[Bug debug/53704] New: [4.8 Regression] ICE: in based_loc_descr, at dwarf2out.c:10027 after revision 188621

[Bug debug/53704] New: [4.8 Regression] ICE: in based_loc_descr, at dwarf2out.c:10027 after revision 188621

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53704

             Bug #: 53704
           Summary: [4.8 Regression] ICE: in based_loc_descr, at
                    dwarf2out.c:10027 after revision 188621
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: debug
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: dominiq@lps.ens.fr
                CC: burnus@net-b.de, ebotcazou@gcc.gnu.org,
                    iains@gcc.gnu.org
              Host: x86_64-apple-darwin10
            Target: x86_64-apple-darwin10
             Build: x86_64-apple-darwin10


On x86_64-apple-darwin10, revision 188621 causes/exposes the following
regressions:

Running target unix/-m32
FAIL: gfortran.dg/minmax_char_1.f90  -O3 -g  (internal compiler error)
FAIL: gfortran.dg/minmax_char_1.f90  -O3 -g  (test for excess errors)
...
Running target unix/-m64
FAIL: gfortran.dg/array_constructor_type_7.f03  -O3 -g  (internal compiler
error)
FAIL: gfortran.dg/array_constructor_type_7.f03  -O3 -g  (test for excess
errors)

In both cases, the internal compiler error appears with optimization at -O1 or
above and -g and is

[macbook] f90/bug% gfc -m32 -O1 -g minmax_char_1_red.f90
minmax_char_1_red.f90: In function 'equal':
minmax_char_1_red.f90:28:0: internal compiler error: in based_loc_descr, at
dwarf2out.c:10027

I have reduced minmax_char_1.f90 to

program test

  character(len=3) :: s
  character(len=6) :: t
  s = "gee"
  t = "crunch"

  call foo("gee   ","crunch",s,t)

contains

  subroutine foo(res_max, res_min, a, b, c, d)
    character(len=*) :: res_min, res_max
    character(len=*), optional :: a, b, c, d

    if (.not. equal(min(a,b,c,d), res_min)) call abort
    if (.not. equal(max(a,b,c,d), res_max)) call abort
  end subroutine foo

  pure function equal(a,b)
    character(len=*), intent(in) :: a, b
    logical :: equal

    equal = (len(a) == len(b)) .and. (a == b)
  end function equal

end program test

The ICE goes away if I remove one of the IF in the subroutine foo. Note also
that minmax_char_1.f90 passes with -m64 and array_constructor_type_7.f03 with
-m32.

[Bug debug/53704] [4.8 Regression] ICE: in based_loc_descr, at dwarf2out.c:10027 after revision 188621

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53704

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-06-17
     Ever Confirmed|0                           |1

--- Comment #1 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-06-17 11:52:33 UTC ---
> Running target unix/-m64
> FAIL: gfortran.dg/array_constructor_type_7.f03  -O3 -g  (internal compiler
> error)
> FAIL: gfortran.dg/array_constructor_type_7.f03  -O3 -g  (test for excess
> errors)
> 
> In both cases, the internal compiler error appears with optimization at -O1 or
> above and -g and is
> 
> [macbook] f90/bug% gfc -m32 -O1 -g minmax_char_1_red.f90
> minmax_char_1_red.f90: In function 'equal':
> minmax_char_1_red.f90:28:0: internal compiler error: in based_loc_descr, at
> dwarf2out.c:10027

But there is a kludge for this in function_possibly_abstracted_p, why doesn't
it work on Darwin as well?

[Bug debug/53704] [4.8 Regression] ICE: in based_loc_descr, at dwarf2out.c:10027 after revision 188621

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53704

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ice-on-valid-code
   Target Milestone|---                         |4.8.0

[Bug debug/53704] [4.8 Regression] ICE: in based_loc_descr, at dwarf2out.c:10027 after revision 188621

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53704

--- Comment #2 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-06-19 15:04:09 UTC ---
I guess is_fortran never returns true on Darwin, which would mean that the
debug info for Fortran is already pretty broken...

[Bug debug/53704] [4.8 Regression] ICE: in based_loc_descr, at dwarf2out.c:10027 after revision 188621

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53704

--- Comment #3 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-06-19 15:06:39 UTC ---
Created attachment 27654
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27654
Tentative fix

[Bug debug/53704] [4.8 Regression] ICE: in based_loc_descr, at dwarf2out.c:10027 after revision 188621

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53704

--- Comment #4 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-06-19 15:13:55 UTC ---
> I guess is_fortran never returns true on Darwin, which would mean that the
> debug info for Fortran is already pretty broken...

Likewise for Ada.

[Bug debug/53704] [4.8 Regression] ICE: in based_loc_descr, at dwarf2out.c:10027 after revision 188621

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53704

--- Comment #5 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-06-19 15:20:33 UTC ---
Created attachment 27655
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27655
Tentative fix #2

Slight variation.

[Bug debug/53704] [4.8 Regression] ICE: in based_loc_descr, at dwarf2out.c:10027 after revision 188621

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53704

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-06-20 07:16:11 UTC ---
(In reply to comment #5)
> Created attachment 27655 [details]
> Tentative fix #2
> 
> Slight variation.

Preapproved for trunk.

[Bug debug/53704] [4.8 Regression] ICE: in based_loc_descr, at dwarf2out.c:10027 after revision 188621

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53704

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
         AssignedTo|unassigned at gcc dot       |ebotcazou at gcc dot
                   |gnu.org                     |gnu.org

--- Comment #7 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-06-20 08:07:25 UTC ---
Thanks.  Now trying to get my hands on a Darwin box...

