https://gcc.gnu.org/bugzilla/show_bug.cgi?id=44677
--- Comment #21 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Created attachment 61171 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=61171&action=edit gcc16-pr44677-wip.patch So, I tried to implement this partially, so far in the C FE. As we've been implementing the current behavior for more than 14 years, I think users should be able to specify what exactly they want in this area, so the WIP patch currently has 3 levels, level 1 is what is currently implemented, level 2 doesn't consider pre/post inc/decrements as reads in addition to that and level 3 in addition doesn't consider x += y as read of x unless x is referenced from inside of y. Both of these were pretty simple (of course need testsuite coverage). The question is if we could also handle x = x + y or x = y + x similarly (whether in level 3 or new level 4), there the mark_exp_read already happens quite early and not sure how hard it would be to have an exception for the lhs (but only to the outermost operation, perhaps wrapped in ()s, but not more levels. And perhaps only if there is no cast needed. Anyway, I think I should first try to implement what the patch does in C in C++ if it isn't too hard and only then think about further changes.
