RewriteRule's applyFirst was brittle with respect to the default setting of the TraversalKind (revealed by D72534). This patch builds awareness of traversal kinds directly into rewrite rules so that they are insensitive to any changes in defaults.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
It might make sense for now (in order to unblock you) to make the Transformer library explicitly require the AsIs mode. I am not so familiar with the transformer library, but I think you can do that by adding traverse(AsIs, ...) in Transformer::registerMatchers.
The real fix is probably in adjusting transformer::detail::buildMatchers. I can look into that in a few hours.
Thanks for the suggestion - I agree that changing registerMatchers would be a short term fix, but I think its better to just fix buildMatchers directly. I'm working on a patch now, so need to look into it (though thanks for the offer!)
clang/lib/Tooling/Transformer/RewriteRule.cpp | ||
---|---|---|
154 | No. Abortive effort to support pulling it from the context. However, turned out the context is not available when this is called. Moreover, I've decided it's the wrong UX altogether. Matchers are not polymorphic (usually) -- they only make sense with respect to a single TK, whether explicit or implicit. So, my goal is that RewriteRules will specify how "open" matchers are interpreted -- that is, the matchers TK will be set at rule creation time rather than at rule execution time (by pulling from the context). I don't achieve that goal in this revision because I'm only updating applyFirst, but I should follow up with changes makeRule and the corresponding docs to make this clear.` That said, I chose TK_IgnoreUnlessSpelledInSource as the default setting for "open" matchers to be consistent with the consensus that its worth trying this out as a default. |
I don't see any callers using the new argument. Do we need this flexibility?