https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102409
--- Comment #2 from Tobias Burnus <burnus at gcc dot gnu.org> --- I wonder whether the following causes that it goes wrong (libcpp/macro.c): /* Push the context of a macro with hash entry NODE onto the context stack. If we can successfully expand the macro, we push a context containing its yet-to-be-rescanned replacement list and return one. If there were additionally any unexpanded deferred #pragma directives among macro arguments, push another context containing the pragma tokens before the yet-to-be-rescanned replacement list and return two. Otherwise, we don't push a context and return zero. LOCATION is the location of the expansion point of the macro. */ static int enter_macro_context (cpp_reader *pfile, cpp_hashnode *node, * * * If I do the following change, it works for the small/large testcase: --- a/libcpp/macro.c +++ b/libcpp/macro.c @@ -1262 +1262 @@ collect_args (cpp_reader *pfile, const cpp_hashnode *node, - else if (token->type == CPP_PRAGMA) + else if (false) But that gives an ICE for gcc.dg/cpp/pr34692.c. I note that both here and for the old cpp testcase, pfile->state.prevent_expansion = 1 and parsing_args = 2. Thus, the question is how to determine when to expand it and when not.
