[PATCH] D15225: [Driver] Sanitizer support based on runtime library presence

Reid Kleckner via Phabricator via cfe-commits cfe-commits at lists.llvm.org
Tue Jul 31 13:38:15 PDT 2018

rnk added a comment.

In https://reviews.llvm.org/D15225#1183132, @george.karpenkov wrote:

> @rnk could you clarify how did it break the distributed asanified build? If the slave compiler supports asan it should have this runtime, right?

Most open source distributed build systems wrap only compilation steps, not link steps. There's no reason to ship the runtime library just for compilation.

In https://reviews.llvm.org/D15225#1183135, @george.karpenkov wrote:

> @rnk Regarding the error message, would it be possible to compromise on something like "sanitizer runtime not available"?
>  I understand that the exact error message would be very useful for you, but it's just confusing for a user getting a toolchain with Xcode, since they can't just add the required file into the toolchain.

I think you are underestimating the likelihood that users will repackage and redistribute the C/C++ toolchain parts of the XCode as part of their normal build processes. I think in principle it's better for clang to remain self-contained, for it to just "know" which platforms have working sanitizer libraries. It feels like this change is trying to check whether the link will succeed during compilation, which is kind of putting the cart before the horse.



More information about the cfe-commits mailing list