GHC 9.4 compatibility + Multiple Home Units #2994
Conversation
|
Perhaps we should move all the code actions that depend on |
wz1000
requested review from
Ailrun,
Anton-Latukha,
July541,
OliverMadine,
berberman and
kokobd
as code owners
|
I've added one commit which adds the required files for GHC 9.4 for nix. |
konn
left a comment
konn
left a comment
There was a problem hiding this comment.
Although I haven't tested in my environment, it seems ok as for the Splice Plugin.
Just curious: is this change to make the splice plugin dependent on hls-refactor-plugin also fixes the Splice Plugin also on GHC 9.2, or do we need an extra effort?
wz1000
force-pushed
the
wip/9.4
branch
2 times, most recently
from
f425458 to
4232bd3
Compare
wz1000
requested review from
drsooch,
eddiemundo,
georgefst and
uhbif19
as code owners
wz1000
force-pushed
the
wip/9.4
branch
2 times, most recently
from
3308000 to
6a6ffa1
Compare
wz1000
force-pushed
the
wip/9.4
branch
5 times, most recently
from
a888817 to
88fd42f
Compare
I've just added naively the different nix lines. The configuration file comes from a copy of the one for 9.2. With that, we can open a shell with `nix develop .\#haskell-language-server-941-dev` and type `cabal build`. (cherry picked from commit 48084ab)
|
@guibou could you help with the nix failure? |
pepeiborra
left a comment
pepeiborra
left a comment
There was a problem hiding this comment.
Minor comments only, as the slice of HLS that I understand shrinks with every GHC update.
This is a massive amount of work, thank you @wz1000 !!!
| then uses_ GetModSummary fs | ||
| else uses_ GetModSummaryWithoutTimestamps fs | ||
| return (map (NodeKey_Module . msKey) dep_mss) | ||
| ms <- msrModSummary <$> use_ GetModSummary file |
There was a problem hiding this comment.
Can use GetModSummaryWithoutTimestamps when fullModSummary is false here?
There was a problem hiding this comment.
yes, its for the new ModuleGraph which includes all ModSummarys paired with their direct dependencies (Earlier GHC versions only had the ModSummarys). It shouldn't be used for much other than mgTransDeps gives us a reasonably fast and cached way of querying for transitive dependencies (which is used in hscCompileCoreExprHook to get all the Linkables needed for a splice). I don't think omitting timestamps will matter here, especially since we omitted timestamps in the ModSummarys we put into the ModuleGraph on older GHCs.
| session' <- liftIO $ mergeEnvs hsc mss inLoadOrder depSessions | ||
| let inLoadOrder = map (\HiFileResult{..} -> HomeModInfo hirModIface hirModDetails Nothing) ifaces | ||
| #if MIN_VERSION_ghc(9,3,0) | ||
| mss_imports <- uses_ GetLocatedImports (file : deps) |
There was a problem hiding this comment.
A comment explaining why collecting the mod summaries of the direct dependencies here
There was a problem hiding this comment.
You were right, there was something fishy here. We were including the ModSummary for the current module in GHC 9.4 but not before. I've changed this to always include the current ModSummary, which allows us to not query imports of direct dependencies. We still need to get the ModSummarys of direct dependencies in order to know their Modules, which we need to reference in the ModuleGraph.
I've written a short note documenting all the assumptions we now make.
wz1000
force-pushed
the
wip/9.4
branch
5 times, most recently
from
1269371 to
bb49204
Compare
Yes, for sure, I'll have a look at it tomorrow. |
TODO