Once we compute a generic signature from a generic signature builder,
all queries involving that generic signature will go through a separate
(canonicalized) builder, and the original builder can no longer be used.
The canonicalization process then creates a new, effectively identical
generic signature builder. How silly.
Once we’ve computed the signature of a generic signature builder, “register”
it with the ASTContext, allowing us to move the existing generic signature
builder into place as the canonical generic signature builder. The builder
requires minimal patching but is otherwise fully usable.
Thanks to Slava Pestov for the idea!
Funnel all places where we create a generic signature builder to compute
the generic signature through a single entry point in the GSB
(`computeGenericSignature()`), and make `finalize` and `getGenericSignature`
private so no new uses crop up.
Tighten up the signature of `computeGenericSignature()` so it only works on
GSB rvalues, and ensure that all clients consider the GSB dead after that
point by clearing out the internal representation of the GSB.
Given an existential type E, this creates a new canonical
generic signature <T : E>.
If E is a protocol composition containing members of protocol,
class or AnyObject type, the resulting generic signature will
split up the T : E requirement into a canonical and minimal
set of conformance, superclass and layout requirements.
If/when we implement generalized existentials, existential
types will be defined in terms of an arbitrary set of
generic requirements applied to a single generic parameter;
at that point, this method will be replaced with a generic
signature stored inside the type itself.
For now, this will be used to simplify some logic in the
SILOptimizer.
This implementation required a compromise between parser
performance and AST structuring. On the one hand, Parse
must be fast in order to keep things in the IDE zippy, on
the other we must hit the disk to properly resolve 'canImport'
conditions and inject members of the active clause into the AST.
Additionally, a Parse-only pass may not provide platform-specific
information to the compiler invocation and so may mistakenly
activate or de-activate branches in the if-configuration decl.
The compromise is to perform condition evaluation only when
continuing on to semantic analysis. This keeps the parser quick
and avoids the unpacking that parse does for active conditions
while still retaining the ability to see through to an active
condition when we know we're moving on to semantic analysis anyways.
A protocol extension initializer creates a new instance of the
static type of Self at the call site.
However a convenience initializer in a class is expected to
initialize an instance of the dynamic type of the 'self' value,
because convenience initializers can be inherited by subclasses.
This means that when a convenience initializer delegates to a
protocol extension initializer, we have to substitute the
'Self' type in the protocol extension generic signature with
the DynamicSelfType, and not the static type.
Since the substitution is formed from the type of the 'self'
parameter in the class convenience initializer, the solution is
to change the type of 'self' in a class convenience initializer
to DynamicSelfType, just like we do for methods that return
'Self'.
This fixes cases where we allowed code to type check that
should not type check (if the protocol extension initializer
has 'Self' in contravariant position, and we pass in an
instance of the static type).
It also fixes a miscompile with valid code -- if the protocol
extension initializer was implemented by calling 'Self()',
it would again use the static type and not the dynamic type.
Note that the SILGen change is necessary because Sema now creates
CovariantReturnExprs that convert a static class type to
DynamicSelfType, but the latter lowers to the former at the
SIL level, so we have to peephole away unnecessary unchecked_ref_cast
instructions in this case.
Because this change breaks source compatibility, it is guarded
by a '-swift-version 5' check.
I don't have reduced test cases. The original test cases
were a series of frontend invocations in -parse-stdlib
mode.
While the original bugs seem to have been fixed, while
verifying I found a few places where we weren't checking
for null decls property in the ASTContext.
Probably not too useful to check this in, but I don't see it
causing any harm, either.
To remove some callers of 'is<InOutType>' after Sema, start using what will soon be a structural invariant - the only expressions that can possibly have 'inout' type are semantically InOut expressions.
As a step toward eliminating the single input type
representation of function parameters, add more constraints on that
input type. It can be one of:
* A tuple type, for multiple parameters,
* A parenthesized type, for a single parameter, or
* A type variable type, for specific cases in the type checker
Enforce these constraints for *canonical* types as well, so the
canonical form of:
typealias MyInt = Int
typealias MyFuncType = (MyInt) -> Int
is now:
(Int) -> Int
rather than:
Int -> Int
This affects canonicalization of FunctionType and
GenericFunctionType. Enchance both, as well as their Can*Type
counterparts, with "get" operators that take an array of
AnyFunctionType::Param, and start switching a few clients over to this
new, preferred API.
AnyFunctionType::Param carries around information about decomposed
parameters now. Information about default arguments must be computed
separately with swift::computeDefaultMap.
