Re: -ffunction-sections -fdata-sections vs. string-inst

[Loren James Rittle <rittle at latour dot rsch dot comm dot mot dot com>]

Hi Benjamin,

I tried your latest patch in an already bootstrapped and tested tree
on i686-unknown-freebsd3.4 using the same procedure I posted last
time.  I no longer see any unreasonable expansion on
21_strings/append.cc and 21_strings/find.cc.

However, FYI, I now see new core dumps on 7 statically built
executables: append.st-exe, bitset_ctor.st-exe, ctor_copy_dtor.st-exe,
element_access.st-exe, insert.st-exe, ios_members.st-exe,
substr.st-exe

This may or may not be what you reported:
> I am getting some cores in static binaries, and I'm still in the middle
> of trying to see what's going on. I'll post a better patch tonight.

Here is the backtrace in all my new cases:

#0  count_fdes (this_fde=0x8059108) at ../../egcs/gcc/frame-dwarf2.c:257
#1  0x80498ff in frame_init (ob=0x80595c0) at ../../egcs/gcc/frame-dwarf2.c:324
#2  0x8049a56 in find_fde (pc=0x8048eb8) at ../../egcs/gcc/frame-dwarf2.c:364
#3  0x8049fb5 in __frame_state_for (pc_target=0x8048eb8, state_in=0xbfbfc638)
    at ../../egcs/gcc/frame-dwarf2.c:628
#4  0x8048ecc in __throw () at ../../egcs/gcc/libgcc2.c:3937

I believe this is exposing a bug in ld on my platform since if I
remove -Wl,--gc-sections from mkcheck, all those tests run fine again.

My opinion only, but I don't think this should stop your patch going
in since I have looked at the build logs closely and think it is doing
the right thing.  The test should continue to include
-Wl,--gc-sections in order to continue to flag the problem, but users
that have a buggy ld (if my supposition is correct) can just avoid
including that flag when they link.

I will try to form a small testcase to go in dejagnu for this issue
since then more GCC/binutil hackers will see it.  <...thinking...>
Sweet, the obvious testcase fails:

int main (void)
{
  try
    {
      throw 1;
    }
  catch (...) {}
  return 0;
}

g++ -ffunction-sections -fdata-sections -Wl,--gc-sections test.C; a.out #OK
g++ -ffunction-sections -fdata-sections -Wl,--gc-sections -static test.C; \
a.out #dumps core while processing the __throw

I can also confirm that linux fails with this case as well.

Regards,
Loren