[PATCH] D16092: [Coverage] Do not allocate memory for coverage map data (Linux)

[Vedant Kumar via llvm-commits]

vsk added a comment.

Along with the compiler-rt test I think this looks fine, but caveat: I'm not a linker expert at all!

After some discussion with @pete and @lgerbarg, it looks like there isn't a short term way to achieve this with Mach-O. To recap:

1. Passing in `-Wl,-dead_strip -fprofile-instr-generate -fcoverage-mapping` doesn't get rid of __llvm_covmap because the linker is conservative about sections it doesn't 'know' about.
2. Because this section is in the __DATA segment, it will be mapped into the process along with the rest of the segment. We could pull this out into a separate segment (along with debug info?) and experiment.


http://reviews.llvm.org/D16092