[Bug c++/116418] New: [11/12/13/14/15 Regression] Nested statement expressions with decltype auto in template break

[Bug c++/116418] New: [11/12/13/14/15 Regression] Nested statement expressions with decltype auto in template break

[valentin at tolmer dot fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116418

            Bug ID: 116418
           Summary: [11/12/13/14/15 Regression] Nested statement
                    expressions with decltype auto in template break
           Product: gcc
           Version: 15.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: valentin at tolmer dot fr
  Target Milestone: ---

$ cat source.cpp
template <typename>
void fn() {
  ({ decltype(auto) v = ({3;}); }); // breaks
  ({ auto v = ({3;}); });           // works
}
void fn2() {
  ({ decltype(auto) v = ({3;}); });  // works
}
$ g++ -std=c++20 source.cpp -c -O0
$ g++ -std=c++20 source.cpp -c -O1
source.cpp: In function 'void fn()':
source.cpp:3:30: error: statement-expression in a constant expression
[-Wtemplate-body]
    3 |   ({ decltype(auto) v = ({3;}); }); // breaks
      |                              ^
Compiler returned: 1

I tested it on godbolt, and it works on GCC 10.5, but not in 11.1, all the way
up to trunk. It's dependent on the optimizations, I've had some form of it
happen in our codebase in different places depending on the compiler version
(so upgrading GCC becomes a breaking change).
I tried to pinpoint the optimization that causes the diagnosis, but listing all
the flags for -O1 in https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html
doesn't trigger the issue; adding -O1 does. Similarly, adding -O1 and disabling
all these flags triggers the issue.

Note that the templated function doesn't have to be templated to trigger the
bug.

[Bug c++/116418] [12/13/14/15 Regression] Nested statement expressions with decltype auto in template break

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116418

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |12.5
            Summary|[11/12/13/14/15 Regression] |[12/13/14/15 Regression]
                   |Nested statement            |Nested statement
                   |expressions with decltype   |expressions with decltype
                   |auto in template break      |auto in template break

[Bug c++/116418] [12/13/14/15 Regression] Nested statement expressions with decltype auto in template break

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116418

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |rejects-valid
             Status|UNCONFIRMED                 |WAITING
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2024-08-20

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
The testcase isn't complete, fn() isn't instantiated so I can't reproduce the
issue.  Adding

void bar()
{
  fn<int>();
}

doesn't make it fail with GCC 11.1 for me.

[Bug c++/116418] [12/13/14/15 Regression] Nested statement expressions with decltype auto in template break

[valentin at tolmer dot fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116418

--- Comment #2 from Valentin Tolmer <valentin at tolmer dot fr> ---
Sorry, typo in the last sentence: "the templated function doesn't have to be
instantiated to trigger the bug".

See https://godbolt.org/z/3xosx5dn4 for a reproduction.

Adding an instantiation gives an additional knock-on error:

<source>: In instantiation of 'void fn() [with <template-parameter-1-1> =
int]':
<source>:10:12:   required from here
<source>:3:21: error: variable or field 'v' declared void
    3 |   ({ decltype(auto) v = ({3;});}); // breaks
      |                     ^

Richard, are you sure you compiled with -O1?

[Bug c++/116418] [12/13/14/15 Regression] statement expressions as initializer for decltype auto breaks in templates with optimization turned on

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116418

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Blocks|                            |55004
            Summary|[12/13/14/15 Regression]    |[12/13/14/15 Regression]
                   |Nested statement            |statement expressions as
                   |expressions with decltype   |initializer for decltype
                   |auto in template break      |auto breaks in templates
                   |                            |with optimization turned on
             Status|WAITING                     |NEW

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Confirmed.

Reduced testcase:
```
template <typename>
void fn() {
  decltype(auto) v = ({3;});  // breaks
}
```

Fails at -O1+. Most likely a check for optimization in the front-end trying to
optimize the initializer and using the const expression path for that.


