Wingman: Low gas warning

[isovector]

Make Wingman inform you if it ran out of gas when attempting to synthesize an expression, rather than just failing with a "couldn't be solved" error message.

Feature request from @masaeedu.

[isovector]

@pepeiborra the dynamically linked builds are now failing in CI with an undefined linker reference to my new constructor. Usually when I see this it means my cabal file's modules are not up to date, but that doesn't appear to be the case here. Any idea what's up?

[pepeiborra]

"Linking" the tactics test suite fails to find the symbol for the new constructor. I'm not sure what "linking" means here since we are asking Cabal to use dynamic linking for executables, it could be some sort of validation step, or TH if any.

We are doing dynamic linking with stack: this is probably not well trodden and involves telling cabal to link haskell packages dynamically (for executables) without stack being aware. What could be going wrong here?

If you have the time, please put together a minimal repro and send it upstream to stack or wherever it belongs.

In terms of fixing it, there's a couple of options:

1. Disable the caching in CircleCI.
2. Disable dynamic linking but find a way to reduce the memory footprint of linking, likely by switching to the gold linker.

I don't have time to investigate any further, hopefully you or any of the usual maintainers can take a look!

[isovector]

For anyone who might be reading the thread:

I'm not sure what "linking" means here since we are asking Cabal to use dynamic linking for executables

This is what appears in CI, and in my local builds:

hls-tactics-plugin         > Linking .stack-work/dist/x86_64-linux/Cabal-3.2.1.0/build/tests/tests ...
hls-tactics-plugin         > .stack-work/dist/x86_64-linux/Cabal-3.2.1.0/build/tests/tests-tmp/CodeAction/AutoSpec.dyn_o:function CodeActionziAutoSpec_spec1_info: error: undefined reference to 'hlszmtacticszmpluginzm1zi2zi0zi0zmFpMEBH6ApWkAbtR1G6wNS6_WingmanziTypes_NotEnoughGas_closure'
hls-tactics-plugin         > collect2: error: ld returned 1 exit status
hls-tactics-plugin         > `gcc' failed in phase `Linker'. (Exit code: 1)

[Ailrun]

Disable the caching in CircleCI.

IMO this is a non-option as it makes CircleCI be a serious bottleneck of PR processes. When the cache had a problem, CircleCI took about 5 hours, and we often have more than 3 somewhat active PRs, so the CI process will take about a day just by itself if we disable the cache.

[jneira]

Hmm i think the problem could be caused by .stack-work being cached. In theory local packages should not be cached but built from scratch for each run. The major part of saved time comes from caching dependencies, and that cache is in .stack
The dynamic linking has uncovered a stack bug i guess.

@isovector could you test locally if stack clean --full resolves the link problem?

[jneira]

Let see how much time cost build local packages from scratch: jneira#25

[jneira]

Ok, with the actual full cache (https://app.circleci.com/pipelines/github/jneira/haskell-language-server/642/workflows/0dfabe4c-3b34-49a3-8df6-e799103089db):

job	id	time
ghc-default	5425	24m 54s
ghc-9.0.1	5424	16m 35s
ghc-8.8.4	5429	25m 51s
ghc-8.8.3	5432	33m 49s
ghc-8.6.5	5426	20m 47s
ghc-8.6.4	5427	21m 32s
ghc-8.10.5	5430	24m 43s
ghc-8.10.4	5423	22m 13s
ghc-8.10.3	5431	27m 33s
ghc-8.10.2	5428	20m 56s

Caching only deps (https://app.circleci.com/pipelines/github/jneira/haskell-language-server/644/workflows/f66595fc-2e5e-46c5-b549-842e59340e84):

job	id	time
ghc-default	5446	27m 31s
ghc-9.0.1	5444	17m 4s
ghc-8.8.4	5447	37m 28s
ghc-8.8.3	5448	13m 21s
ghc-8.6.5	5449	20m 42s
ghc-8.6.4	5445	19m 10s
ghc-8.10.5	5451	1h 7m 30s (wout cache)
ghc-8.10.4	5450	15m 16s
ghc-8.10.3	5443	33m 14s
ghc-8.10.2	5452	12m 23s

Times are not uniform in the 2nd case so maybe it could be closed to full cache.

[pepeiborra]

The numbers look good @jneira, please send a PR to merge it

[Ailrun]

LGTM

[jneira]

thanks for your phenomal work on wingman 🙂

[jneira]

Several jobs were stalled since 17 hours with the message:

Can't find any online and idle self-hosted or hosted runner in the current repository, account/organization that matches the required labels: 'ubuntu-latest'
Waiting for a self-hosted or a hosted runner to pickup this job...

As reported by @Ailrun in some pr i dont find out right now
I've cancelled all those jobs and will restart them in priority pr order (wingman ones first)
I've reported the error in gha: actions/runner#1230 (comment)

[jneira]

At least the testing workflow of this one has started to run 🤞