Michael137 wrote: > > Another possible solution: Use two `NonEquivalentDecls` sets, one for > > `IgnoreTemplateParmDepth = true` and one for `false`. This may use less > > memory (no rarely used third value in the key) but requires more code > > changes. > > I like this idea, and I think it can be implemented without significant code > changes -- see my inline comments for details. > > (Disclaimer: I didn't check my change so there might be other locations where > you didn't need to change between the old "set of pairs" and your "set of > 3-tuples", but the suggested "two sets of pairs in an array" would behave > differently -- but it would be a simplification even if you needed one or two > lines of additional logic.) > > However the current "set of 3-tuples" approach is also OK if you prefer it.
One thing I don't like about having two separate caches is the prospect of what this looks like if we add another parameter affecting the equivalence. For example, we already have another parameter `StrictTypeSpelling`, which by the looks of it could affect the result of an equivalence check. That just doesn't feel like a future-proof solution. Encoding the flag that affects that hash of the cache entry in the entry itself (like @balazske did in the first iteration of this PR) would be my preferred approach. https://github.com/llvm/llvm-project/pull/115518 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
