- User Since
- Jan 2 2013, 4:34 PM (220 w, 6 d)
Fri, Mar 24
After briefly attempting to rework D31352 to work without this patch, I now see why you want to do this. Working with module flags is hard and working with named md nodes is easier. Module flags still seem appropriate for integer settings that need to match during LTO, but I don't think there are many appending module flags that wouldn't be better served by named md nodes.
Thu, Mar 23
I'm opposed to getting rid of the dllexport bit on GlobalValue, and so far on the bug, both @majnemer and @rjmccall seem to be opposed to it as well. I also asked @chandlerc about it in person, and he wasn't in favor, although I'm not sure how strong his opinion was.
Wed, Mar 22
We can't just keep passing more analysis pointers throughout our simplification utilities. We might want some wrapper of common, almost globally used analyses that optionally contains DL, TLI, DT, AC, etc.
Yep, thanks, looks good!
I think you need some driver logic to make sure the xray lists show up in .d files so that modifications to them trigger rebuilds in most build systems. See the SanitizerArgs.cpp driver logic for this.
Tue, Mar 21
This broke a CFI stats test, which I XFAILd in a follow up:
I agree with this in principle. The high-level LLVM_ENABLE_LTO flag shouldn't affect compiler-rt. We should have a different way of getting compiler-rt bitcode libraries if we want them.
- Fix bug in addAttribute implementation
Mon, Mar 20
I'm going to go ahead and land this to try to green the Chromium bots. We can reorganize later if we like.
I think it would be better if Config->OutputFile is always either empty or the true final output name. Otherwise we might query it somewhere in the middle of options processing and get an incorrect value.
Can you give an example of what this would enable?
Fri, Mar 17
Thu, Mar 16
- Delete stale typedef, sink arrayref helper
Wed, Mar 15
Tue, Mar 14
Cool, thanks for the reviews, this is definitely in better shape now. This is off by default, so I'm going to commit and experiment with it on a few codebases.
- Test that explicit casts suppress the warning