[PATCH] D29110: LangRef: Document the allowed metadata dropping transforms.

[Mehdi AMINI via Phabricator via llvm-commits]

mehdi_amini added a comment.

Another analogy is instructions vs intrinsics: the former are first class and need specific handling in every pass to understand the semantic of the instruction, while the latter can be opaque because we provide a mechanism of generic "property" (through attributes), and only these attributes / properties needs to be understood by transformations/analyses. Metadata are opaque the same way as intrinsics, but there is no generic mechanism to attach any "semantic property" to them.


https://reviews.llvm.org/D29110