[CS] Don’t fail constraint generation for ErrorExpr or if type fails to resolve #60062
Conversation
|
@swift-ci Please smoke test |
|
@swift-ci Please SourceKit stress test |
lib/Sema/CSSimplify.cpp
Outdated
| @@ -11208,8 +11208,17 @@ ConstraintSystem::simplifyApplicableFnConstraint( | |||
| if (shouldAttemptFixes()) { | |||
| if (auto *typeVar = type2->getAs<TypeVariableType>()) { | |||
| auto *locator = typeVar->getImpl().getLocator(); | |||
| if (typeVar->isPlaceholder() || hasFixFor(locator)) | |||
| if (typeVar->isPlaceholder() || hasFixFor(locator)) { | |||
There was a problem hiding this comment.
Could you please do a couple of things here:
- Remove
typeVar->isPlaceholder()check since it's always false because type variable cannot be a placeholder anymore - Move whole
if (shouldAttemptFixes())to just beforeTypeMatchOptions subflags = getDefaultDecompositionOptions(flags); - use
desugar2instead ofsimplifyType(type2);
There was a problem hiding this comment.
use
desugar2instead ofsimplifyType(type2);
We need to simplify the type because getFixedTypeRecursive doesn’t simplify AnyMetatypeTypes that have a type variable as instance type, so in desugar2 the type variable won’t have been replaced by a placeholder yet. The alternative would be to add handling for AnyMetatypeType in getFixedTypeRecursive. Do you have a preference what would be the better solution?
There was a problem hiding this comment.
Hm I keep forgetting about all the nuances of getFixedTypeRecursive vs. simplifyType. Maybe we should just call getFixedTypeRecursive on an instance type if type2 was a metatype for this new check.
ahoppen
force-pushed
the
pr/placeholder-for-errors
branch
from
6eb3603 to
a9514f9
Compare
|
@swift-ci Please smoke test macOS |
|
@swift-ci Please SourceKit stress test |
ahoppen
force-pushed
the
pr/placeholder-for-errors
branch
3 times, most recently
from
d3a70ab to
95abc5b
Compare
|
@swift-ci Please smoke test macOS |
test/Parse/matching_patterns.swift
Outdated
| if case let .Naught(value1, value2, value3) = n {} // expected-error{{pattern with associated values does not match enum case 'Naught'}} | ||
| // expected-note@-1 {{remove associated values to make the pattern match}} {{20-44=}} | ||
| // expected-error@-2 {{could not infer type of variable 'value1'}} |
There was a problem hiding this comment.
I think this is actually a regression of sorts - the main problem here is described by the first error and problems with value1..3 are just a consequence...
test/stmt/foreach.swift
Outdated
|
|
||
| for (x, y, _) in arr2 {} | ||
| // expected-error@-1 {{pattern cannot match values of type '(a: Int, b: String)'}} | ||
| // expected-error@-2 {{could not infer type of variable 'x'}} |
There was a problem hiding this comment.
Here is the same, maybe we could somehow coalesce fixes
ahoppen
force-pushed
the
pr/placeholder-for-errors
branch
from
95abc5b to
27ce696
Compare
|
@swift-ci Please smoke test |
ahoppen
force-pushed
the
pr/placeholder-for-errors
branch
from
27ce696 to
88fc983
Compare
|
@swift-ci Please smoke test |
…to resolve Instead of failing constraint generation by returning `nullptr` for an `ErrorExpr` or returning a null type when a type fails to be resolved, return a fresh type variable. This allows the constraint solver to continue further and produce more meaningful diagnostics. Most importantly, it allows us to produce a solution where previously constraint generation for a syntactic element had failed, which is required to type check multi-statement closures in result builders inside the constraint system.
…nside a named pattern We need this to resolve a test failure in optional.swift.
The issue has been fixed by allowing holes for variables in paterns.
ahoppen
force-pushed
the
pr/placeholder-for-errors
branch
from
88fc983 to
5ba66aa
Compare
|
@swift-ci Please smoke test |
|
Resolves: #60485 |
…rExpr It appears like this was missed in swiftlang#60062.
…rExpr It appears like this was missed in swiftlang#60062.
…rExpr It appears like this was missed in swiftlang#60062.
Instead of failing constraint generation by returning
nullptrfor anErrorExpror returning a null type when a type fails to be resolved, return a fresh type variable. This allows the constraint solver to continue further and produce more meaningful diagnostics.Most importantly, it allows us to produce a solution where previously constraint generation for a syntactic element had failed, which is required to type check multi-statement closures in result builders inside the constraint system.
Resolves: #60029
Resolves: #60485