cyndyishida wrote: > but I'm concerned that this does not guarantee the resulting command line > only refers to inputs in the sysroot. Imagine that we run this without the > header search and VFS pruning optimizations - the command line will likely > refer to unused project-specific source/build directories.
I would think that if `ModuleDepCollector` only found dependency inputs that resolve to sysroot locations, the command line for building the specific module wouldn't matter. Of course, I would also expect that `Precompile module` invocations don't point to paths they don't need, but wouldn't be required for tracking whether a module is in sysroot. Is there a concern that the command lines (e.g. header search paths with different files populated) between scanning and compiling could result in different input files being resolved? > One way of doing this is going through the *Options classes that contribute > to CompilerInvocation and adding an API that allows for iteration through all > input/output paths. Do you have an intuition of whether this would be noticeably expensive for the scanner? https://github.com/llvm/llvm-project/pull/130634 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
