Make polymorphic functions more efficient and expressive #17548
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
smarter
force-pushed
the
poly-closure2
branch
5 times, most recently
from
7535ede
to
0022e5b
Compare
odersky
approved these changes
Jun 6, 2023
| @@ -151,7 +151,7 @@ object untpd extends Trees.Instance[Untyped] with UntypedTreeInfo { | |||
| case class CapturesAndResult(refs: List[Tree], parent: Tree)(implicit @constructorOnly src: SourceFile) extends TypTree | |||
|
|
|||
| /** Short-lived usage in typer, does not need copy/transform/fold infrastructure */ | |||
| case class DependentTypeTree(tp: List[Symbol] => Type)(implicit @constructorOnly src: SourceFile) extends Tree | |||
| case class DependentTypeTree(tp: (List[TypeSymbol], List[TermSymbol]) => Type)(implicit @constructorOnly src: SourceFile) extends Tree | |||
There was a problem hiding this comment.
Maybe we simplify this so that a DependentTypeTree just takes a (tp: List[Symbol] => Type)?
There was a problem hiding this comment.
Using a List[Symbol] means that we need to spend extra time concatenating and partitioning the parameters, and we need to document this behavior, so I'm not sure it's really more simple, but I'm happy to do the change if you prefer it that way.
There was a problem hiding this comment.
I agree, in the end it does not seem to be a simplification. So OK to keep as is.
There was a problem hiding this comment.
When I started my comment stream I thought it would come out simpler than it did 😄
| val tpe = tpFun(termParamss.head) | ||
| // A lambda has at most one type parameter list followed by exactly one term parameter list. | ||
| val tpe = (paramss: @unchecked) match | ||
| case TypeSymbols(tparams) :: TermSymbols(vparams) :: Nil => tpFun(tparams, vparams) |
There was a problem hiding this comment.
If we simplify DependentTypeTrees, these would become:
case TypeSymbols(tparams) :: TermSymbols(vparams) :: Nil => tpFun(tparams ++ vparams)
case TermSymbols(vparams) :: Nil => tpFun(vparams)| val resultTpt = pt.dealias match | ||
| case RefinedType(parent, nme.apply, poly @ PolyType(_, mt: MethodType)) if parent.classSymbol eq defn.PolyFunctionClass => | ||
| untpd.DependentTypeTree((tsyms, vsyms) => | ||
| mt.resultType.substParams(mt, vsyms.map(_.termRef)).substParams(poly, tsyms.map(_.typeRef))) |
There was a problem hiding this comment.
If we simplify DependentTypeTrees, this would become:
untpd.DependentTypeTree(syms =>
val (tsyms, vsyms) = syms.partition(_.isType)
mt.resultType.substParams(mt, vsyms.map(_.termRef)).substParams(poly, tsyms.map(_.typeRef)))
|
So I am OK to merge as is. |
Previously, the expression
val a: (x: Int) => Bar[x.type] = ??? : ((x: Int) => Foo[x.type])
lead to the following error:
Found: (x: Int) => Foo[x.type]
Required: (x: Int) => Bar[x².type]
where: x is a reference to a value parameter
x² is a reference to a value parameter
There's two problems here:
1. In the second lambda, x is supposed to be renamed to x², but the binder
wasn't renamed.
2. Since the binders introducing x are part of the printed types, there isn't
any ambiguity and no renaming is actually needed.
(1.) is a straight-up bug in the printing logic that we fix. (2.) is maybe a
matter of opinion, but for clarity we choose to address it by using the same
superscript for every parameter reference with the same name whose binder has
been printed. This does meant that `(x: Int) => (x: Int) => x.type` won't be
disambiguated, but something like that would be better addressed by printing a
warning about name shadowing anyway.
This was dropped from the syntax, so this isn't needed anymore.
Reuse the existing `DependentTypeTree` mechanism already in place for monomorphic lambdas to compute the result type of polymorphic lambdas based on their expected type.
Previously, we desugared them manually into anonymous class instances, but by using a Closure node instead, we ensure that they get translated into indy lambdas on the JVM. Also cleaned up and added a TODO in the desugaring of polymorphic function types into refinement types since I realized that purity wasn't taken into account.
…uilds This avoids finding compilation errors late in the development cycle. Having explicit nulls on is especially important because typed and untyped trees are now only distinguished using `Null`, so on the non-bootstrapped build we don't get compiler errors when passing an untyped Tree where a typed Tree is expected.
Just like for monomorphic closures this isn't currently supported.
The check used to disallow any reference of the form `X.this.type` even though `X` is never the refinement class. This change mirrors the check done with `ThisType` a couple of line below.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR enhances polymorphic function types in two ways:
Additionally, we fix the logic for renaming bound variables when pretty-printing lambdas and fix the handling of
thisin refinements.