Tests on opam versions =< 2.3 are failing with solver timeouts #354
Assignees
Comments
shonfeder
changed the title
shonfeder
added a commit
to shonfeder/opam-repo-ci
that referenced
this issue
Sep 17, 2024
Trying to fix ocurrent#354. Recreates ocurrent#342 (for which the branch has been deleted). If this timeout is long enough, it will address the timeouts on opam 2.0, 2.1, and 2.2, since this env var has been supported at least since 2.0: https://github.com/ocaml/opam/blob/7d4a0f2e0fefe748efbd7358775b1a0bd8267544/src/solver/opamSolverConfig.ml#L131-L132 This can increase the build time of some tests, but I think it should be worth it if it can give proper results: - Only few builds (~3) per PR run on these older opam versions. - We are wasting a lot of cumulative preson time having to check the CI - results for PR that should be green if the CI wasn't giving these (arguably) false negatives.
Merged
shonfeder
changed the title
punchagan
pushed a commit
to shonfeder/opam-repo-ci
that referenced
this issue
Sep 19, 2024
Trying to fix ocurrent#354. Recreates ocurrent#342 (for which the branch has been deleted). If this timeout is long enough, it will address the timeouts on opam 2.0, 2.1, and 2.2, since this env var has been supported at least since 2.0: https://github.com/ocaml/opam/blob/7d4a0f2e0fefe748efbd7358775b1a0bd8267544/src/solver/opamSolverConfig.ml#L131-L132 This can increase the build time of some tests, but I think it should be worth it if it can give proper results: - Only few builds (~3) per PR run on these older opam versions. - We are wasting a lot of cumulative preson time having to check the CI - results for PR that should be green if the CI wasn't giving these (arguably) false negatives.
Merged
|
An easy, and I think reasonable, way to reduce the negative impact from this is just to not report solver failures as test failures. I think this is the right approach since, solver failures (especially on the old opam versions) hare not due to any defect or problem in the packages. |
|
A few years ago I had decent results by using |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
A handful of tests fail with solver timeouts on many package PRs. See, e.g., ocaml/opam-repository#26473 (comment)
Using opam 2.3 along with configuring opam to use the 0install solver (e.g., in #343) has been a big a help in reducing solver timeouts, but the problem still persists, and it creates false positives in the repo CI and slows down package publication. It could also mask genuine test failures.
We still run tests on opam 2.0, 2.1, and 2.2, since those are available on some distros, so we cannot simply switch all tests over to using this approach (see #352). An alternative approach proposed by @mtelvers in #342 is to increase the timeout.
The text was updated successfully, but these errors were encountered: