whisperity added a comment. In D69560#2487117 <https://reviews.llvm.org/D69560#2487117>, @aaron.ballman wrote:
> In D69560#2487093 <https://reviews.llvm.org/D69560#2487093>, @whisperity > wrote: > >> Not sure what CVR-modelling's default should be... it finds less when "off", >> but too easily silences crucial issues (such as `memcpy(T*, const T*)`). > > My instinct is that if you accidentally swap a qualified parameter you'll get > some other diagnostic about dropped qualifiers and so perhaps the default > should be `off`, but perhaps there's something I'm not thinking of there. That only happens if at a call site the //argument// is qualified. void foo(const int& ir1, int& ir2); void test() { int i1 = 8; int i2 = 10; foo(i2, i1); // No warning here. const int ci = 12; foo(i1, ci); // Error here. } So while `const Something` and `Something` are distinct types conceptually (and as of now, according to I.24), an unqualified `Something` can still be passed in a swapped order at any given call site. Will think about this. I'll continue implementing some of the heuristics, re-run the measurements, and then I'll update the patch with the rename. CHANGES SINCE LAST ACTION https://reviews.llvm.org/D69560/new/ https://reviews.llvm.org/D69560 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
