http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41260#c31

-Nick

On Sep 18, 2009, at 8:16 AM, Jack Howarth wrote:

/* Support -mmacosx-version-min by supplying different (stub) libgcc_s.dylib
   libraries to link against, and by not linking against libgcc_s on
   earlier-than-10.3.9.

   Note that by default, -lgcc_eh is not linked against!  This is
   because in a future version of Darwin the EH frame information may
   be in a new format, or the fallback routine might be changed; if
   you want to explicitly link against the static version of those
   routines, because you know you don't need to unwind through system
   libraries, you need to explicitly say -static-libgcc.

   If it is linked against, it has to be before -lgcc, because it may
   need symbols from -lgcc.  */
#undef REAL_LIBGCC_SPEC
#define REAL_LIBGCC_SPEC                                                   \
   "%{static-libgcc|static: -lgcc_eh -lgcc;                                \
      shared-libgcc|fexceptions|fgnu-runtime:                              \
       %:version-compare(!> 10.5 mmacosx-version-min= -lgcc_s.10.4)        \
       %:version-compare(>= 10.5 mmacosx-version-min= -lgcc_s.10.5)        \
       -lgcc;                                                              \
      :%:version-compare(>< 10.3.9 10.5 mmacosx-version-min= -lgcc_s.10.4) \
       %:version-compare(>= 10.5 mmacosx-version-min= -lgcc_s.10.5)        \
       -lgcc}"

Would it be as simple as adding...

%:version-compare(>= 10.6 mmacosx-version-min= -lgcc_s.10.5)        \
       -lSystem -lgcc}"

or could we even use...

%:version-compare(>= 10.6 mmacosx-version-min= -lgcc_s.10.5)        \
       -lSystem -lgcc_s}"

I think the second case would solve the outstanding issue of the
TLS emutls not being linked in on darwin...

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39888

which would definitely be nice and take a lot of the testsuite
out of the unsupported result mode.
   If anyone would like to propose a specific patch, I would be
more than happy to test it against current gcc trunk.
                    Jack

I think there is an extra dimension to darwin that might be confusing  
things.  Darwin uses two-level-name-space.  That means that at build  
time the linker records where it found each dylib (SO) symbol.  (It  
records the path the dylib supplied as its "install name" - not just  
the leafname as SO_NEEDED does.)

On a SnowLeopard system you *can* link against /usr/lib/libgcc_s.
10.5.dylib, but the linker will not record any symbols coming from  
it.  In fact, the link order does not matter.  That is because /usr/
lib/libgcc_s.10.5.dylib has magic symbols in it that say if you are  
targeting 10.6 then _Unwind_Resume  (and other other symbols) are not  
in that dylib, so the linker looks elsewhere and finds them in  
libSystem.B.dylib.  In other words, the compiler changes to  
SnowLeopard to omit /re-order the linking with -lgcc_s when targeting  
10.6 was just an optimization and not required.

So, when these test cases are run, is the binary linked against /usr/
lib/libgcc_s.10.5.dylib? or against some just built libgcc_s.
10.5.dylib? or against some just build libgcc_s.dylib?  If either of  
the latter, then if you changed the FSF build of libgcc_s for darwin  
to have the right magic symbols, then when targeting 10.6, the linker  
will ignore those dylibs and record that the symbols should be found  
in /usr/lib/libSystem.B.dylib at runtime.

That does mean you will *not* be implicitly testing the libgcc just  
built (since at runtime the OS implementation of libgcc functions in  
libSystem.dylib will be used), but I think this test suite is supposed  
to be testing the compiler.  So that should be OK.

It also means any new symbols introduced in main line libgcc *will* be  
recorded as coming from /custom/path/libgcc_s.1.dylib.  Which again is  
what you want (as long as the functions are independent and don't need  
to be used in sets). And very similar to the libgcc_ext.dylib idea  
from http://gcc.gnu.org/bugzilla/show_bug.cgi?id=3988

-Nick

On Sep 18, 2009, at 9:47 AM, Jack Howarth wrote:

On Sep 18, 2009, at 10:43 AM, Jack Howarth wrote:

Even if you managed to get the test suite to run where everything is  
consistently using the just built libgcc_s.dylib, how will this work  
for end users of the gcc-4.5+ who want to ship an app built with  
gcc-4.5+?  Their code needs to use the OS unwinder.

So it seems to me you should be testing the compiler against the OS  
unwinder - not against the just built unwinder.

Has something changed in the FSF unwinder that clients of gcc will want?

-Nick

P.S. One minor pet peeve of mine is that the exception handling is  
well layered (e.g. _cxa_* layer on top the _Unwind_* layer (which on  
darwin is now built upon the libunwind layer)).  Except for one  
glaring problem.  The compiler mostly emits calls to _cxa_* functions,  
but it makes one call to _Unwind_Resume.   Ah!!   And those are in  
different dylibs!  If instead the compiler made some call like  
__cxa_resume() in libstdc++.dylib which turned around and called  
_Unwind_Resume() in libgcc_s.dylib, then much of the problem above  
would never happen.

