Inference of type witnesses for associated types was previously
implemented as part of value witness matching in the constraint
solver. This led to a number of serious problems, including:
- Recursion problems with the solver hunting for a type witness,
which triggers more attemts to match value witnesses...
- Arbitrarily crummy attempts to break the recursion causing
type-check failures in fun places.
- Ordering dependencies abound: different results depending on which
value witnesses were satisfied first, failures because of the order
in which we attempted to infer type witnesses, etc.
This new implementation of type witness inference uses a separate pass
that occurs whenever we're looking for any type witness, and solves
all of the type witnesses within a given conformance
simultaneously. We still look at potential value witnesses to infer
type witnesses, but we match them structurally, without invoking the
constraint solver.
There are a few caveats to this implementation:
* We're not currently able to infer type witnesses from value
witnesses that are global operators, so some tricks involving global
operators (*cough* ~> *cough*) might require some manually-specified
type witnesses. Note that the standard library doesn't include any
such cases.
* Yes, it's another kind of solver. At simple one, fortunately.
On the other hand, this implementation should be a big step forward:
* It's far more predictable, order-invariant, and non-recursive.
* The diagnostics for failures to infer type witnesses have
improved.
Fixes rdar://problem/20598513.
Swift SVN r27616
(Note that this registry isn't fully enabled yet; it's built so that
we can test it, but has not yet taken over the primary task of
managing conformances from the existing system).
The conformance registry tracks all of the protocols to which a
particular nominal type conforms, including those for which
conformance was explicitly specified, implied by other explicit
conformances, inherited from a superclass, or synthesized by the
implementation.
The conformance registry is a lazily-built data structure designed for
multi-file support (which has been a problematic area for protocol
conformances). It allows one to query for the conformances of a type
to a particular protocol, enumerate all protocols to which a type
conforms, and enumerate all of the conformances that are associated
with a particular declaration context (important to eliminate
duplicated witness tables).
The conformance registry diagnoses conflicts and ambiguities among
different conformances of the same type to the same protocol. There
are three common cases where we'll see a diagnostic:
1) Redundant explicit conformance of a type to a protocol:
protocol P { }
struct X : P { }
extension X : P { } // error: redundant explicit conformance
2) Explicit conformance to a protocol that collides with an inherited
conformance:
protocol P { }
class Super : P { }
class Sub : Super, P { } // error: redundant explicit conformance
3) Ambiguous placement of an implied conformance:
protocol P1 { }
protocol P2 : P1 { }
protocol P3 : P1 { }
struct Y { }
extension Y : P2 { }
extension Y : P3 { } // error: ambiguous implied conformance to 'P1'
This happens when two different explicit conformances (here, P2 and
P3) placed on different declarations (e.g., two extensions, or the
original definition and other extension) both imply the same
conformance (P1), and neither of the explicit conformances imply
each other. We require the user to explicitly specify the ambiguous
conformance to break the ambiguity and associate the witness table
with a specific context.
Swift SVN r26067
Most tests were using %swift or similar substitutions, which did not
include the target triple and SDK. The driver was defaulting to the
host OS. Thus, we could not run the tests when the standard library was
not built for OS X.
Swift SVN r24504
We don't properly open up the existential to make this work, which
leads to an IRGen crash. Reject the uses of generics that would cause
such a crash rdar://problem/17491663.
Swift SVN r21946
When a non-final class satisfies a method requirement that returns
Self, it must do so with a method that also returns (dynamic)
Self. This ensures conformance will be inheritable, closing off an
awful type-safety hole <rdar://problem/16880016>. Other
non-contravariant uses of Self in the signatures of requirements cause
the protocol to be unusable by non-final classes.
I had to leave a tiny little gaping hole for the ~> operator, whose
removal is covered by <rdar://problem/17828741>. We can possibly put
this on firm footing with clever handling of generic witnesses, but
it's not important right now.
Swift SVN r20626
attribute is a "modifier" of a decl, not an "attribute" and thus shouldn't
be spelt with an @ sign. Teach the parser to parse "@foo" but reject it with
a nice diagnostic and a fixit if "foo" is a decl modifier.
Move 'dynamic' over to this (since it simplifies some code), and switch the
@optional and @required attributes to be declmodifiers (eliminating their @'s).
Swift SVN r19787
Better to describe how the protocol can be used than how it can't. Also include a mention of Self type requirements as a source of non-existentiability.
Swift SVN r19207
These types are often useless and confusing to users who expect to be able to use Sequence or Generator as types in their own right like in C# or Java. While we're here, relax the rules for self-conformance to admit methods returning 'Self'. Covariant return types should not actually prevent a protocol type from conforming to itself, and the stdlib makes particular use of protocols with 'init' requirements which implicitly return Self.
Swift SVN r18989
'Self' can be used within parameters whenever the corresponding
parameter in a subclass will be contravariant, and in result types
when the method returns dynamic Self. This also applies to subscript
indices. More of <rdar://problem/16996872>.
Swift SVN r18788
One difficulty in generating reasonable diagnostic data for type check failures has been the fact that many constraints had been synthesized without regard for where they were rooted in the program source. The result of this was that even though we would store failure information for specific constraints, we wouldn't emit it for lack of a source location. By making location data a non-optional component of constraints, we can begin diagnosing type check errors closer to their point of failure.
Swift SVN r18751
A protocol conformance of a class A to a protocol P can be inherited
by a subclass B of A unless
- A requirement of P refers to Self (not an associated type thereof)
in its signature,
+ *except* when Self is the result type of the method in P and the
corresponding witness for A's conformance to B is a DynamicSelf
method.
Remove the uses of DynamicSelf from the literal protocols, going back
to Self. The fact that the conformances of NSDictionary, NSArray,
NSString, etc. to the corresponding literal protocols use witnesses
that return DynamicSelf makes NSMutableDictionary, NSMutableArray,
NSMutableString, and other subclasses still conform to the
protocol. We also correctly reject attempts to (for example) create an
NSDecimalNumber from a numeric literal, because NSNumber doesn't
provide a suitable factory method by which any subclass can be literal
convertible.
Swift SVN r14204
As part of this, take away the poor attempt at recovering by adding an
explicit protocol conformance. The recovery mode isn't all that useful
in a system with only explicit conformance, and it messes with
diagnostics further down the line. We can bring it back later once
we're happy with explicit conformance checking.
Swift SVN r11503
There's a lot more work to do here, but start to categorize tests
along the lines of what a specification might look like, with
directories (chapters) for basic concepts, declarations, expressions,
statements, etc.
Swift SVN r9958