In anticipation of removing this bit, move it from the
recursive type property into TupleType - its only real
user. This necessitates uglifying a bit of logic in the
short term that used to speak broadly of materializability
to instead speak about LValues and Tuples of InOut values
independently.
With the introduction of special decl names, `Identifier getName()` on
`ValueDecl` will be removed and pushed down to nominal declarations
whose name is guaranteed not to be special. Prepare for this by calling
to `DeclBaseName getBaseName()` instead where appropriate.
This changes `getBaseName()` on `DeclName` to return a `DeclBaseName`
instead of an `Identifier`. All places that will continue to be
expecting an `Identifier` are changed to call `getBaseIdentifier` which
will later assert that the `DeclName` is actually backed by an
identifier and not a special name.
For transitional purposes, a conversion operator from `DeclBaseName` to
`Identifier` has been added that will be removed again once migration
to DeclBaseName has been completed in other parts of the compiler.
Unify approach to printing declaration names
Printing a declaration's name using `<<` and `getBaseName()` is be
independent of the return type of `getBaseName()` which will change in
the future from `Identifier` to `DeclBaseName`
Instead, introduce a new hasDependentMember() recursive property.
The only place that cares about this is associated type inference,
where I changed all existing hasTypeParameter() checks to instead
check (hasTypeParameter() || hasDependentMember()). We could
probably refine this over time and remove some of the
hasTypeParameter() checks, but I'm being conservative for now.
Fixes <https://bugs.swift.org/browse/SR-4575> and
<rdar://problem/31603113>.
A `GenericSignatureBuilder` only needs to be finalized when we need to
compute a generic signature from it or otherwise require diagnostics.
Canonical generic signature builders don’t need either of those, so unless we
are performing the expensive idempotency checking, only do the minimal work
to “process delayed requirements”.
This gives a 3x speedup in type-checking time for the “trivial” example
let x = [1, 2]
because we do a lot less work on deserialization of generic environments.
Part of rdar://problem/32116933.
Rather than having `ASTContext:: getOrCreateCanonicalGenericEnvironment()`
ask its `GenericSignatureBuilder` parameter recompute the generic signature
at nontrivial cost, just pass the known signature through.
When casting from an object type to a bridged Swift value type, classifyDynamicCast would use the cast classification for the target type's bridged object type, which would be trivially WillSucceed for thinks like NSNumber-to-Int or NSError-to-SomeError, even though the bridging itself could fail. Fixing this fixes SR-2920|rdar://problem/31404281.
Like NSObject, CFType has primitive operations CFEqual and CFHash,
so Swift should allow those types to show up in Hashable positions
(like dictionaries). The most general way to do this was to
introduce a new protocol, _CFObject, and then have the importer
automatically make all CF types conform to it.
This did require one additional change: the == implementation that
calls through to CFEqual is in a new CoreFoundation overlay, but the
conformance is in the underlying Clang module. Therefore, operator
lookup for conformances has been changed to look in the overlay for
an imported declaration (if there is one).
This re-applies 361ab62454, reverted in
f50b1e73dc, after a /very/ long interval
where we decided if it was worth breaking people who've added these
conformances on their own. Since the workaround isn't too difficult---
use `#if swift(>=3.2)` to guard the extension introducing the
conformance---it was deemed acceptable.
https://bugs.swift.org/browse/SR-2388
Replace `NameOfType foo = dyn_cast<NameOfType>(bar)` with DRY version `auto foo = dyn_cast<NameOfType>(bar)`.
The DRY auto version is by far the dominant form already used in the repo, so this PR merely brings the exceptional cases (redundant repetition form) in line with the dominant form (auto form).
See the [C++ Core Guidelines](https://github.com/isocpp/CppCoreGuidelines/blob/master/CppCoreGuidelines.md#es11-use-auto-to-avoid-redundant-repetition-of-type-names) for a general discussion on why to use `auto` to avoid redundant repetition of type names.
If SubstitutionMap is asked to form a substitution for a generic
parameter that has been made concrete by the generic signature,
substitute into the concrete type. This allows us to better deal with
non-canonical types.
Add CGFloat (and other types that this applies to) as a trivial type even after
the one-time initialization of the ForeignRepresentableCache.
This allows
let str = ""
import Foundation
let pt = CGPoint(x: 1.0, y: 2.0)
to work.
Before we would populate the cache the first time on the first line "let str =
..." and because the CoreGraphics module was not loaded we would not add CGFloat
as a trivial type. When we come to query for CGFloat on the third line we would
return NSNumber instead of CGFloat as a type and that would crash IRGen.
rdar://31610342
