VitaNuo wrote: Thank you for the review @ChuanqiXu9!
> (1) Hardcoding is generally not good. And even if we have to, e.g., (we don't > have an automatic mechanism to check if we need to defer an attribute), let's > avoid hardcoding in both reader side and writer side. We can make this by > adding a new bit in the record of attribute to indicate if we need to delay > it or not in the reader side generally. Added a `DeferDeserialization` bit to `Attr` class that is set to `true` for the `preferred_name` and to `false` for all other attributes. Depending on this field, the skip count field is set on the record (to the count of record fields to skip on initial read for the `preferred_name`, and to 0 otherwise). > (2) Then I think we can keep the signature of readAttr as is, where it will > always skip the check for the delay bit and always try to read it. Then we > can add a readOrDeferFor(Decl *D) method as you did. The readOrDeferFor will > try to read the deferring bits directly and if it is true, it will record the > length skipCount and D to PendingDeferredAttributes . Otherwise, it will call > readAttr. I think it is more clear. Yes this makes sense. It seems like one field (skip count) should be enough. https://github.com/llvm/llvm-project/pull/122726 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
