Re: nasty crash

GE> Maybe I am misreading the code but I think the ++specpdl_ stuff in GE> unbind_to_hairy() is necessary because it is called from the GE> UNBIND_TO* macros. Right. GE> my interpretation of the code is: GE> - the UNBIND_TO* macros handle the normal variable case. GE> - unbind_to_hairy() is basically the old (pre-marzilla patch) GE> unbind_to() GE> - in UNBIND_TO* we first decrement the specpdl counters and then only GE> if we are not dealing with a normal variable call GE> unbind_to_hairy(). The first thing unbind_to_hairy() has to do is to GE> undo the decrement done in the macro. GE> Martin, I'd suggest to move the increments into the macros to make GE> this more obvious. But that would make the macro a nanosecond slower by making it expand to larger code. We could of course add a comment for clarity instead. GE> Ben, am I missing something here? What did you mean by 'logic error'? It's of course still possible that there's a specpdl logic error here, given that no one really understands the code. But it's probably more subtle. Maybe it has to do with the #if 0 I wrapped around the following code in unbind_to_hairy(). My assumption was that if this broke, we would find out rather quickly - and it's been more than a year now... #if 0 /* martin */ #ifndef EXCEEDINGLY_QUESTIONABLE_CODE /* There should never be anything here for us to remove. If so, it indicates a logic error in Emacs. Catches should get removed when a throw or signal occurs, or when a catch or condition-case exits normally. But it's too dangerous to just remove this code. --ben */ /* Furthermore, this code is not in FSFmacs!!! Braino on mly's part? */ /* If we're unwound past the pdlcount of a catch frame, that catch can't possibly still be valid. */ while (catchlist && catchlist->pdlcount > specpdl_depth_counter) { catchlist = catchlist->next; /* Don't mess with gcprolist, backtrace_list here */ } #endif #endif