https://github.com/haoNoQ commented:
So from the user's perspective we're checking the flag twice: once at the `}` to see if we need to ramp up the analysis, then again at the diagnostic location to see if the diagnostic should be discarded. The warning will only be emitted if it's enabled at both source locations. I like this. I think this is a very good practical tradeoff. It behaves very counter-intuitively if you try to enable the diagnostic at one line with a pragma but it doesn't show up until you wrap your entire function in a pragma. But this is a very unlikely situation to occur naturally. Even if you do use the pragma to enable the diagnostic, why would you minimize the scope this way? (I also really like @cor3ntin's idea of emitting an explanatory warning if we can detect this case.) The opposite situation, where the diagnostic is disabled on an individual line but enabled globally, is significantly more likely. And it is also the one that acts in an intuitive manner! @ziqingluo-90 @jkorous-apple You might want to take a look at how this interacts with `-Wunsafe-buffer-usage`. If you still need to analyze the entire TU at once, the approach proposed in this patch probably won't work for you. https://github.com/llvm/llvm-project/pull/136323 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
