[Bug sanitizer/65610] Compare debug failure with -g3 -fsanitize=undefined -fno-sanitize=vptr -O3

[Bug sanitizer/65610] Compare debug failure with -g3 -fsanitize=undefined -fno-sanitize=vptr -O3

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65610

--- Comment #1 from Tobias Burnus <burnus at gcc dot gnu.org> ---
Note that the "-fno-sanitize=vptr" is crucial: without it doesn't give an ICE.

[Bug tree-optimization/65610] [5 Regression] Compare debug failure with -g3 -fsanitize=undefined -fno-sanitize=vptr -O3

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65610

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hubicka at gcc dot gnu.org
          Component|sanitizer                   |tree-optimization
   Target Milestone|---                         |5.0
            Summary|Compare debug failure with  |[5 Regression] Compare
                   |-g3 -fsanitize=undefined    |debug failure with -g3
                   |-fno-sanitize=vptr -O3      |-fsanitize=undefined
                   |                            |-fno-sanitize=vptr -O3

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Seems to be a bug somewhere in ipa-polymorphic-call.c.
In particular, it seems noncall_stmt_may_be_vtbl_ptr_store returns different
results between -g0 and -g - the function apparently walks BLOCKs and
apparently with -g we have in there BLOCK_ABSTRACT_ORIGIN of some inlined dtor,
while for -g0 something has optimized the blocks away.

Honza, can you please have a look?

[Bug tree-optimization/65610] [5 Regression] Compare debug failure with -g3 -fsanitize=undefined -fno-sanitize=vptr -O3

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65610

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Perhaps one possibility would be even for -g0 preserve those specific BLOCKs
(those satisfying
    if (BLOCK_ABSTRACT_ORIGIN (block)
        && TREE_CODE (BLOCK_ABSTRACT_ORIGIN (block)) == FUNCTION_DECL)
      {
        tree fn = BLOCK_ABSTRACT_ORIGIN (block);

        if (flags_from_decl_or_type (fn) & (ECF_PURE | ECF_CONST))
          return false;
        return (TREE_CODE (TREE_TYPE (fn)) == METHOD_TYPE
                && (DECL_CXX_CONSTRUCTOR_P (fn)
                    || DECL_CXX_DESTRUCTOR_P (fn)));
      }
).

[Bug tree-optimization/65610] [5 Regression] Compare debug failure with -g3 -fsanitize=undefined -fno-sanitize=vptr -O3

[hubicka at ucw dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65610

--- Comment #4 from Jan Hubicka <hubicka at ucw dot cz> ---
> Perhaps one possibility would be even for -g0 preserve those specific BLOCKs
> (those satisfying

Yep, we should do that. Who is removing them?

[Bug tree-optimization/65610] [5 Regression] Compare debug failure with -g3 -fsanitize=undefined -fno-sanitize=vptr -O3

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65610

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Haven't debugged that part yet.
Looking at decl_maybe_in_construction_p, we'd probably need to treat functions
with DECL_ABSTRACT_ORIGIN being ctor or dtor.
I can look into this on Monday...

[Bug tree-optimization/65610] [5 Regression] Compare debug failure with -g3 -fsanitize=undefined -fno-sanitize=vptr -O3

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65610

--- Comment #6 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
i guess it is remove_unused_scope_block_p who remove the blocks.  I guess
we want to consider blocks as live when BLOCK_ABSTRACT_ORIGIN is function decl,
it is DECL_CXX_CONSTRUCTOR or DESTURCTOR and moreover it is polymorphic, i.e.
method_class_type (TREE_TYPE (fn)) has TYPE_BINFO and that binfo passes
polymorphic_type_binfo_p

We may want to make predicates polymorphic_type_ctor/dtor to test that :)

I will try to look into it tomorrow, but given the timezone, I would not mind
if you beat me :)

[Bug ipa/65610] [5 Regression] Compare debug failure with -g3 -fsanitize=undefined -fno-sanitize=vptr -O3

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65610

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 35180
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=35180&action=edit
gcc5-pr65610.patch

Untested fix.  Not at all sure about the if (!cfun->after_inlining) guard,
dunno when we can call the ipa-polymorphic-call.c stuff, if it can happen also
later on, the guard should be removed.

[Bug ipa/65610] [5 Regression] Compare debug failure with -g3 -fsanitize=undefined -fno-sanitize=vptr -O3

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65610

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Author: jakub
Date: Mon Mar 30 21:56:02 2015
New Revision: 221781

URL: https://gcc.gnu.org/viewcvs?rev=221781&root=gcc&view=rev
Log:
    PR ipa/65610
    * ipa-utils.h (inlined_polymorphic_ctor_dtor_block_p): Declare.
    * ipa-polymorphic-call.c (inlined_polymorphic_ctor_dtor_block_p): New
    function.
    (decl_maybe_in_construction_p, noncall_stmt_may_be_vtbl_ptr_store):
    Use it.
    * ipa-prop.c (param_type_may_change_p): Likewise.
    * tree-ssa-live.c: Include ipa-utils.h and its dependencies.
    (remove_unused_scope_block_p): Add in_ctor_dtor_block
    argument.  Before inlining, preserve
    inlined_polymorphic_ctor_dtor_block_p blocks and the outermost block
    with FUNCTION_DECL BLOCK_ABSTRACT_ORIGIN inside of them.  Adjust
    recursive calls.
    (remove_unused_locals): Adjust remove_unused_scope_block_p caller.

    * g++.dg/ubsan/pr65610.C: New test.

Added:
    trunk/gcc/testsuite/g++.dg/ubsan/pr65610.C
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/ipa-polymorphic-call.c
    trunk/gcc/ipa-prop.c
    trunk/gcc/ipa-utils.h
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-ssa-live.c

[Bug ipa/65610] [5 Regression] Compare debug failure with -g3 -fsanitize=undefined -fno-sanitize=vptr -O3

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65610

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

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

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed.