[Bug debug/94459] New: Missing c++ debug information for 'auto&' return type

[Bug debug/94459] New: Missing c++ debug information for 'auto&' return type

[ssbssa at yahoo dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94459

            Bug ID: 94459
           Summary: Missing c++ debug information for 'auto&' return type
           Product: gcc
           Version: 9.3.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: debug
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ssbssa at yahoo dot de
  Target Milestone: ---

For the following example:

template <typename T>
struct MyClass
{
  T value;
  auto get()
  {
    return value;
  }
  auto &get_ref()
  {
    return value;
  }
  auto &&get_r_ref()
  {
    return (T&&)value;
  }
  const auto c_get()
  {
    return value;
  }
  const auto &c_get_ref()
  {
    return value;
  }
  const auto &&c_get_r_ref()
  {
    return (T&&)value;
  }
};

int main()
{
  MyClass<int> mc{1};
  return (mc.get() + mc.get_ref() + mc.get_r_ref()
      + mc.c_get() + mc.c_get_ref() + mc.c_get_r_ref());
}

For the simple 'auto' case, the debugger knows the real return type:

(gdb) pt MyClass<int>::get
type = int (MyClass<int> * const)

But not for the others:

(gdb) pt MyClass<int>::get_ref
type = void &(MyClass<int> * const)
(gdb) pt MyClass<int>::get_r_ref
type = void &&(MyClass<int> * const)
(gdb) pt MyClass<int>::c_get
type = void (MyClass<int> * const)
(gdb) pt MyClass<int>::c_get_ref
type = void &(MyClass<int> * const)
(gdb) pt MyClass<int>::c_get_r_ref
type = void &&(MyClass<int> * const)

For 'auto', extra type information is stored on the definition die, this was
implemented here:
https://gcc.gnu.org/git/?p=gcc.git;a=commitdiff;h=2e5e7103a39315664f9a625bea42981f5251c27e


So I came up with the following patch:

--- gcc/dwarf2out.c     2020-03-12 12:07:21.000000000 +0100
+++ gcc/dwarf2out.c     2020-04-02 14:54:03.680451100 +0200
@@ -23016,6 +23016,11 @@
          if (is_cxx () && debug_info_level > DINFO_LEVEL_TERSE)
            {
              dw_die_ref die = get_AT_ref (old_die, DW_AT_type);
+             if (die->die_tag == DW_TAG_reference_type
+                 || die->die_tag == DW_TAG_rvalue_reference_type)
+               die = get_AT_ref (die, DW_AT_type);
+             if (die->die_tag == DW_TAG_const_type)
+               die = get_AT_ref (die, DW_AT_type);
              if (die == auto_die || die == decltype_auto_die)
                add_type_attribute (subr_die, TREE_TYPE (TREE_TYPE (decl)),
                                    TYPE_UNQUALIFIED, false, context_die);

This seems to work fine:

(gdb) pt MyClass<int>::get
type = int (MyClass<int> * const)
(gdb) pt MyClass<int>::get_ref
type = int &(MyClass<int> * const)
(gdb) pt MyClass<int>::get_r_ref
type = int &&(MyClass<int> * const)
(gdb) pt MyClass<int>::c_get
type = const int (MyClass<int> * const)
(gdb) pt MyClass<int>::c_get_ref
type = const int &(MyClass<int> * const)
(gdb) pt MyClass<int>::c_get_r_ref
type = const int &&(MyClass<int> * const)

But I can't run the testsuite (I'm on Windows), so I can't be sure that this
would break anything else.

[Bug debug/94459] Missing c++ debug information for 'auto&' return type

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94459

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

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

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Looking through DW_TAG_const_type seems insufficient to me, can't there be
other qualifications (at least DW_TAG_volatile_type, perhaps in various
orders)?
So wouldn't be better a loop as long as die and die->die_tag is one of
dwarf_qual_info[?].t or DW_TAG_{rvalue_}reference, use die = get_AT_ref (die,
DW_AT_type); ?

[Bug debug/94459] Missing c++ debug information for 'auto&' return type

[ssbssa at yahoo dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94459

--- Comment #2 from Domani Hannes <ssbssa at yahoo dot de> ---
(In reply to Jakub Jelinek from comment #1)
> Looking through DW_TAG_const_type seems insufficient to me, can't there be
> other qualifications (at least DW_TAG_volatile_type, perhaps in various
> orders)?
> So wouldn't be better a loop as long as die and die->die_tag is one of
> dwarf_qual_info[?].t or DW_TAG_{rvalue_}reference, use die = get_AT_ref
> (die, DW_AT_type); ?

You're probably right, I wasn't sure if the use of 'auto' allows e.g.
'volatile' as well, so I just stuck with the ones I knew about.

[Bug debug/94459] Missing c++ debug information for 'auto&' return type

[ssbssa at yahoo dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94459

--- Comment #3 from Domani Hannes <ssbssa at yahoo dot de> ---
(In reply to Jakub Jelinek from comment #1)
> Looking through DW_TAG_const_type seems insufficient to me, can't there be
> other qualifications (at least DW_TAG_volatile_type, perhaps in various
> orders)?
> So wouldn't be better a loop as long as die and die->die_tag is one of
> dwarf_qual_info[?].t or DW_TAG_{rvalue_}reference, use die = get_AT_ref
> (die, DW_AT_type); ?

