We recently hit an issue where an llbuild change broke Swift CI, because our PR testing doesn't test everything downstream correctly. To fix this, and better match how we are testing the rest of our toolchain projects, I want to add testing of sourcekit-lsp to the llbuild PR tests. In looking into that, I discovered our existing PR tests for llbuild have some other deficiencies, and I'd like some feedback on how they should be configured.
Claims to test --preset=buildbot_linux_1604, but actually tests the 1404 preset
Question: should this just change to 1604 like it claimed? This is fairly close to testing the whole toolchain like we do in PR tests for Swift itself.
Edited to clarify: the Linux smoke bot does run the llbuild tests. It was another case where the jenkins page is giving an incorrect description of what the job runs. The only bot that doesn't run the llbuild tests is the non-smoke macOS bot.
I haven't read/thought about all of this stuff, but I want to give a big +1 to adding presets. All bots should use presets for reproducibility. IMO if we ever do not use a preset it is a bug.
Claims to test --preset=buildbot_linux_1604 , but actually tests the 14 04 preset
Question : should this just change to 1604 like it claimed? This is fairly close to testing the whole toolchain like we do in PR tests for Swift itself.
Smoke test changes look good. I have no problem with, and would indeed prefer, that we move the full linux tests to 16.04 to match Swift itself. And for the full macOS tests, it seems a bug to me that it has skip-test-llbuild.
New Preset: buildbot_llbuild_macos_platform,tools=RA,stdlib=RA
This bot also runs an xcodebuild command that's not part of build-script. For now, we should keep that. Someone should really add that to build-script, but it's beyond the scope of what I'm doing here.