/status/fetch.php?pkg=libcgns&arch=ppc64el&ver=3.4.0-1%7Eexp1&stamp=1572269974&raw=0
[1] https://gcc.gnu.org/onlinedocs/gfortran/Argument-passing-conventions.html

Further investigation shows that release 3.3.0-6 of libcgns FTBFS the very same
way with GCC 8.3.0, while it succeeded with GCC 8.2.0 [2].

[2]
https://buildd.debian.org/status/fetch.php?pkg=libcgns&arch=ppc64el&ver=3.3.0-6%2Bb2&stamp=1542797186&raw=0

I then ran a bisect on the GCC svn branch 'gcc-8-branch' and found out that the
failure was introduced by the r269349 changeset [3].

[3] https://gcc.gnu.org/viewcvs/gcc?view=revision&revision=269349

This is a backport of the trunk r268992 changetset [4] introduced during GCC-9
development.

[4] https://gcc.gnu.org/viewcvs/gcc?view=revision&revision=268992