[Bug libstdc++/95200] New: user-defined hash function is not copied correctly if unordered_map is declared using an incomplete type

[Bug libstdc++/95200] New: user-defined hash function is not copied correctly if unordered_map is declared using an incomplete type

[jevgenijsz1 at verifone dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95200

            Bug ID: 95200
           Summary: user-defined hash function is not copied correctly if
                    unordered_map is declared using an incomplete type
           Product: gcc
           Version: 9.3.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: libstdc++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jevgenijsz1 at verifone dot com
  Target Milestone: ---

Created attachment 48558
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=48558&action=edit
g++ -v output

The failing code requires at least 3 source files, I did not find a cleaner way
to reproduce the issue. I'm not 100% sure this is a bug in libstdc++ however I
couldn't find a reason for why this should not work as expected.

Code:
=====
map_obj.h
=====

#ifndef MAP_OBJ_H_
#define MAP_OBJ_H_

#include <unordered_map>

enum class EnumType;

typedef std::unordered_map<EnumType, bool> map;

class MapObj {
 public:
  map getEnumMap() const;
  map enum_map_;
};

#endif
=====

=====
map_obj.cpp
=====
#include "map_obj.h"

map MapObj::getEnumMap() const {
  return enum_map_;
}
=====

=====
main.cpp
=====
#include <functional>

enum class EnumType : int {
  VALUE0,
  VALUE1,
};

namespace std {
template <>
struct hash<EnumType> {
  size_t operator()(EnumType type) const { return
std::hash<int>()(static_cast<int>(type)); }
};
}

#include "map_obj.h"

int main() {
  auto event = MapObj();
  event.enum_map_[EnumType::VALUE0] = true;
  event.enum_map_[EnumType::VALUE1] = false;
  event.getEnumMap().at(EnumType::VALUE0);
  event.getEnumMap().at(EnumType::VALUE1);
  return 0;
}
=====

using: 
gcc version 9.3.1 20200408 (Red Hat 9.3.1-2) (GCC)
libstdc++ 3.4.28

command line to build: g++ -save-temps -Wall -Wextra -o map_test main.cpp
map_obj.cpp

expected result: runnnig map_test returns 0

actual result: terminate called after throwing an instance of
'std::out_of_range'
  what():  _Map_base::at

[Bug libstdc++/95200] user-defined hash function is not copied correctly if unordered_map is declared using an incomplete type

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95200

Jonathan Wakely <redi at gcc dot gnu.org> changed:

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

--- Comment #1 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to jevgenijsz1 from comment #0)
> however I couldn't find a reason for why this should not work as expected.

[res.on.functions] paragraph 2:

In particular, the effects are undefined in the following cases:
- [...]
- If an incomplete type ([basic.types]) is used as a template argument when
instantiating a template component or evaluating a concept, unless specifically
allowed for that component.

[Bug libstdc++/95200] user-defined hash function is not copied correctly if unordered_map is declared using an incomplete type

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95200

--- Comment #2 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Although the actual cause of the behaviour you see is due to violating a
different requirement. The explicit specialization hash<EnumType> is not
declared in map_obj.h which means it gets implicitly instantiated using the
default definition of std::hash for enumeration types. An explicit
specialization needs to be declared before any use that would cause the
implicit instantiation to happen.

[Bug libstdc++/95200] user-defined hash function is not copied correctly if unordered_map is declared using an incomplete type

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95200

--- Comment #3 from Jonathan Wakely <redi at gcc dot gnu.org> ---
That's [temp.expl.spec] paragraph 7:

If a template, a member template or a member of a class template is explicitly
specialized then that specialization shall be declared before the first use of
that specialization that would cause an implicit instantiation to take place,
in every translation unit in which such a use occurs; no diagnostic is
required.

[Bug libstdc++/95200] user-defined hash function is not copied correctly if unordered_map is declared using an incomplete type

[jevgenijsz1 at verifone dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95200

--- Comment #4 from jevgenijsz1 at verifone dot com ---
Jonathan,

Thank you for looking into this. I guess what confused me was copy assignment
requirements in 26.2.7 that explicitly say hash function is copied and the hash
function is available at the time of copy

[Bug libstdc++/95200] user-defined hash function is not copied correctly if unordered_map is declared using an incomplete type

[jevgenijsz1 at verifone dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95200

--- Comment #5 from jevgenijsz1 at verifone dot com ---
Moreover if I was to put a breakpoint in struct hash<EnumType> I can see that
it is being used 4 times in the code example posted: twice to hash the EnumType
on insertion and twice on access (however the second access attempt throws an
exception)

[Bug libstdc++/95200] user-defined hash function is not copied correctly if unordered_map is declared using an incomplete type

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95200

--- Comment #6 from Jonathan Wakely <redi at gcc dot gnu.org> ---
It's undefined behaviour so anything can happen.