GHC 9.14 support#2604
Conversation
|
Thanks @gergoerdi! I'm stuck at the same point, roughly. I don't have the failure on Incr though. I think I fixed that one in 5a6b33e2d, merged in ucsd-progsys/liquid-fixpoint#819 and #2609. I'll let you know if I discover what's making a difference for SKIIdx and SKILam. |
|
This fixes the failing tests: |
The development version is only expected to build with one GHC. When upgrading to GHC 9.14, CI would need to be updated as well. If you want to support more than one GHC version, the considerations of #2379 are relevant. |
We had pretty much the same issue a few versions ago. It turned out they added a new benchmarking function and ended up using it in the definition of id. To see if that’s what’s going on again, we should dump the Core, find the function name, and then assert in the Prelude that it’s equal to id |
|
GHC 9.14.1: GHC 9.12.2: I think its the breakpoint thingy, it should be reflected as the identity function |
2d65c1c to
3c6b3e4
Compare
|
I rebased develop, and contributed some commits.
Might work. It is a change that can be separated from the upgrade, though. I think I would prefer to assume-reflect it most of the time, to verify with a simpler definition of |
I don't think we should break user code, we should be the ones providing that assume reflect |
👍 Feel free to make a pull request. |
f73974a to
4f94bd3
Compare
|
Thanks @gergoerdi again for submitting the fixes, and thanks @TeofilC for providing the patches in |
This draft PR adds GHC 9.14 support. It is a draft because I haven't gone through the trouble of factoring out the compatibility parts so now it only builds with GHC 9.14.
The more pertinent reason that I am making this PR in this state is because there are two mysterious (to me) test failures on the branch:
ple-posbenchmark-icfp15-pos