Revert "Send unhandled exceptions to the user (#2484)"

[jneira]

• This reverts commit 0211f75.
• That commit causes ghcide tests in windows hangs for all ghcs
• The test suite was marked incorrectly as ok due a bug in the workflows (fixed with Fail if main or pre jobs are cancelled #2493)
• See Send unhandled exceptions to the user #2484 for more details

[pepeiborra]

Can you reproduce the problem locally? I don't have a Windows machine.

Can you share a link to a CI job showing the problem? I looked at the original PR and I don't see any hanging.

Approving based on trust.

[jneira]

Can you reproduce the problem locally? I don't have a Windows machine.

I did not try yet, will do

Can you share a link to a CI job showing the problem? I looked at the original PR and I don't see any hanging.

Collecting here the links:

• Last run from the original pr cancelled after 4 hours : https://github.com/haskell/haskell-language-server/runs/4546179353?check_suite_focus=true. This was incorrectly merged due to the mentioned bug in the workflows.
• Run exhibiting the issue in the haskell repo after merging the pr: https://github.com/haskell/haskell-language-server/runs/4557599786?check_suite_focus=true (from pr Fail if main or pre jobs are cancelled #2493)
• Run in my fork with the pr included exhibiting the issue: https://github.com/jneira/haskell-language-server/runs/4558206432?check_suite_focus=true (with pr test doc change jneira/haskell-language-server#55)
• Succesful run in my fork after reverting: https://github.com/jneira/haskell-language-server/runs/4559275575?check_suite_focus=true (with pr Revert "Send unhandled exceptions to the user (#2484)" jneira/haskell-language-server#56)

Approving based on trust.

thanks!

[pepeiborra]

I have looked at the 3 linked failures and I don't see any hanging. The jobs seem to be taking a very long time to build ghcide and then the ghcide test suite, which suggests a problem with CI caching?
I am out of touch with the CI setup after the recent changes, but I don't think we should be reverting this PR without more evidence.

[pepeiborra]

Withholding approval until we have more evidence that this PR caused hangs

[jneira]

I've added a fourth link to the original pr last failing run before the merge in master.

Not sure what additional evidence can I include apart the already linked:

• the original pr failing run the ghcide test suite: cancelled after 4h 25 min in the middle of initialize response capabilities test group
• a posterior build with the pr included cancelled after 2h 9m hanged in the same test group
• one build in my fork with the commit included, cancelled after 4h 39m hanged in the same test group.
• the successful run in my fork reversing the pr
• the successful run of this pr reverting the pr

do you thin a local reproduction would be needed? even if i don't get to reproduce it locally it would not confirm than the pr is not the culprit and we need the ci green

[jneira]

to check if the cache is related (don't see how the cache can make hang a test at runtime but still) i am gonna open another pr invalidating it

[pepeiborra]

Is it the "text doc sync" test that is handling? Please try a local repro and share the log of messages

[jneira]

all runs, the failing and successful ones are using the same cache so...

[michaelpj]

Hmm, maybe this is actually a caching issue? It does seem like there's more recompilation going on that I would expect if the caching was working.

[jneira]

Is it the "text doc sync" test that is handling? Please try a local repro and share the log of messages

I am not at home right now but I ve opened a pr in my repo with LSP msgs enabled in the first call to tests

jneira#57

[jneira]

The run https://github.com/haskell/haskell-language-server/runs/4561717143?check_suite_focus=true
from the pr #2499 with a fresh cache has failed with the same hang so i think it is clear the cache is not related with the bug

[jneira]

I think is is clear this revert unblocks the ci and there are several prs in the queue to be merged.
We could continue working in the diagnose and the fix afterwards.

The pr with lsp messages enabled has ended, the full log can be downloaded here: https://github.com/jneira/haskell-language-server/suites/4693272570/logs?attempt=1

[soulomoon]

This is still ghosting us sometimes, see https://github.com/haskell/haskell-language-server/actions/runs/8924119343/job/24509825742?pr=4202