Referenced Bugs:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55004
[Bug 55004] [meta-bug] constexpr issues

[Bug c++/116418] [12/13/14/15 Regression] statement expressions as initializer for decltype auto breaks in templates with optimization turned on

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116418

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

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

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Can't reproduce either, tried GCC 11, 12, 13, 14, trunk, -O1 -std=c++20 or -O1
-std=c++23.

[Bug c++/116418] [12/13/14/15 Regression] statement expressions as initializer for decltype auto breaks in templates with optimization turned on and debug info turned on

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116418

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[12/13/14/15 Regression]    |[12/13/14/15 Regression]
                   |statement expressions as    |statement expressions as
                   |initializer for decltype    |initializer for decltype
                   |auto breaks in templates    |auto breaks in templates
                   |with optimization turned on |with optimization turned on
                   |                            |and debug info turned on

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #4)
> Can't reproduce either, tried GCC 11, 12, 13, 14, trunk, -O1 -std=c++20 or
> -O1 -std=c++23.

Oh it is with -g too.

[Bug c++/116418] [12/13/14/15 Regression] statement expressions as initializer for decltype auto breaks in templates with optimization turned on and debug info turned on

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116418

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #5)
> (In reply to Jakub Jelinek from comment #4)
> > Can't reproduce either, tried GCC 11, 12, 13, 14, trunk, -O1 -std=c++20 or
> > -O1 -std=c++23.
> 
> Oh it is with -g too.

`-gno-statement-frontiers ` fixes the issue.

[Bug c++/116418] [12/13/14/15 Regression] statement expressions as initializer for decltype auto breaks in templates with optimization turned on and debug info turned on due to gstatement-frontiers

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116418

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

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

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Ah, with -g -O1 -std=c++20 it started with
r11-7452-g04b10596fe2fec1aa6652c15fd389087a78a235f

[Bug c++/116418] [12/13/14/15 Regression] statement expressions as initializer for decltype auto breaks in templates with optimization turned on and debug info turned on due to gstatement-frontiers

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116418

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2

[Bug c++/116418] [12/13/14/15 Regression] statement expressions as initializer for decltype auto breaks in templates with optimization turned on and debug info turned on due to gstatement-frontiers

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116418

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
void foo ();
template <typename>
void bar ()
{
  decltype(auto) v = ({ foo (); 3; });
}

doesn't need -O1 nor -g to trigger the error, and can be reproduced with
-std=c++14 and higher.  Still started with that commit, before that it has been
accepted.
strip_typedefs_expr errors on STATEMENT_LIST, which is called by the
canonicalize_type_argument call which has been added in that change.
I think decltype(auto) deduction shouldn't require a constant expression in the
initializer.

[Bug c++/116418] [12/13/14/15 Regression] statement expressions as initializer for decltype auto breaks in templates with optimization turned on and debug info turned on due to gstatement-frontiers

[ppalka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116418

Patrick Palka <ppalka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |ppalka at gcc dot gnu.org
             Status|NEW                         |ASSIGNED

[Bug c++/116418] [12/13/14/15 Regression] statement expressions as initializer for decltype auto breaks in templates with optimization turned on and debug info turned on due to gstatement-frontiers

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116418

--- Comment #9 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Patrick Palka <ppalka@gcc.gnu.org>:

https://gcc.gnu.org/g:12bdcc3d7970860b9d66ed4dea203bde8fd68d4d

commit r15-3611-g12bdcc3d7970860b9d66ed4dea203bde8fd68d4d
Author: Patrick Palka <ppalka@redhat.com>
Date:   Thu Sep 12 12:45:03 2024 -0400

    c++: decltype(auto) deduction of statement-expression [PR116418]

    r8-7538 for PR84968 made strip_typedefs_expr diagnose STATEMENT_LIST
    so that we reject statement-expressions in noexcept-specifiers to
    match our behavior in template arguments (which the parser diagnoses
    directly).

