https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65780

            Bug ID: 65780
           Summary: [5 Regression] Uninitialized common handling in
                    executables
           Product: gcc
           Version: 5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jakub at gcc dot gnu.org

int optopt;

int
main ()
{
  optopt = 4;
  return 0;
}

fails to link with GCC 5 on powerpc64{,le}-linux:
gcc -O2 -o a a.c
/usr/bin/ld: /tmp/cckV8RvX.o: In function `main':
a.c:(.text.startup+0x6): unresolvable R_PPC64_TOC16_HA against
`optopt@@GLIBC_2.3'
/usr/bin/ld: a.c:(.text.startup+0xe): unresolvable R_PPC64_TOC16_LO against
`optopt@@GLIBC_2.3'
/usr/bin/ld: final link failed: Nonrepresentable section on output
collect2: error: ld returned 1 exit status

The problem seems to be that targetm.binds_local_p now returns true for !shlib
DECL_COMMON with NULL (or error_mark_node) DECL_INITIAL.
Looking at the difference between 4.9 and 5, there is:
  if (defined_locally && weak_dominate && !shlib)
    resolved_locally = true;
and finally (similar to 4.9)
  /* Uninitialized COMMON variable may be unified with symbols
     resolved from other modules.  */
  if (DECL_COMMON (exp)
      && !resolved_locally
      && (DECL_INITIAL (exp) == NULL
          || (!in_lto_p && DECL_INITIAL (exp) == error_mark_node)))
    return false;
Looking at what x86_64-linux linker does in this case, assuming that an
unitialized common binds locally in the executable if it is defined there works
fine, but clearly not on ppc*.  Either it is a linker bug there, or we for
powerpc* (some other targets too) should treat the uninitialized DECL_COMMON
defined_locally && weak_dominate && !shlib differently - not set
resolved_locally in that case?
Can people try other targets?  The testcase relies on optopt being an integer
variable defined in libc shared library, if that is not the case, create a
shared library with that symbol and link it to the binary.

Reply via email to