We currently have EmbedBitcode, which injects -fembed-bitcode into each invocation. This works pretty well on targets like macOS (where -fembed-bitcode has well-defined bitcode merging behavior), but not as well on Linux (where is just mashes bitcode streams together, creating an unparseable blob).
We should add another action, one that's roughly on par with WLLVM or GLLVM in terms of being able to extract a combined bitcode module from a binary by walking backwards through a list of referenced intermediate files.
We currently have
EmbedBitcode, which injects-fembed-bitcodeinto each invocation. This works pretty well on targets like macOS (where-fembed-bitcodehas well-defined bitcode merging behavior), but not as well on Linux (where is just mashes bitcode streams together, creating an unparseable blob).We should add another action, one that's roughly on par with WLLVM or GLLVM in terms of being able to extract a combined bitcode module from a binary by walking backwards through a list of referenced intermediate files.