The introduction of forward-scan matching for trailing closures
(SE-0286) failed to account for unresolved member expressions,
sometimes causing a crash in SILGen. Fixes rdar://problem/67781123.
This approach, suggested by Xiaodi Wu, provides better source
compatibility for existing Swift code, by breaking ties in favor of the
existing Swift semantics. Each time the backward-scan rule is needed
(and differs from the forward-scan result), we will produce a warning
+ Fix-It to prepare for Swift 6 where the backward rule can be
removed.
My experiment to improve source compatibility by also performing a
backward scan removed the SE-0286 heuristic that skipped binding
the unlabeled trailing closure to a defaulted parameter when that
would fail. Reinstate that heuristic, which makes more existing code
work with the forward-scan behavior.
This makes my source-compatibility improvements a quality-of-implementation
Whenever we form a call that relies on the deprecated "backward" scan,
produce a warning to note the deprecation along with a Fix-It to label
the parameter appropriately (and suppress the warning). For example:
warning: backward matching of the unlabeled trailing closure is
deprecated; label the argument with 'g' to suppress this warning
trailingClosureEitherDirection { $0 * $1 }
^
(g: )
To better preserve source compatibility, teach the constraint
solver to try both the new forward scanning rule as well as the
backward scanning rule when matching a single, unlabeled trailing
closure. In the extreme case, where the unlabeled trailing closure
matches different parameters with the different rules, and yet both
produce a potential match, introduce a disjunction to explore both
possibilities.
Prefer solutions that involve forward scans to those that involve
backward scans, so we only use the backward scan as a fallback.
SE-0248 changes the backward-scan matching behavior for the unlabeled
trailing closure into a forward scan. In circumstances where this
could silently change the meaning of a call to a particular
function, i.e., when there are two defaulted closure parameters such
that a given closure to match either one of them, produce an warning
that describes the change in behavior. For example:
t4.swift:2:24: warning: since Swift 5.3, unlabeled trailing
closure argument matches parameter 'x' rather than parameter 'z'
trailingClosureSingle2 { $0 }
^
t4.swift:2:24: note: label the argument with 'z' to retain the
pre-Swift 5.3 behavior
trailingClosureSingle2 { $0 }
^
(z: )
t4.swift:2:24: note: label the argument with 'x' to silence this
warning for Swift 5.3 and newer
trailingClosureSingle2 { $0 }
^
(x: )
t4.swift:1:6: note: 'trailingClosureSingle2(x:y:z:)' contains
defaulted closure parameters 'x' and 'z'
func trailingClosureSingle2(x: (Int) -> Int = { $0 } , y: (Int) ->
Int = { $0 }, z: (Int) -> Int = { $0 }) {}
^ ~
This explains the (rare) case where SE-0286 silently changes the
meaning of a program, offering Fix-Its to either restore the
pre-SE-0286 behavior or silence the warning, as appropriate.
The change to the forward-scanning rule regressed some diagnostics,
because we no longer generated the special "trailing closure mismatch"
diagnostic. Reinstate the special-case "trailing closure mismatch"
diagnostic, but this time do so as part of the normal argument
mismatch diagnostics so it is based on type information.
While here, clean up the handling of missing-argument diagnostics to
deal with (multiple) trailing closures properly, so that we can (e.g)
suggest adding a new labeled trailing closure at the end, rather than
producing nonsensical Fix-Its.
And, note that SR-12291 is broken (again) by the forward-scan matching
rules.
Once the first argument for a variadic function-typed parameter has been
matched, allow an unlabeled trailing closure to match, rather than
banning all uses of the unlabeled trailing closure with variadic
parameters.
The "fuzzy" forward scan matching algorithm was only applied when there
was a single, unlabeled trailing closure, but was disabled in the
presence of multiple trailing closures. Extend the "fuzzy" match to
account for multiple trailing closures, by restricting the search for
"a later parameter that needs an argument" to stop when we find a
parameter that matches the first (labeled) trailing closure.
Introsuce a new "forward" algorithm for trailing closures where
the unlabeled trailing closure argument matches the next parameter in
the parameter list that can accept an unlabeled trailing closure.