    Later r11-7452 made decltype(auto) deduction canonicalize the expression
    (as an implementation detail) which in turn calls strip_typedefs_expr,
    and so ever since we inadvertently reject decltype(auto) deduction of a
    statement-expression.

    This patch just removes the diagnostic in strip_typedefs_expr and instead
    treats statement-expressions similar to lambda-expressions.  The function
    doesn't seem like the right place for such a diagnostic and so it seems
    easier to just accept rather than try to reject them in a suitable place.

            PR c++/116418

    gcc/cp/ChangeLog:

            * tree.cc (strip_typedefs_expr) <case STATEMENT_LIST>: Replace
            this error path with ...
            <case STMT_EXPR>: ... this, returning the original tree.

    gcc/testsuite/ChangeLog:

            * g++.dg/eh/pr84968.C: No longer expect an ahead of time diagnostic
            for the statement-expresssion.  Instantiate the template and expect
            an incomplete type error instead.
            * g++.dg/ext/stmtexpr26.C: New test.

    Reviewed-by: Jason Merrill <jason@redhat.com>

[Bug c++/116418] [12/13/14/15 Regression] statement expressions as initializer for decltype auto breaks in templates with optimization turned on and debug info turned on due to gstatement-frontiers

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116418

--- Comment #10 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-14 branch has been updated by Patrick Palka
<ppalka@gcc.gnu.org>:

https://gcc.gnu.org/g:abdea396e12e66185d810435bbc363845cf4664a

commit r14-10697-gabdea396e12e66185d810435bbc363845cf4664a
Author: Patrick Palka <ppalka@redhat.com>
Date:   Thu Sep 12 12:45:03 2024 -0400

    c++: decltype(auto) deduction of statement-expression [PR116418]

    r8-7538 for PR84968 made strip_typedefs_expr diagnose STATEMENT_LIST
    so that we reject statement-expressions in noexcept-specifiers to
    match our behavior in template arguments (which the parser diagnoses
    directly).

    Later r11-7452 made decltype(auto) deduction canonicalize the expression
    (as an implementation detail) which in turn calls strip_typedefs_expr,
    and so ever since we inadvertently reject decltype(auto) deduction of a
    statement-expression.

    This patch just removes the diagnostic in strip_typedefs_expr and instead
    treats statement-expressions similar to lambda-expressions.  The function
    doesn't seem like the right place for such a diagnostic and so it seems
    easier to just accept rather than try to reject them in a suitable place.

            PR c++/116418

    gcc/cp/ChangeLog:

            * tree.cc (strip_typedefs_expr) <case STATEMENT_LIST>: Replace
            this error path with ...
            <case STMT_EXPR>: ... this, returning the original tree.

    gcc/testsuite/ChangeLog:

            * g++.dg/eh/pr84968.C: No longer expect an ahead of time diagnostic
            for the statement-expresssion.  Instantiate the template and expect
            an incomplete type error instead.
            * g++.dg/ext/stmtexpr26.C: New test.

    Reviewed-by: Jason Merrill <jason@redhat.com>
    (cherry picked from commit 12bdcc3d7970860b9d66ed4dea203bde8fd68d4d)

[Bug c++/116418] [12/13 Regression] statement expressions as initializer for decltype auto breaks in templates with optimization turned on and debug info turned on due to gstatement-frontiers

[ppalka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116418

Patrick Palka <ppalka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[12/13/14/15 Regression]    |[12/13 Regression]
                   |statement expressions as    |statement expressions as
                   |initializer for decltype    |initializer for decltype
                   |auto breaks in templates    |auto breaks in templates
                   |with optimization turned on |with optimization turned on
                   |and debug info turned on    |and debug info turned on
                   |due to gstatement-frontiers |due to gstatement-frontiers

--- Comment #11 from Patrick Palka <ppalka at gcc dot gnu.org> ---
Fixed for GCC 14.3 so far.