[Bug c++/108887] [13 Regression] ICE in process_function_and_variable_attributes since r13-3601

["jakub at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108887

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The ICE is actually in cgraph code, so it might as well be just some latent
cgraph bug triggered by the C++ changes.
What I see is that first_analyzed is set to a cgraph node for
_ZZN27LinkPaginationTimelineModel12fillTimelineEvENKUlP15AbstractAccountE0_clES1_
But that node is then removed in:
#0  0x00007ffff7a4ec8f in __memset_evex_unaligned_erms () from /lib64/libc.so.6
#1  0x000000000094ec0b in ggc_free (p=0x7fffe9e93bb0) at
../../gcc/ggc-page.cc:1630
#2  0x0000000000a24fb6 in symbol_table::release_symbol (this=0x7fffea130000,
node=<error reading variable: Cannot access memory at address
0xa5a5a5a5a5a5a5c5>)
    at ../../gcc/cgraph.h:2853
#3  0x0000000000a1d498 in cgraph_node::remove (this=<error reading variable:
Cannot access memory at address 0xa5a5a5a5a5a5a5c5>) at
../../gcc/cgraph.cc:1974
#4  0x0000000000a08369 in symtab_node::remove (this=<error reading variable:
Cannot access memory at address 0xa5a5a5a5a5a5a5c5>) at ../../gcc/symtab.cc:481
#5  0x00000000005866e4 in record_mangling (decl=<function_decl 0x7fffea319200
operator()>, need_warning=true) at ../../gcc/cp/decl2.cc:4751
#6  0x00000000005f5d5d in mangle_decl (decl=<function_decl 0x7fffea319200
operator()>) at ../../gcc/cp/mangle.cc:4196
#7  0x0000000001456c70 in decl_assembler_name (decl=<function_decl
0x7fffea319200 operator()>) at ../../gcc/tree.cc:743
#8  0x0000000001457234 in assign_assembler_name_if_needed (t=<function_decl
0x7fffea319200 operator()>) at ../../gcc/tree.cc:858
#9  0x0000000000a2cf4d in cgraph_node::analyze (this=<cgraph_node * const
0x7fffea144990 "operator()"/9>) at ../../gcc/cgraphunit.cc:669
#10 0x0000000000a2effe in analyze_functions (first_time=true) at
../../gcc/cgraphunit.cc:1238
but nothing updates first_analyzed.

record_mangling has:
  /* If this is already an alias, remove the alias, because the real
     decl takes precedence.  */
  if (*slot && DECL_ARTIFICIAL (*slot) && DECL_IGNORED_P (*slot))
    if (symtab_node *n = symtab_node::get (*slot))
      if (n->cpp_implicit_alias)
        {
          n->remove ();
          *slot = NULL_TREE;
        }

So, to some extent this is related to PR107897.
The ICE is obviously gone with -fabi-compat-version=0 and so if we wouldn't
emit compatibility aliases for lambdas based on PR107897, this issue would be
latent again.