This allows us to coerce closure expressions to function types with lifetime
dependencies.
Since captures are added to the parameter list when lowering closures to SIL
functions, we also need to update the result index when lowering their lifetime
dependencies.
This is important for things like instance methods of actors in
particular because otherwise it won't be possible to compute a
correct isolation for the thunk.
This fix enables fully unapplied references to actor-isolated
instance methods and other functions with isolated parameters.
Resolves: rdar://148395744
We no longer need to track the `ForEachStmtInfo` in the
`SyntacticElementTarget`, and we can remove the special diagnostic
logic for `next` and `makeIterator` since those are type-checked
separately now.
Copy, Borrow, {Any, Optional}Try, VarargExpansion expressions
have to perform the same transformation - resolve the type and
use it to coerce their sub-expression. This change introduces a
single method to do just that instead of copying the same code
around.
Since `try!` now forces l-value -> r-value conversion during
CSGen, let's simplify solution application to `try` expressions
by coercing sub-expression to a type of a `try` itself which
would introduce all necessary loads.
The change in 8ab8b2e3e9, was too
broad. We want to coerce the subexpression of `try!` to an rvalue,
but not the subexpression of a `try`.
If the subexpression of a `try` becomes an rvalue even though the
type of the parent expression is an lvalue, we can end up with
infinite recursion in coerceToType(), as demonstrated by the
test case.
Fixes https://github.com/swiftlang/swift/issues/85034.
This appears to only support synthesized conformances because
operators in such cases use different names - `__derived_*`.
These days devirtualization like this is performed as part of
mandatory inlining. And this approach doesn't stack well with
features like `MemberImportVisibility` because there is change
to check whether witness is available or not.
This means we now either produce a bare ErrorType, or an ErrorType
with a generic parameter original type for a generic parameter hole.
We ought to further consolidate this logic by sinking the generic
parameter original type replacement into `simplifyType` itself, but
I'm leaving that for a future patch since it affects completion
results and I want to try keep this close to NFC.
Generic methods declared in protocols (and extensions thereof) cannot
be used on existential values, because there is no way to specialize
them for all potential types. Diagnose such cases in Embedded Swift
mode and via `-Wwarning EmbeddedRestrictions`.
This adds a bunch more warnings to the standard library that we'll
need to clean up, probably by `#if`'ing more code out.
Part of rdar://119383905.
Aidan Hall
Anthony Latsis
Aidan Hall
Kavon Farvardin
Pavel Yaskevich
Hamish Knight
Elsa Keirouz
Mykola Pokhylets
Slava Pestov
Becca Royal-Gordon
Pavel Yaskevich
Tim Kientzle
Allan Shortlidge
Doug Gregor