[Bug plugins/62252] a callback to event PLUGIN_FINISH_TYPE segfaults

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62252

Andrew Pinski  changed:

   What|Removed |Added

   Severity|normal  |enhancement

[Bug plugins/62252] a callback to event PLUGIN_FINISH_TYPE segfaults

[klemen.jan.enova at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62252

klemen.jan.enova at gmail dot com changed:

   What|Removed |Added

 CC||klemen.jan.enova at gmail dot 
com

--- Comment #4 from klemen.jan.enova at gmail dot com ---
Created attachment 33391
  -- https://gcc.gnu.org/bugzilla/attachment.cgi?id=33391action=edit
cp-parser-patch

Fixes the bug. If I use debug_tree(type) in handle_struct(), it prints a
RECORD_TYPE only once, because the type declaration and definition happens only
once, other uses of the identifier struct S are variable declarations.

[Bug plugins/62252] a callback to event PLUGIN_FINISH_TYPE segfaults

[klemen.jan.enova at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62252

--- Comment #5 from klemen.jan.enova at gmail dot com ---
The build passed make check, but I said invoke_plugin_callback() should be
before cp_parser_set_decl_spec_type(), so I will try that way too.

[Bug plugins/62252] a callback to event PLUGIN_FINISH_TYPE segfaults

[manu at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62252

Manuel López-Ibáñez manu at gcc dot gnu.org changed:

   What|Removed |Added

 Status|UNCONFIRMED |WAITING
   Last reconfirmed||2014-08-24
 CC||manu at gcc dot gnu.org
 Ever confirmed|0   |1

--- Comment #1 from Manuel López-Ibáñez manu at gcc dot gnu.org ---
It seems it is your plugin crashing. Have you tried debugging why it crashes?

0x7f520d010780 handle_struct(void*, void*)
/home/kje/example/example.c:20
0x8fb0b9 invoke_plugin_callbacks_full(int, void*)

[Bug plugins/62252] a callback to event PLUGIN_FINISH_TYPE segfaults

[klemen.jan.enova at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62252

--- Comment #2 from klemen.jan.enova at gmail dot com ---
TREE_CODE(type) returns an ERROR_MARK. So, (tree) event_data must not alias
with a tree. This plugin is from
https://github.com/gcc-mirror/gcc/blob/master/gcc/testsuite/g%2B%2B.dg/plugin/dumb_plugin.c.
There are no checks there, it just assumes it is a type represented as a tree.

[Bug plugins/62252] a callback to event PLUGIN_FINISH_TYPE segfaults

[klemen.jan.enova at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62252

--- Comment #3 from klemen.jan.enova at gmail dot com ---
It works on gcc, though.

in gcc/c/c-parser.c:

 if (!typespec_ok)
   goto out;


 invoke_plugin_callbacks (PLUGIN_FINISH_TYPE, t.spec);
 declspecs_add_type (loc, specs, t);

in gcc/cp/parser.c:

  type_spec = cp_parser_class_specifier (parser);
  invoke_plugin_callbacks (PLUGIN_FINISH_TYPE, type_spec);
  /* If that worked, we're done.  */
  if (cp_parser_parse_definitely (parser))
   {
 if (declares_class_or_enum)
   *declares_class_or_enum = 2;
 if (decl_specs)
   cp_parser_set_decl_spec_type (decl_specs,
 type_spec,
 token,
 /*type_definition_p=*/true);
 return type_spec;
   }


I see a pattern. The call to invoke_plugin_callbacks() should probably be
before something_that_adds_the_type_to_decl_spec(). Also, there is a test in
both cases (typespec_ok VS cp_parser_parse_definitely). I will try to compile
gcc with this patch, and see what comes out.