Details
Diff Detail
- Repository
- rL LLVM
- Build Status
Buildable 26677 Build 26676: arc lint + arc unit
Event Timeline
Super cool!
llvm/utils/gn/build/libs/pthread/BUILD.gn | ||
---|---|---|
9 | Maybe add a # On Android, bionic has built-in support for pthreads. like comment (I had to look that up just now). | |
llvm/utils/gn/build/toolchain/BUILD.gn | ||
49 | (this might be a git merge issue -- be sure to run git ls-files '*.gn' '*.gni' | xargs -n 1 gn format though :-) | |
158 | Whoa, do deps on toolchains work now? Is that new? | |
169 | That's alright for now, but I think eventually we want to remove the target_foo stuff from the toolchain definition again and instead use configs for this (like https://cs.chromium.org/chromium/src/build/config/compiler/BUILD.gn?type=cs&q=build/config/compiler&sq=package:chromium&g=0&l=217) | |
172 | The -fuse-ld=lld will probably need revisiting -- probably want to also add that on Linux when e.g. clang_base_path is set, and maybe if some use_lld arg is explicitly set to on. But ok for now. |
llvm/utils/gn/build/toolchain/BUILD.gn | ||
---|---|---|
158 | It's been around for a while, but it's discouraged. gn help toolchain says for toolchain deps: This concept is somewhat inefficient to express in Ninja (it requires a lot of duplicate of rules) so should only be used when absolutely necessary. I don't think there's a cleaner way for multi-stage builds though since you need to somehow express that dependency. | |
169 | That would be my preference. Rather than combining different ways of specifying flags, we should just introduce default configs for each platform and then add those to defaults for each target type (e.g. https://fuchsia.googlesource.com/build/+/master/config/BUILDCONFIG.gn#146) |
llvm/utils/gn/build/toolchain/BUILD.gn | ||
---|---|---|
172 | I think that should also be a config that's set as a default. If it's done that way, targets can then opt-out by using: configs -= ["//build/config:lld"] |
llvm/utils/gn/build/toolchain/BUILD.gn | ||
---|---|---|
49 | Yes, I think I might have forgotten to format D56576. This change restores the correct formatting though. | |
169 | We'll see how that goes. One issue is that we'll need some way of writing the target flags to a file to run the tests: |
Maybe add a # On Android, bionic has built-in support for pthreads. like comment (I had to look that up just now).