[Bug c++/105104] New: [coroutines] ICE during GIMPLE pass: coro-early-expand-ifns

[Bug c++/105104] New: [coroutines] ICE during GIMPLE pass: coro-early-expand-ifns

[jehelset at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105104

            Bug ID: 105104
           Summary: [coroutines] ICE during GIMPLE pass:
                    coro-early-expand-ifns
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jehelset at gmail dot com
  Target Milestone: ---

Created attachment 52716
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=52716&action=edit
Output from "-fdump-tree-coro-early-expand-ifns=stdout"

Compiler segfaults on `if (EDGE_COUNT (pred->succs) <= EDGE_COUNT
(succ->preds))`, as `succ` is null. 

Backtrace:
#0  0x000000000107a2ae in find_edge (pred=0x7fffc626f068, succ=0x0) at
../../gcc/cfganal.cc:523
#1  0x0000000001864cc7 in convert_single_case_switch (swtch=0x7fffc625d2d0,
gsi=...) at ../../gcc/tree-cfgcleanup.cc:118
#2  0x0000000001864d9e in cleanup_control_expr_graph (bb=0x7fffc626f068,
gsi=...) at ../../gcc/tree-cfgcleanup.cc:145
#3  0x00000000018652be in cleanup_control_flow_bb (bb=0x7fffc626f068) at
../../gcc/tree-cfgcleanup.cc:274
#4  0x000000000186712f in cleanup_control_flow_pre () at
../../gcc/tree-cfgcleanup.cc:908
#5  0x00000000018678f2 in cleanup_tree_cfg_noloop (ssa_update_flags=0) at
../../gcc/tree-cfgcleanup.cc:1073
#6  0x0000000001867bbb in cleanup_tree_cfg (ssa_update_flags=0) at
../../gcc/tree-cfgcleanup.cc:1183
#7  0x00000000016893c6 in execute_function_todo (fn=0x7fffcf0580b8, data=0x60)
at ../../gcc/passes.cc:2051
#8  0x00000000016881d9 in do_per_function (callback=0x1689361
<execute_function_todo(function*, void*)>, data=0x60) at
../../gcc/passes.cc:1688
#9  0x000000000168972e in execute_todo (flags=96) at ../../gcc/passes.cc:2139
#10 0x000000000168a838 in execute_one_pass (pass=0x408fb20) at
../../gcc/passes.cc:2675
#11 0x000000000168aa9b in execute_pass_list_1 (pass=0x408fb20) at
../../gcc/passes.cc:2738
#12 0x000000000168ab25 in execute_pass_list (fn=0x7fffcf0580b8, pass=0x408f6a0)
at ../../gcc/passes.cc:2749
#13 0x00000000010f6cbf in cgraph_node::analyze (this=0x7fffcf07d880) at
../../gcc/cgraphunit.cc:685
#14 0x00000000010f8daa in analyze_functions (first_time=true) at
../../gcc/cgraphunit.cc:1240
#15 0x00000000010fc094 in symbol_table::finalize_compilation_unit
(this=0x7ffff70c6000) at ../../gcc/cgraphunit.cc:2500
#16 0x00000000017eb979 in compile_file () at ../../gcc/toplev.cc:479
#17 0x00000000017eea61 in do_compile (no_backend=false) at
../../gcc/toplev.cc:2168
#18 0x00000000017eee49 in toplev::main (this=0x7fffffffdf62, argc=33,
argv=0x7fffffffe098) at ../../gcc/toplev.cc:2320
#19 0x0000000002fd18b1 in main (argc=33, argv=0x7fffffffe098) at
../../gcc/main.cc:39

Dump from tree-coro-early-expand-ifns put in attachment, because of
template-spew.

Compiled with e3d2b0d040e9baf6c0548b865ed5244dec464cc1.