The "can accept an unlabeled trailing closure" criteria looks at the
parameter itself. The parameter accepts an unlabeled trailing closure
if all of the following are true:
* The parameter is not 'inout'
* The adjusted type of the parameter (defined below) is a function type
The adjusted type of the parameter is the parameter's type as
declared, after performing two adjustments:
* If the parameter is an @autoclosure, use the result type of the
parameter's declared (function) type, before performing the second
adjustment.
* Remove all outer "optional" types.
For example, the following function illustrates both adjustments to
determine that the parameter "body" accepts an unlabeled trailing
closure:
func doSomething(body: @autoclosure () -> (((Int) -> String)?))
This is a source-breaking change. However, there is a "fuzzy" matching
rule that that addresses the source break we've observed in practice,
where a defaulted closure parameter precedes a non-defaulted closure
parameter:
func doSomethingElse(
onError: ((Error) -> Void)? = nil,
onCompletion: (Int) -> Void
) { }
doSomethingElse { x in
print(x)
}
With the existing "backward" scan rule, the trailing closure matches
onCompletion, and onError is given the default of "nil". With the
forward scanning rule, the trailing closure matches onError, and there
is no "onCompletion" argument, so the call fails.
The fuzzy matching rule proceeds as follows:
* if the call has a single, unlabeled trailing closure argument, and
* the parameter that would match the unlabeled trailing closure
argument has a default, and
* there are parameters *after* that parameter that require an argument
(i.e., they are not variadic and do not have a default argument)
then the forward scan skips this parameter and considers the next
parameter that could accept the unlabeled trailing closure.
Note that APIs like doSomethingElse(onError:onCompletion:) above
should probably be reworked to put the defaulted parameters at the
end, which works better with the forward scan and with multiple
trailing closures:
func doSomethingElseBetter(
onCompletion: (Int) -> Void,
onError: ((Error) -> Void)? = nil
) { }
doSomethingElseBetter { x in
print(x)
}
doSomethingElseBetter { x in
print(x)
} onError: { error in
throw error
}
Currently absence of `subtyping` is the only problem detected and diagnosed specifically
for `inout` parameters, but there could be type mismatches in `inout` positions as well
and we can use `argument-to-parameter mismatch fix to detect and diagnose them.
This PR migrates instance member on type and type member on instance diagnostics handling to use the new diagnostics framework (fixes) and create more reliable and accurate diagnostics in such scenarios.
- Attempting to construct class object using metatype value via
non-required initializer
- Referencing initializer of protocol metatype base
Both of the diagnostics are used by `AllowInvalidInitRef` fix.
* Make sure that base and unwrapped types aren't null
* Don't allocate `ForceOptional` fix if nothing has been unwrapped
in `simplifyApplicableFnConstraint`
* Add sugar reconstitution support to `FailureDiagnostic::resolveType`
* Format diagnostics a bit better
When the compiler fails to find an overload with suitable parameter or return types, it often attaches a note listing the available overloads so that users can find the one they meant to use. The overloads are currently ordered in a way that depends on the order they were declared, so swift-evolve would sometimes cause tests involving these diagnostics to fail.
This change emits the list in a textually-sorted order instead. The names were already being sorted as they were inserted into a std::set, so this shouldn’t significantly slow down the diagnostic.
This builds on initial commit which added `RelabelArguments` fix
to the solver that only supported `missingLabels` at that moment,
but now it supports all three posibilities - missing/extraneous and
incorrect labels.
When we determine that an optional value needs to be unwrapped to make
an expression type check, use notes to provide several different
Fix-It options (with descriptions) rather than always pushing users
toward '!'. Specifically, the errors + Fix-Its now looks like this:
error: value of optional type 'X?' must be unwrapped to a value of
type 'X'
f(x)
^
note: coalesce using '??' to provide a default when the optional
value contains 'nil'
f(x)
^
?? <#default value#>
note: force-unwrap using '!' to abort execution if the optional
value contains 'nil'
f(x)
^
!
Fixes rdar://problem/42081852.
For `use(self.init())`, target of RebindSelfInConstructorExpr should be
call expression instead of paren expression.
rdar://problem/41416911
Possibly: rdar://problem/41593987
