Clang now accepts an option -fapinotes-swift-version=XX that describes
which Swift version to use when applying API notes to the Clang
AST. Pass this option down to Clang based on the Swift major version
number (e.g., 3 or 4), allowing API notes to introduce
Swift-version-specific behavior.
Fixes rdar://problem/28617631.
Drop the ImportNameSwiftContext, which was merely used to pass around
some global state. Instead, pass the relevant parts
directly. Additionally, remove the 2-phase initialization of a
NameImporter from the module writers, as they are only invoked once
per Clang instance anyways, we can hold a local instance. Also
streamlines some APIs.
NFC
During normal import, the name for a foreign entity may be requested
many times. As we have more and more complex logic to compute these
new names, the overhead of potential re-computations adds up. This
cache is hit about 1/3rd of the time when running a simple
CoreGraphics test case.
NFC
Banish the abomination that is clangSemaOverride, a previously
necessary evil. When building the module caches, different Clang
instances will be spawned than the one used by the normal
importer. Since we want to reuse code and get the same name both ways,
this meant threading through alternative clang Semas and preprocessors
throughtout, some of the time. This broke the abstraction and
encapsulation of the Impl, complicated the programming model, and
otherwise made effective caching hard.
Now that we’ve done enough ImportName refactoring, we can create a
NameImporter per Clang instance, and encapsulate naming therein. We
can now remove the sema overrides, as we have already done to the
preprocessor overrides.
This shifts the 2-phase initialization problem to the Impl and the
Clang module writers.
NFC
Delay initialization of the EnumInfoCache until a Clang instance is
ready, simplifying its interface and allowing us to finally make this
per-Clang-instance. This will allow us to further de-couple ImportName
from the importer imply, as well as allow us to use a more efficient
and simpler caching mechanism. It is now owned by the NameImporter.
NFC.
Move SwiftLookupTableReader/Writer into .cpp file, as they are
irrelevant to the rest of the importer, and this allows us to make
them more ImportName aware in the future. Pull more code out of
ClangImporter.cpp.
NFC.
SwiftNameLookupExtension and ClangImporter::Implementation were
friends, but as time goes on they have drifted apart. As part of the
ImportName refactoring, these are being decoupled to facilitate
multiple-name importing, and fight the existing false encapsulation
present in the Impl.
SwiftNameLookupExtension is now spun off into its own entity, and can
evolve to have and use its own de-coupled NameImporter.
Finally, SwiftNameLookupExtension no longer stors a reference to
ClangImporter::Implementation, a false encapsulation. Instead, it uses
a NameImporter of its own, and merely keeps a reference to the lookup
table map.
Future directions include making the NameImporter specific to a Clang
instance, meaning that we can effectively cache Clang entities better
(and not resort to EnumInfoCache's stringly keyed caches). We can also
then drop all notions of a "ClangOverride" and other messy phase
entanglement.
NFC
Change more methods on the Impl to be static, passing down an
ASTContext if necessary. While this does ugly up the functions
slightly with the extra parameter, it decouples them from the Impl,
which has been a false abstraction when building the lookup
tables. These can all be moved to a more appropriate place in the
future.
NFC
addEntryToLookupTable is one of the laggards holding back the
Impl-free Clang lookup table effort. Make a static variant, now that
we have the convenient NameImporter.
NFC
Name Importer wraps references to the Swift context and other needed
encapsulation. The eventual goal is to allow name lookup tables to
live separately from the Impl, and do better Clang-instance-based
caching in the future, as well as general separation of concerns.
NFC
Enable newly-introduced Clang functionality to load API notes stored
alongside the headers for a module or, for a framework, in an APINotes
subdirectory within the framework. Fixes rdar://problem/28455644.
This flag switches the "effective language version" of the compiler,
at least to any version supported (as of this change: "3" or "3.0").
At the moment nothing uses it except the language version build
configuration statements (#if swift(...)) and various other places
that report, encode, or otherwise check version numbers.
In the future, it's intended as scaffolding for backwards compatibility.
Fixes SR-2582
By refactoring out PlatformAvailability from the ClangImporter, we can
more easily refactor out isUnavailableInSwift from the impl, which
will free us up to do more flexible import naming.
Introduces new files ClangAdapter.h/cpp, which will serve as a
convenient place to put code reasoning about Clang details. Refactors
out most Clang-related is*, has*, and get* methods from the
ImporterImpl. In the future, an adapter class could help serve to
seperate the concerns of the importer from the details of how to
correctly use Clang APIs.
Refactors much of the import name determination from ClangImporter.cpp
into ImportName.cpp.
This is not a perfect separation, but it's a step in the right
direction. Long term, we'd like to get as much functionality off of
the ClangImporter::Implementation class and onto more specialized and
seperable (and stateless!) components.
Refactors out some definitions and types from the
ClangImporter::Implementation into a new component ImportName. Future
work will include more separation and finally some redesigning of name
determination components.
We need to enable blocks support in the ClangImporter on
non-Darwin platforms for libdispatch (and transitively foundation).
The simplest way to do this is to just enable blocks unconditionally.
Also enable two Linux tests that are no longer XFAILS
...instead of picking one definition arbitrarily. This comes from the
new "lookup table" design in Swift 3---we no longer just look for any
"visible" (imported) macro definition, but instead need to know them
up front. This works fine when there's only one definition per module,
but for modules like 'OpenGL' on macOS, with mutually-exclusive
submodules 'GL' and 'GL3', the compiler was arbitrarily deciding that
all of the macros the submodules had in common belonged to 'GL'.