[Bug c++/105104] [coroutines] ICE during GIMPLE pass: coro-early-expand-ifns

[jehelset at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105104

--- Comment #1 from John Eivind Helset <jehelset at gmail dot com> ---
Indices of `bb` and `case_bb` in `convert_single_case_switch` are: 
(gdb) p bb->index
$18 = 154
(gdb) p case_bb->index
$19 = 157

If that helps.

[Bug c++/105104] [coroutines] ICE during GIMPLE pass: coro-early-expand-ifns

[jehelset at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105104

--- Comment #2 from John Eivind Helset <jehelset at gmail dot com> ---
I'll try to create an MRE eventually...

[Bug c++/105104] [coroutines] ICE during GIMPLE pass: coro-early-expand-ifns

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105104

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2022-03-30
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |WAITING

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
Please provide preprocessed source of a testcase.

[Bug c++/105104] [coroutines] ICE during GIMPLE pass: coro-early-expand-ifns

[jehelset at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105104

--- Comment #4 from John Eivind Helset <jehelset at gmail dot com> ---
It seems a non-void return-type from await-resume of a final awaitable,
combined with at least -O1 causes a segfault: https://godbolt.org/z/rzq8dM7Pr

Not sure if it's the same as the one I initially encountered yet.

[Bug c++/105104] [coroutines] ICE during GIMPLE pass: coro-early-expand-ifns

[jehelset at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105104

--- Comment #5 from John Eivind Helset <jehelset at gmail dot com> ---
Created attachment 52729
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=52729&action=edit
Patch with testcase.

Tried adding a testcase to the g++.dg/coroutines testsuite. Used dg-ice, but it
seems to complain about excess errors anyways:

FAIL: g++.dg/coroutines/pr105104.C (test for excess errors)
Excess errors:
during GIMPLE pass: coro-early-expand-ifns

[Bug c++/105104] [coroutines] ICE during GIMPLE pass: coro-early-expand-ifns

[iains at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105104

Iain Sandoe <iains at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|2022-03-30 00:00:00         |2022-4-18
                 CC|                            |iains at gcc dot gnu.org,
                   |                            |jakub at gcc dot gnu.org
             Status|WAITING                     |NEW
           Keywords|                            |ice-on-valid-code

--- Comment #6 from Iain Sandoe <iains at gcc dot gnu.org> ---
Initial analysis - perhaps some issue with constexpr processing:

When we have a void await_return() in  the final_awaiter class:

 final.suspend:
 frame_ptr->_Coro_resume_fn = 0B;
 {
 struct final_awaiter Fs [value-expr: frame_ptr->Fs_1_2];

 frame_ptr->_Coro_resume_index = 4;
 _15 = &frame_ptr->_Coro_self_handle;
 D.9844 = std::__n4861::coroutine_handle<return_object::promise_type>::operator
std::__n4861::coroutine_handle<void> (_15);
 return_object::promise_type::final_awaiter::await_suspend (D.9844);
 D.9768 = .CO_YIELD (4, 1, &resume.4, &destroy.4, frame_ptr);
 retval.2 = D.9768;
 switch (retval.2) <default: <D.9771>, case 0: <D.9769>, case 1: <D.9770>>
 <D.9769>:
 .CO_SUSPN (&actor.suspend.ret);
 <D.9770>:
 goto resume.4;
 <D.9771>:
 goto destroy.4;
 destroy.4:
 goto coro.delete.promise;
 resume.4:
 return_object::promise_type::final_awaiter::await_resume ();
 }

when the await_resume() function is made to return an int.

 final.suspend:
 frame_ptr->_Coro_resume_fn = 0B;
 {
<snip>
 destroy.4:
 goto coro.delete.promise;
 resume.4:
 }

So we have a dangling label at the end of that scope (which then gives rise to
the crash).

* If I remove the 'contexpr' from the await_resume() function, then it all
works as expected.

* AFAICT, the content is correct in "expand_one_await_expression()".

* Altering the coroutines code to cast the result of the await_resume() to void
makes no difference.