As suggested, I'm now using this:

--- gcc/dwarf2out.c     2020-03-12 12:07:21.000000000 +0100
+++ gcc/dwarf2out.c     2020-04-02 20:10:05.829023900 +0200
@@ -23016,6 +23016,11 @@
          if (is_cxx () && debug_info_level > DINFO_LEVEL_TERSE)
            {
              dw_die_ref die = get_AT_ref (old_die, DW_AT_type);
+             while (die->die_tag == DW_TAG_reference_type
+                    || die->die_tag == DW_TAG_rvalue_reference_type
+                    || die->die_tag == DW_TAG_const_type
+                    || die->die_tag == DW_TAG_volatile_type)
+               die = get_AT_ref (die, DW_AT_type);
              if (die == auto_die || die == decltype_auto_die)
                add_type_attribute (subr_die, TREE_TYPE (TREE_TYPE (decl)),
                                    TYPE_UNQUALIFIED, false, context_die);

I don't think that restrict/atomic is possible for auto return types, but I may
be wrong about that.

[Bug debug/94459] Missing c++ debug information for 'auto&' return type

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94459

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2020-04-03
             Status|UNCONFIRMED                 |ASSIGNED
     Ever confirmed|0                           |1
           Assignee|unassigned at gcc dot gnu.org      |jakub at gcc dot gnu.org

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 48191
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=48191&action=edit
gcc10-pr94459.patch

Full patch I'll be testing.  Related to PR53756.

[Bug debug/94459] Missing c++ debug information for 'auto&' return type

[ssbssa at yahoo dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94459

--- Comment #5 from Hannes Domani <ssbssa at yahoo dot de> ---
I also should mention that this is a precursor to fix gdb_bug 24154:
https://sourceware.org/bugzilla/show_bug.cgi?id=24154

[Bug debug/94459] Missing c++ debug information for 'auto&' return type

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94459

--- Comment #6 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jakub Jelinek <jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:0be9efad938726721fd8c8c35609b1e1d7d30035

commit r10-7560-g0be9efad938726721fd8c8c35609b1e1d7d30035
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Sun Apr 5 00:28:28 2020 +0200

    debug: Improve debug info of c++14 deduced return type [PR94459]

    On the following testcase, in gdb ptype S<long>::m1 prints long as return
    type, but all the other methods show void instead.
    PR53756 added code to add_type_attribute if the return type is
    auto/decltype(auto), but we actually should look through references,
    pointers and qualifiers.
    Haven't included there DW_TAG_atomic_type, because I think at least ATM
    one can't use that in C++.  Not sure about DW_TAG_array_type or what else
    could be deduced.

    > http://eel.is/c++draft/dcl.spec.auto#3 says it has to appear as a
    > decl-specifier.
    >
    > http://eel.is/c++draft/temp.deduct.type#8 lists the forms where a
template
    > argument can be deduced.
    >
    > Looks like you are missing arrays, pointers to members, and function
return
    > types.

    2020-04-04  Hannes Domani  <ssbssa@yahoo.de>
                Jakub Jelinek  <jakub@redhat.com>

            PR debug/94459
            * dwarf2out.c (gen_subprogram_die): Look through references,
pointers,
            arrays, pointer-to-members, function types and qualifiers when
            checking if in-class DIE had an 'auto' or 'decltype(auto)' return
type
            to emit type again on definition.

            * g++.dg/debug/pr94459.C: New test.

    Co-Authored-By: Hannes Domani <ssbssa@yahoo.de>

[Bug debug/94459] Missing c++ debug information for 'auto&' return type

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94459

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed on the trunk so far.

[Bug debug/94459] Missing c++ debug information for 'auto&' return type

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94459

--- Comment #8 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-9 branch has been updated by Jakub Jelinek
<jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:b5039b7259e64a92f5c077fe4d023556d6b12550

commit r9-8479-gb5039b7259e64a92f5c077fe4d023556d6b12550
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Tue Apr 7 21:01:40 2020 +0200

    debug: Improve debug info of c++14 deduced return type [PR94459]

    On the following testcase, in gdb ptype S<long>::m1 prints long as return
    type, but all the other methods show void instead.
    PR53756 added code to add_type_attribute if the return type is
    auto/decltype(auto), but we actually should look through references,
    pointers and qualifiers.
    Haven't included there DW_TAG_atomic_type, because I think at least ATM
    one can't use that in C++.  Not sure about DW_TAG_array_type or what else
    could be deduced.

    > http://eel.is/c++draft/dcl.spec.auto#3 says it has to appear as a
    > decl-specifier.
    >
    > http://eel.is/c++draft/temp.deduct.type#8 lists the forms where a
template
    > argument can be deduced.
    >
    > Looks like you are missing arrays, pointers to members, and function
return
    > types.

    2020-04-04  Hannes Domani  <ssbssa@yahoo.de>
                Jakub Jelinek  <jakub@redhat.com>

            PR debug/94459
            * dwarf2out.c (gen_subprogram_die): Look through references,
pointers,
            arrays, pointer-to-members, function types and qualifiers when
            checking if in-class DIE had an 'auto' or 'decltype(auto)' return
type
            to emit type again on definition.

            * g++.dg/debug/pr94459.C: New test.

    Co-Authored-By: Hannes Domani <ssbssa@yahoo.de>

[Bug debug/94459] Missing c++ debug information for 'auto&' return type

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94459

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

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

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed for 8.5 in r8-10481-gbd748b6b448ef47f56922ec67244037915f721c2 and by the
above commit for 9.4+ too.