We have to work with selectors quite often, so provide an efficient
representation for them. Switch ObjCAttr over to this representation,
which has the nice property that it efficiently represents implicit
@objc attributes with names and allows us to overwrite the Objective-C
name without losing all source information. Addresses
<rdar://problem/16478678>, and sets us up for dealing with selectors
better.
Swift SVN r16327
These bits are orthogonal to each other, so combine them into one, and diagnose attempts to produce a type that's both. Spot-fix a bunch of places this revealed by inspection that we would have crashed in SILGen or IRGen if blocks were be handled.
Swift SVN r16088
... and fix a few other bugs:
* always set the inherited protocols on the ProtocolDecl in the type checker,
so that we can remove a hack in ProtocolDecl::requiresClassSlow();
* diagnose DeclAttributes that are inverted when this is not allowed.
Swift SVN r15992
This allows us to consistently print the AST, no matter if it was just parsed
or deserialized.
Note that we still serialize the isObjC bit from Decl, because it can be set or
cleared by the typechecker. It is possible to have isObjC=true when there is
not attribute (the ObjC'ness was inferred), and it is possible that
isObjC=false while there is an attribute (when the attribute does not pass the
semantic check). While we can represent the former with an implicit attribute,
the latter is harder to represent (maybe with an invalid bit on the attribute?)
Swift SVN r15935
To generalize our serialization logic for more attributes, serialize
each DeclAttribute object in a separate bitcode record.
For simple declaration attributes (no arguments), all of this
serialization logic can be fully automatically generated, and is
done so in this patch. This currently includes @final, but will
expand over time.
To illustrate the plumbing end-to-end, move the serialization logic
for asmnmame over to the new mechanism.
Swift SVN r15933
More importantly, when writing substitution conformances /within the same
module/, use an "incomplete" form of the NormalProtocolConformance layout
that doesn't include any of the substitutions or defaulted definitions.
This avoids a serialization cycle when the witness for a protocol itself
ends up conforming to the protocol.
I couldn't come up with a reduced test cases, but both cases filed by Dave
now work.
<rdar://problem/16468715>
Swift SVN r15912
Language features like erasing concrete metatype
values are also left for the future. Still, baby steps.
The singleton ordinary metatype for existential types
is still potentially useful; we allow it to be written
as P.Protocol.
I've been somewhat cavalier in making code accept
AnyMetatypeType instead of a more specific type, and
it's likely that a number of these places can and
should be more restrictive.
When T is an existential type, parse T.Type as an
ExistentialMetatypeType instead of a MetatypeType.
An existential metatype is the formal type
\exists t:P . (t.Type)
whereas the ordinary metatype is the formal type
(\exists t:P . t).Type
which is singleton. Our inability to express that
difference was leading to an ever-increasing cascade
of hacks where information is shadily passed behind
the scenes in order to make various operations with
static members of protocols work correctly.
This patch takes the first step towards fixing that
by splitting out existential metatypes and giving
them a pointer representation. Eventually, we will
need them to be able to carry protocol witness tables
Swift SVN r15716
We'll need types to be convertible from multiple kinds of inouts, which currently can't be represented with protocol conformance since we only allow one protocol conformance per type per protocol. Instead just look for a magic "__inout_conversion" static method in the type; this is lame but easy, and inout conversions shouldn't be available outside of the stdlib anyway.
Swift SVN r15599
This patch adds in the necessary infrastructure for lazily deserializing
witness tables. This is done by following the same approach as the
deserialization/serialization of SILFunction.
Now if one calls SILModule::lookUpWitnessTable and the given witness table is a
definition, the SILModule will attempt to deserialize it from one of the other
modules.
Swift SVN r15403
Parse function declarations with the form
func murder inRoom(room: Int) weapon(Int) {}
where the function name ("murder") is separated from the parameter
names. This is the same style used in initializers, i.e.,
init withCString(cstr: CString) encoding(Encoding)
Swift SVN r15140
Instead of referencing them by name, serialize a reference to their
storage decl instead, then note which accessor to retrieve.
Uncovered by Chris's override work in r15051...thanks, Chris!
