[Fix] Polymorphic field typechecking #1872
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
Using a polymorphic type annotation on a record field, as in
`{id : forall a. a -> a = std.function.id}` would unexpectedly file with
a failure to unify `forall a. a -> a` with `a -> a`, which indicates
that the type annotation wasn't properly instantiated before being used
to check the value of the field.
It turns out the typechecking of the field per se is fine, but the issue
comes from the recursive environment: in a recursive record, we need to
build an environment with types for all record fields _before_ we can
start to actually typecheck each field. When the field isn't annotated,
we use a fresh unification variable. When typechecking the corresponding
field, we unify what's in the type environment (the unification
variable, but which might have been unified with another type at this
point) with the type inferred for the field.
However, when the field has a type annotation, we use this annotation
for the recursive environment instead (which is important to keep
polymorphic types...polymorphic). In this case, we would unify a
polymorphic type with the inferred type, which is the place that was
missing the instantiation.
One possibility is to just add the missing instantiation, but in fact
this unification is useless when there is a type annotation: to avoid
this useless work, we rather remember which fields are annotated, and we
entirely skip the unification for them.
yannham
commented
Mar 27, 2024
| @@ -2857,7 +2900,8 @@ fn instantiate_foralls( | |||
| .. | |||
| } = ty | |||
| { | |||
| let kind = (&var_kind).into(); | |||
| let kind: VarKindDiscriminant = (&var_kind).into(); | |||
There was a problem hiding this comment.
Cosmetic change: I find it better to put an explicit type annotation when there's an into() that isn't used right away.
yannham
commented
Mar 27, 2024
Comment on lines
+2881
to
+2892
| // | ||
| // As this function can be called on monomorphic types, we only increment the level when we | ||
| // really introduce a new block of rigid type variables. | ||
| if matches!( | ||
| ty, | ||
| UnifType::Concrete { | ||
| typ: TypeF::Forall { .. }, | ||
| .. | ||
| } | ||
| ) { | ||
| ctxt.var_level.incr(); | ||
| } |
There was a problem hiding this comment.
Unrelated small change that was on my radar for some time. It doesn't change anything with respect to correctness (the typechecker was already correct before), but would just introduce more levels that necessary.
jneem
approved these changes
Mar 28, 2024
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.
Closes #1690
Using a polymorphic type annotation on a record field, as in
{id : forall a. a -> a = std.function.id}would unexpectedly fail with a failure to unifyforall a. a -> awitha -> a, which indicates that the type annotation wasn't properly instantiated before being used to check the value of the field.It turns out the typechecking of the field per se is fine, but the issue comes from the recursive environment: in a recursive record, we need to build an environment with the types of all record fields before we can start to actually typecheck each field. When the field isn't annotated, we use a fresh unification variable. When typechecking the corresponding field, we unify what's in the type environment (the unification variable, but which might have been unified with another type at this point) with the type inferred for the field. Doing so, we reject things like
{a = true, b = a + 1} : _.However, when the field has a type annotation, we use this annotation for the recursive environment instead (which is important e.g. to keep polymorphic types...polymorphic). In this case, the additional unification step described above would unify a polymorphic type with the inferred type, which is the place that was missing the instantiation.
One possibility is to just add the missing instantiation, but in fact this unification is useless when there is a type annotation, as the typechecking of the field with a polymorphic annotation will perform the same work anyway (infer a type and check that it's a subtype of the instantiated annotation). We thus rather remember which fields are annotated and we entirely skip the unification in this case.