[Bug debug/53704] [4.8 Regression] ICE: in based_loc_descr, at dwarf2out.c:10027 after revision 188621

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53704

--- Comment #8 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-06-20 08:41:58 UTC ---
The failures are gone with the patch in comment #5 (why not
DW_LANG_Fortran95?). However it looks more like papering over the problem
rather than really fixing it:

as said in comment #0, the ICE in is gone if I comment one of the "if (.not.
equal ...", but comes back if after that, I duplicate the "call foo ...". In
addition the test passes on powerpc-apple-darwin9.8.0, but yields several
warnings (not seen for the original tests in gfortran.dg)

warning: DWARFDebugInfoEntry::AppendDependants() -- check on this item
TAG_subrange_type: attr =  AT_upper_bound  form = FORM_ref4
warning: DWARFDebugInfoEntry::AppendDependants() -- check on this item
TAG_subrange_type: attr =  AT_upper_bound  form = FORM_ref4
warning: DWARFDebugInfoEntry::AppendDependants() -- check on this item
TAG_subrange_type: attr =  AT_upper_bound  form = FORM_ref4
warning: DWARFDebugInfoEntry::AppendDependants() -- check on this item
TAG_subrange_type: attr =  AT_upper_bound  form = FORM_ref4
warning: DWARFDebugInfoEntry::AppendDependants() -- check on this item
TAG_subrange_type: attr =  AT_upper_bound  form = FORM_ref4
warning: DWARFDebugInfoEntry::AppendDependants() -- check on this item
TAG_subrange_type: attr =  AT_upper_bound  form = FORM_ref4
warning: DWARFDebugInfoEntry::AppendDependants() -- check on this item
TAG_subrange_type: attr =  AT_upper_bound  form = FORM_ref4

[Bug debug/53704] [4.8 Regression] ICE: in based_loc_descr, at dwarf2out.c:10027 after revision 188621

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53704

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-06-20 08:44:19 UTC ---
Because DW_LANG_Fortran90 is already in DWARF2, but DW_LANG_Fortran95 is only
in DWARF3 and later.  If you request DWARF2 and strictly that, no extensions,
which is what Darwin does, then you get what you request.

[Bug debug/53704] [4.8 Regression] ICE: in based_loc_descr, at dwarf2out.c:10027 after revision 188621

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53704

--- Comment #10 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-06-20 08:57:20 UTC ---
> The failures are gone with the patch in comment #5 (why not
> DW_LANG_Fortran95?). However it looks more like papering over the problem
> rather than really fixing it:

The kludge is already in function_possibly_abstracted_p, the patch only fixes
an obvious oversight for Darwin.

> as said in comment #0, the ICE in is gone if I comment one of the "if (.not.
> equal ...", but comes back if after that, I duplicate the "call foo ...". In
> addition the test passes on powerpc-apple-darwin9.8.0, but yields several
> warnings (not seen for the original tests in gfortran.dg)
> 
> warning: DWARFDebugInfoEntry::AppendDependants() -- check on this item
> TAG_subrange_type: attr =  AT_upper_bound  form = FORM_ref4
> warning: DWARFDebugInfoEntry::AppendDependants() -- check on this item
> TAG_subrange_type: attr =  AT_upper_bound  form = FORM_ref4
> warning: DWARFDebugInfoEntry::AppendDependants() -- check on this item
> TAG_subrange_type: attr =  AT_upper_bound  form = FORM_ref4
> warning: DWARFDebugInfoEntry::AppendDependants() -- check on this item
> TAG_subrange_type: attr =  AT_upper_bound  form = FORM_ref4
> warning: DWARFDebugInfoEntry::AppendDependants() -- check on this item
> TAG_subrange_type: attr =  AT_upper_bound  form = FORM_ref4
> warning: DWARFDebugInfoEntry::AppendDependants() -- check on this item
> TAG_subrange_type: attr =  AT_upper_bound  form = FORM_ref4
> warning: DWARFDebugInfoEntry::AppendDependants() -- check on this item
> TAG_subrange_type: attr =  AT_upper_bound  form = FORM_ref4

Thanks for the testing.  I'm afraid I cannot have access to a Darwin 9 box so,
if this doesn't show up on Darwin 10, someone else will need to have a look.
I presume that some code guarded by is_fortran will also need to be guarded by
a check on the DWARF version/strictness.

[Bug debug/53704] [4.8 Regression] ICE: in based_loc_descr, at dwarf2out.c:10027 after revision 188621

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53704

--- Comment #11 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-06-20 11:58:38 UTC ---
BTW the patch survived a clean bootstrap with all languages, but go, enabled
and a full regtesting.

[Bug debug/53704] [4.8 Regression] ICE: in based_loc_descr, at dwarf2out.c:10027 after revision 188621

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53704

--- Comment #12 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-06-22 07:18:56 UTC ---
Author: ebotcazou
Date: Fri Jun 22 07:18:50 2012
New Revision: 188874

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=188874
Log:
    PR debug/53704
    * dwarf2out.c (gen_compile_unit_die): Use DW_LANG_Fortran90 as language
    for GNU Fortran if in strict DWARF2 mode.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/dwarf2out.c

[Bug debug/53704] [4.8 Regression] ICE: in based_loc_descr, at dwarf2out.c:10027 after revision 188621

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53704

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED

--- Comment #13 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-06-22 07:23:19 UTC ---
Patch applied.