Annotate ASTs with types at every node #991
Conversation
- Split out `Swarm.Language.Module`, put annotated AST in a `Module` - Update module typechecking appropriately - Still some `undefined` things that need to be filled in!
Need to rewrite `elaborate`.
Getting very bogged down. Might need to throw this code away and start again with some kind of more principled approach...
| mkProg = TApp (TConst Make) (TText (e ^. entityName)) | ||
| mkPT = ProcessedTerm mkProg (Module mkTy empty) mkReq empty | ||
| let name = e ^. entityName | ||
| mkPT = [tmQ| make $str:name |] |
src/Swarm/Language/Syntax.hs
Outdated
| pattern Syntax :: SrcLoc -> Term -> Syntax | ||
| pattern Syntax l t = Syntax' l t () | ||
|
|
||
| {-# COMPLETE Syntax #-} |
There was a problem hiding this comment.
It specifies a set of constructors/pattern synonyms which should be considered "complete" for the purposes of pattern match coverage checking. i.e. if you pattern-match on everything listed in the COMPLETE pragma then GHC will not warn about missing cases. Using COMPLETE is common when creating pattern synonyms since there's no way in general to know whether an arbitrary set of pattern synonyms is going to cover all possible cases, so without the pragma in general GHC will generate lots of warnings about incomplete pattern matches.
https://ghc.gitlab.haskell.org/ghc/doc/users_guide/exts/pragmas.html#complete-pragmas
`Testing/479-atomic` timed out, not sure why.
Was accidentally deleting the `atomic` constant.
|
I've realized a big issue with the way this works: in particular it doesn't make sense to annotate every node of the AST with a |
|
@byorgey won't every definition and let binding have a I would not mind if the type was stored in the |
Yes, they should. Still trying to figure out a good way to annotate different things with |
|
@byorgey could you merge this as is and improve on it in the main branch? 🙂 The types are morally wrong with the I would like to rebase and merge my OnHover rework and the types added here are invaluable. Not that I am in a rush, I just don't want to have merge conflicts with @kostmo if he works on the LSP further. 😅 |
|
You're right, I will do that. Can't let the perfect become the enemy of the good. 😄 |
|
As discussed above, we can merge this as-is now, and I've created #1042 for tracking the remaining issue. |
Co-authored-by: Restyled.io <[email protected]>
test/unit/TestLanguagePipeline.hs
Outdated
|
|
||
| isVar (TVar {}) = True | ||
| isVar _ = False | ||
| getVars = map (_sTerm &&& _sType) . filter (isVar . _sTerm) . universe |
There was a problem hiding this comment.
See, I put that in there on purpose to give people ideas. 😄 https://hackage.haskell.org/package/lens-5.2/docs/Control-Lens-Plated.html has some pretty powerful tools for working with ASTs.
Values of type
Syntaxare as before: parsed syntax, with each node annotated withSrcLoc.Values of type
Syntax' Polytype, however, have each node annotated with both aSrcLocand aPolytype. (Syntaxis really just a synonym forSyntax' ().)Type inference takes a
Syntaxand outputs aTModule, which now contains aSyntax' Polytype, in other words, a new version of the AST where every node has been annotated with the inferred type of the subterm rooted there.Why is this useful?
OnHoverLSP handler, e.g. to show the type of the term under the mouse.I imagine the code changes might look kind of intimidating but I don't think it's really that bad once you understand what is going on, so I'm happy to answer any questions or explain anything.