Swift SVN r15063
Previously, we were cloning the default arguments completely, which
meant code duplication (when inheriting within a module) or simply a
failure (when inheriting across modules). Now, we reference the
default arguments where we inherited them, eliminating the
duplication. Part of <rdar://problem/16318855>.
Swift SVN r15062
This will help with ensuring that we do not create multiple witness
table "definitions" one of which is null. That situtation yields an
IRGen assertion to be hit since the external declaration (in the guise
of a definition) has a different type from the actual deserialized
definition.
Swift SVN r14999
The driver infers the filename from the module file by replacing the extension,
and passes the explicit path to the swiftdoc file to the frontend. But there
is no option in the driver to control emission of swiftdoc (it is always
emitted, and name is always inferred from the swiftmodule name).
The swiftdoc file consists of a single table that maps USRs to {brief comment,
raw comment}. In order to look up a comment for decl we generate the USR
first. We hope that the performance hit will not be that bad, because most
declarations come from Clang. The advantage of this design is that the
swiftdoc file is not locked to the swiftmodule file, and can be updated,
replaced, and even localized.
Swift SVN r14914
In the short term, we need to be able to emit shared symbols for SILWitnessTables corresponding to Clang-imported modules, and soon, the generic specializer will need to be able to reference *_external witness tables deserialized from library modules.
Swift SVN r14887
Let ArchetypeType nested types and PotentialArchetypes be bound to concrete types in addition to archetypes. Constraints to outer context archetypes still suffer type-checker issues, but constraints to true concrete types should work now.
Swift SVN r14832
The standard library likes to have default definitions for associated types,
which is good. Often the /choice/ of default type, however, is a type that
(indirectly) conforms to the very protocol containing the associated type.
Rather than try to make sure everything is present all at once, just delay
the deserialization of the default definition until it's actually requested.
This does swell the size of AssociatedTypeDecl by two words. I've filed
<rdar://problem/16266669> to remind myself to try to reduce this.
Part of <rdar://problem/16257259>
Swift SVN r14809
Make the name lookup interfaces all take DeclNames instead of identifiers, and update the lookup caches of the various file units to index their members by both compound name and simple name. Serialized modules are keyed by identifiers, so as a transitional hack, do simple name lookup then filter the results by compound name.
Swift SVN r14768
Add __FUNCTION__ to the repertoire of magic source-location-identifying tokens. Inside a function, it gives the function name; inside a property accessor, it gives the property name; inside special members like 'init', 'subscript', and 'deinit', it gives the keyword name, and at top level, it gives the module name. As a bit of future-proofing, stringify the full DeclName, even though we only ever give declarations simple names currently.
Swift SVN r14710
to the archetypes of their generic parameter declarations.
This was a major inconsistency that was causing a lot of
problems with deserialized generic functions in the presence
of chained module files. In particular, what would happen is
that all the deserialized uses of the archetype would be
mapped to a freshly-created archetype, but the GenericParamList
would refer to GenericTypeParamDecls deserialized from
(potentially) a different ModuleFile and therefore using a
different archetype (since the deserializer assumes that
different modules never share archetypes).
The fix relies on processing a generic parameter list before
any references to the archetypes. Unfortunately, we have
bogus references to archetypes in substitution lists scattered
everywhere in SIL and the AST. I've introduced a really
gross hack where we allow archetypes to be created in that
way but just drop them (in favor of the parameter's archetype)
after processing the generic parameter list; this should
work as long as all the decontextualized references are
basically pointless, but like I said, it's gross, and I've
filed rdar://16240384 to clean it up.
Swift SVN r14694
alloc_ref_dynamic allocates an instance of a class type based on the
value in its metatype operand. Start emitting these instructions for
the allocating constructor of a complete object initializer (not yet
tested) and for the allocating constructor synthesized for an imported
Objective-C init method.
Still missing:
- IRGen still does the same thing as alloc_ref right now. That
change will follow.
- There are devirtualization opportunities when we know the value of
the metatype that would turn an alloc_ref_dynamic into an alloc_ref;
I'm not planning to do this optimization.
Swift SVN r14560