The new model tries to decide if it's possible for two modules to be
imported separately, and keeps both macro entries if possible, only
deduplicating equivalent definitions at the last minute (when
importing into Swift). This /still/ doesn't perfectly match the
behavior you'd get in C, where a submodule and its parent module could
theoretically have conflicting definitions and you'd be fine as long
as you only imported one of them, but hopefully (a) it's close enough,
and (b) nobody is doing that. (The Swift compiler will prefer the
definition in the parent module even if the submodule is the only one
imported.)
rdar://problem/26731529
* [ClangImporter] Remove importer-based NS stripping.
As Tony puts it, in the end we wound up with more Foundation
declarations imported as members or keeping "NS" than those that
dropped it, and any further decisions will be made on a case-by-case
basis. Move all of the existing cases of prefix-stripping into
Foundation's API notes and drop the logic from the compiler.
Tested by dumping the generated interface for Foundation and its
submodules for both macOS and the iOS simulator, and comparing the
results. A few cases did slip through here because of the interaction
between "SwiftName" and "Availability: nonswift".
The next commit will re-add "NS" to some stragglers that we missed.
rdar://problem/26880017
* APINotes: Add "NS" back to a few types.
NSKeyedUnarchiverDelegate
NSKeyedArchiverDelegate
NSTextCheckingTypes
NSBinarySearchingOptions
NSEnumerationOptions
NSSortOptions
More rdar://problem/26880017
* Remove now-redundant SwiftNames from API notes.
No change observed in the generated interface of Foundation and its
submodules.
Finishes rdar://problem/26880017.
A given Objective-C error enum, which is effectively an NS_ENUM that
specifies its corresponding error domain, will now be mapped to an
ErrorProtocol-conforming struct that wraps an NSError, much like
NSCocoaError does. The actual enum is mapped to a nested "Code"
enum. For example, CoreLocation's CLError becomes:
struct CLError : ErrorProtocol {
let _nsError: NSError
// ...
@objc enum Code : Int {
case ...
}
}
This implements bullet (2) in the proposed solution of SE-0112, so
that Cocoa error types are mapped into structures that maintain the
underlying NSError to allow more information to be extracted from it.
We map clang::AvailabilityAttr::getReplacement() to
swift::AvailableAttr::Rename, transforming the replacement
name using by looking up the named Clang replacement, and
importing its name.
Fixes <rdar://problem/26301742>.
Due to swift_name and swift_newtype, we are frequently importing onto
different contexts. This was confusing the fixit logic for unavailable
swift2 names, as we were trying to use Clang names when the Swift name
might be totally different (and even a nested type). This change has a
two-fold effect:
1) Globals who are imported onto swift_newtype-ed typedefs should be
considered ImportAsMember.
2) When printing out the name of an ImportAsMember Swift 3 decl, we
need to print out a fully qualified context, which also uses the
Swift names, not the Clang names.
More generally, an unavailable initializer shouldn't stomp on an
available initializer, because it's possible that (for example) a
designated initializer will be unavailable but a factory initializer
will be available, so one still construct objects of that type.
Fixes rdar://problem/26238032.
Otherwise, the lookup table for "CGColor" has two entries, because of
this:
typedef struct CGColor *CGColorRef;
and that interferes with our ability to import things as members of
"CGColor" (as opposed to "CGColorRef"), which affects the fix-its we
generate when you try to use the non-member form.
This isn't necessarily the best long-term solution (as noted in the
FIXME) but it is expedient and won't break any current users.
More rdar://problem/26347297
Previously we imported a Core Foundation type "CCFooRef" as "CCFoo",
but also provided a typealias "CCFooRef". In Swift 3, we decided to
mark "CCFooRef" unavailable to force developers to consistently use
"CCFoo". Now that we have infrastructure to mark /all/ renamed
declarations as unavailable, just use that to track the renaming,
i.e. pretend that "CCFooRef" was the "Swift 2" name for the type.
This doesn't change the conflict resolution behavior: if there's
another name "CCFoo" in the same module, the CF type will be
imported as just "CCFooRef".
Groundwork cleanup for rdar://problem/26347297, which notes that our
import-as-member fix-its use the "Ref" names rather than the short
names.
Previously, we had hacks in place to eagerly emit everything in
the global ExternalDefinitions list. These can now be removed,
at least at the IRGen layer.
More specifically, don't include declarations of methods and properties
in the list of "all imported Objective-C members" if said method or
property is in a generated header. (We actually key off of whether the
enclosing class, protocol, or category is marked as coming from Swift,
but since users aren't supposed to modify generated headers themselves
it's much the same thing.)
This previously caused a crash because we tried to import a Clang member
onto a Swift decl in order to provide the particular member on AnyObject.
rdar://problem/25955831 and probably also rdar://problem/25828987, which
deals with the fix-it to migrate to #selector. (We do an AnyObject-like
lookup to find out which class likely implements the specified selector.)
Leftover bits of SE-0055. Now that pointer nullability is reflected
in the type system, we can properly import throwing methods with
non-object pointer return types.
Note that Swift still won't let you declare them. That's coming next.
