Add a second CI job on x86_64-linux for fsanitize=undefined (later asan?) builds of Lix #403

New issue

Open

opened 2024-06-19 03:44:22 +00:00 by jade · 1 comment

jade commented

2024-06-19 03:44:22 +00:00

Owner

Currently ASan is very horrible due to boost coroutines being busted garbage, and I don't really want to ship it because it randomly fails for nonsense seemingly un-suppressable reasons: https://gerrit.lix.systems/c/lix/+/1481

We have working UBSan support in-tree as of https://gerrit.lix.systems/c/lix/+/1483, so there is nothing really stopping this being done as a hydraJobs entry, besides a little bit of engineering our Nix expressions to support adding the meson option nicely.

We can probably afford to throw more x86_64-linux compute at our builds, and it would be nice to have actual sanitized test runs. After the rather extremely intense results of UBSan'ing and, to the extent possible, ASan'ing, lix in https://gerrit.lix.systems/q/topic:%22undefined-behaviour%22, I really think we might want to CI running the test suite with instrumentation. This would also, I guess, test no-gc builds actually work so would fix https://git.lix.systems/lix-project/lix/issues/319. Currently ASan is very horrible due to boost coroutines being busted garbage, and I don't really want to ship it because it randomly fails for nonsense seemingly un-suppressable reasons: https://gerrit.lix.systems/c/lix/+/1481 We have working UBSan support in-tree as of https://gerrit.lix.systems/c/lix/+/1483, so there is nothing really stopping this being done as a `hydraJobs` entry, besides a little bit of engineering our Nix expressions to support adding the meson option nicely.

jade added the

stability

Area/build-packaging

labels 2024-06-19 03:44:22 +00:00

jade commented

2024-06-19 03:46:23 +00:00

Author

Owner

Another idea for the build packaging is to make clang UBSan builds with the minimal runtime ("suitable for production use", they say) that we could deploy for developers, potentially even in the overlay. This would however cause potentially about 10% regression in perf; we would have to rerun benchmarks. Personally I wouldn't be mad about it on my own machines, but others might think otherwise. https://clang.llvm.org/docs/UndefinedBehaviorSanitizer.html#minimal-runtime

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference: lix-project/lix#403

No description provided.