Author: rth
Date: Sat May 30 00:33:46 2009
New Revision: 147995

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=147995
Log:
        * cfgcleanup.c (try_crossjump_to_edge): Only skip past
        NOTE_INSN_BASIC_BLOCK.
        * cfglayout.c (duplicate_insn_chain): Copy epilogue insn marks.
        Duplicate NOTE_INSN_EPILOGUE_BEG notes.
        * cfgrtl.c (can_delete_note_p): Allow NOTE_INSN_EPILOGUE_BEG
        to be deleted.
        * dwarf2out.c (struct cfa_loc): Change indirect field to bitfield,
        add in_use field.
        (add_cfi): Disable check redefining cfa away from drap.
        (lookup_cfa_1): Add remember argument; handle remember/restore.
        (lookup_cfa): Pass remember argument.
        (cfa_remember): New.
        (compute_barrier_args_size_1): Remove sibcall check.
        (dwarf2out_frame_debug_def_cfa): New.
        (dwarf2out_frame_debug_adjust_cfa): New.
        (dwarf2out_frame_debug_cfa_offset): New.
        (dwarf2out_frame_debug_cfa_register): New.
        (dwarf2out_frame_debug_cfa_restore): New.
        (dwarf2out_frame_debug): Handle REG_CFA_* notes.
        (dwarf2out_begin_epilogue): New.
        (dwarf2out_frame_debug_restore_state): New.
        (dw_cfi_oprnd1_desc): Handle DW_CFA_remember_state,
        DW_CFA_restore_state.
        (output_cfi_directive): Likewise.
        (convert_cfa_to_fb_loc_list): Likewise.
        (dw_cfi_oprnd1_desc): Handle DW_CFA_restore.
        * dwarf2out.h: Update.
        * emit-rtl.c (try_split): Don't split RTX_FRAME_RELATED_P.
        (copy_insn_1): Early out for null.
        * final.c (final_scan_insn): Call dwarf2out_begin_epilogue
        and dwarf2out_frame_debug_restore_state.
        * function.c (prologue, epilogue, sibcall_epilogue): Remove.
        (prologue_insn_hash, epilogue_insn_hash): New.
        (free_after_compilation): Adjust freeing accordingly.
        (record_insns): Create hash table if needed; push insns into
        hash instead of array.
        (maybe_copy_epilogue_insn): New.
        (contains): Search hash table instead of array.
        (sibcall_epilogue_contains): Remove.
        (thread_prologue_and_epilogue_insns): Split eh_return insns
        and mark them as epilogues.
        (reposition_prologue_and_epilogue_notes): Rewrite epilogue
        scanning in terms of basic blocks.
        * insn-notes.def (CFA_RESTORE_STATE): New.
        * jump.c (returnjump_p_1): Accept EH_RETURN.
        (eh_returnjump_p_1, eh_returnjump_p): New.
        * reg-notes.def (CFA_DEF_CFA, CFA_ADJUST_CFA, CFA_OFFSET,
        CFA_REGISTER, CFA_RESTORE): New.
        * rtl.def (EH_RETURN): New.
        * rtl.h (eh_returnjump_p, maybe_copy_epilogue_insn): Declare.

        * config/bfin/bfin.md (UNSPEC_VOLATILE_EH_RETURN): Remove.
        (eh_return_internal): Use eh_return rtx; split w/ epilogue.

        * config/i386/i386.c (gen_push): Update cfa state.
        (pro_epilogue_adjust_stack): Add set_cfa argument.  When true,
        add a CFA_ADJUST_CFA note.
        (ix86_dwarf_handle_frame_unspec): Remove.
        (ix86_expand_prologue): Update cfa state.
        (ix86_emit_restore_reg_using_pop): New.
        (ix86_emit_restore_regs_using_pop): New.
        (ix86_emit_leave): New.
        (ix86_emit_restore_regs_using_mov): Add CFA_RESTORE notes.
        (ix86_expand_epilogue): Add notes for unwinding the epilogue.
        * config/i386/i386.h (struct machine_cfa_state): New.
        (ix86_cfa_state): New.
        * config/i386/i386.md (UNSPEC_EH_RETURN): Remove.
        (eh_return_internal): Merge from eh_return_<mode>,
        use eh_return rtx, split w/ epilogue.

http://gcc.gnu.org/ml/gcc/2008-11/msg00280.html

On Sep 18, 2009, at 12:33 PM, Jack Howarth wrote:

I'm not sure what you mean by depreciation here.  Some perspective:

Darwin (like windows) has it's own system exception handling  
mechanisms and GCC shouldn't try to replace it.  Darwin has always  
been extremely conservative about ABI/API changes: we don't want to  
break our customer apps.  Any changes to libgcc that would break "old"  
unwinder functionality would be unacceptable on our platform,  
regardless of whether the unwinder is part of libsystem or not.

libgcc is still useful for adding other functionality (like emultls as  
you mentioned) as well as other arithmetic support libraries.  I do  
NOT think that "libgcc shouldn't be used on darwin", I just don't  
think the EH pieces should be.

