ldionne added a comment. In D17053#2315384 <https://reviews.llvm.org/D17053#2315384>, @dexonsmith wrote:
> In D17053#632700 <https://reviews.llvm.org/D17053#632700>, @EricWF wrote: > >> Maybe if we want to improve the failure mode we can add a >> `_LIBCPP_ASSERT(__new_last <= __end, "invalid range")`? > > I suspect this assertion would get optimized out, since if `__new_last > > __end` it's undefined behaviour to compare them. Whereas the loop condition > won't get optimized away. > > @ldionne, pointing you at this in case you have an idea (maybe specializing > for raw pointers?), but I'm not planning to move this forward. Thanks for the heads up. We already have `_LIBCPP_ASSERT(!empty(), "vector::pop_back called for empty vector");` in `vector::pop_back()`, but it doesn't trigger because assertions in libc++ are tied to whether the debug mode is enabled, and by default, it's not enabled at all. I believe that fixing this instead is the right way to go, since we'll get this improvement but also several additional assertions that are already in place in libc++. It is on my roadmap to improve that situation. CHANGES SINCE LAST ACTION https://reviews.llvm.org/D17053/new/ https://reviews.llvm.org/D17053 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
