Fix concat-satisfy on GHC 9.2.1 #114
Conversation
Testing the first release should imply that any later release of a major version should also work. E.g., testing 9.0.1 implies that 9.0.2 works, but testing 9.0.2 does not imply that 9.0.1 works. So the earlier the release that we test, the more test coverage we get. However, this doesn’t always hold like it should. E.g., we know that GHC 9.2.2 has breaking changes relative to 9.2.1, even though it shouldn't so we explicit test 9.2.2 as well.
sellout
force-pushed
the
fixing-9.2.1
branch
from
2cb28ff to
623de1e
Compare
The latest release isn’t compatible with concat, so the lack of upper bound meant the new constraints release started breaking concat.
|
Unfortunately, 83f06ae seems to have broken I don't use concat-plugin, so the remaining breakage doesn't affect me, but the fixes here address failures in |
sellout
force-pushed
the
fixing-9.2.1
branch
from
5dab49e to
5aa593b
Compare
|
Thanks, @sellout. I was going to investigate the CI failures (plugin works for me, as do the gold tests), but ran out of time. I'll get back to it next week. Meanwhile, my experience has been that there's no version jump in ghc that doesn't hold the risk of breaking changes. So I suggest also adding the latest known version along each 9.x line at least. |
| - "9.0.2" | ||
| - "9.2.4" | ||
| - "9.4.8" | ||
| - "9.0.1" |
There was a problem hiding this comment.
I think we should also do the last version of each branch, but we can do that separately.
There was a problem hiding this comment.
We get the latest release covered in the Nix builds. Nix tends to not keep all the releases around, so I test the newer ones via Nix and the oldest via Cabal directly.
It does seem like the gold tests are passing in 9.4.8 (on Linux), but not in 9.4.1.
sellout
merged commit d95c79d
into
compiling-to-categories:master
This broadens the CI coverage and fixes an issue that was missed by
testing 9.2.4 instead of 9.2.1.