                 ...
                 DW_CFA_advance_loc4 (64)  #<-- advance to near end of  
function
                 DW_CFA_restore (rbp)
                 DW_CFA_def_cfa (rsp, 8)
                 DW_CFA_nop
                 DW_CFA_nop

The linker's conversion to compact unwind "runs" the dwarf unwind info  
for a function and then records the state at the end.  Adding unwind  
info for the epilog breaks this.  In the long term, I can add  
heuristics to the linker to detect that what looks like unwind info  
for the epilog and stop processing the dwarf instructions.

The short term fix for gcc is to *not* add epilog unwind information  
for Darwin.

Epilog unwind information is never needed for exception processing.  
Its only use is for debugging or sampling when you want to  
asynchronously make a stack back trace.

-Nick

On Sep 18, 2009, at 11:40 AM, Jack Howarth wrote:

-Nick

On Sep 18, 2009, at 2:27 PM, Nick Kledzik wrote:

FAIL: g++.dg/torture/stackalign/eh-alloca-1.C  -O3 -g  execution test
FAIL: g++.dg/torture/stackalign/eh-vararg-1.C  -O3 -g  execution test
FAIL: g++.dg/torture/stackalign/eh-vararg-2.C  -O3 -g  execution test
FAIL: g++.dg/torture/stackalign/throw-1.C  -O3 -g  execution test
FAIL: g++.dg/torture/stackalign/throw-2.C  -O3 -g  execution test
FAIL: g++.dg/torture/stackalign/throw-3.C  -O3 -g  execution test
FAIL: g++.dg/torture/stackalign/eh-alloca-1.C  -O3 -g  execution test
FAIL: g++.dg/torture/stackalign/eh-vararg-1.C  -O3 -g  execution test
FAIL: g++.dg/torture/stackalign/eh-vararg-2.C  -O3 -g  execution test
FAIL: g++.dg/torture/stackalign/throw-1.C  -O3 -g  execution test
FAIL: g++.dg/torture/stackalign/throw-2.C  -O3 -g  execution test
FAIL: g++.dg/torture/stackalign/throw-3.C  -O3 -g  execution test

These eh failures are really weird because they are triggered by the
-g option. Without -g, the resulting test cases pass their execution
tests fine. The problem doesn't exist for -m64 (with or without -g).
I posted assembly files for these testcases at the various
combinations (-m32 -O3, -m32 -O3 -g, -m64 -O3 -g, -m64 -O3)...

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37012#c48
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37012#c49
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37012#c50
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37012#c51

make -k  check RUNTESTFLAGS="--target_board=unix'{-Wl,-no_compact_unwind}'"

make -k check RUNTESTFLAGS="--target_board=unix/-Wl,-no_compact_unwind"

reveals that this approach in fact eliminates all of the eh regressions
in gcc trunk. Unfortunately, it plays havoc with the gcc.dg/pch and
g++.dg/pch test cases causing hundreds of new failures of the form...

FAIL: ./common-1.h -O0 -g (test for excess errors)
FAIL: gcc.dg/pch/common-1.c -O0 -g
FAIL: gcc.dg/pch/common-1.c -O0 -g assembly comparison

Executing on host: /sw/src/fink.build/gcc45-4.4.999-20090914/darwin_objdir/gcc/xgcc -B/sw/src/fink.build/gcc45-4.4.999-20090914/darwin_objdir/gcc/ ./common-1.h  -O0 -g   -Wl,-no_compact_unwind -o common-1.h.gch    (timeout = 300)
Undefined symbols:
  "_main", referenced from:
      start in crt1.10.5.o
ld: symbol(s) not found
collect2: ld returned 1 exit status
compiler exited with status 1
output is:
Undefined symbols:
  "_main", referenced from:
      start in crt1.10.5.o
ld: symbol(s) not found
collect2: ld returned 1 exit status

FAIL: ./common-1.h -O0 -g (test for excess errors)
Excess errors:
Undefined symbols:
  "_main", referenced from:
      start in crt1.10.5.o
ld: symbol(s) not found

pch file 'common-1.h.gch' missing
FAIL: gcc.dg/pch/common-1.c -O0 -g
assembly file 'common-1.s' missing
FAIL: gcc.dg/pch/common-1.c -O0 -g assembly comparison

 - Daniel

http://gcc.gnu.org/ml/gcc-testresults/2009-09/msg01761.html

...eliminating all of the eh regressions in gcc trunk as well as a
few failures in the libjava testsuite present since gcc 4.4. I've
submitted this change for gcc 4.5...

http://gcc.gnu.org/ml/gcc-patches/2009-09/msg01327.html