LLVM is gearing up to move to `std::endianness` and as part of that has
moved `llvm::support::endianness` to `llvm::endianness`
(bbdbcd83e6702f314d147a680247058a899ba261). Rename our uses.
This is phase-1 of switching from llvm::Optional to std::optional in the
next rebranch. llvm::Optional was removed from upstream LLVM, so we need
to migrate off rather soon. On Darwin, std::optional, and llvm::Optional
have the same layout, so we don't need to be as concerned about ABI
beyond the name mangling. `llvm::Optional` is only returned from one
function in
```
getStandardTypeSubst(StringRef TypeName,
bool allowConcurrencyManglings);
```
It's the return value, so it should not impact the mangling of the
function, and the layout is the same as `std::optional`, so it should be
mostly okay. This function doesn't appear to have users, and the ABI was
already broken 2 years ago for concurrency and no one seemed to notice
so this should be "okay".
I'm doing the migration incrementally so that folks working on main can
cherry-pick back to the release/5.9 branch. Once 5.9 is done and locked
away, then we can go through and finish the replacement. Since `None`
and `Optional` show up in contexts where they are not `llvm::None` and
`llvm::Optional`, I'm preparing the work now by going through and
removing the namespace unwrapping and making the `llvm` namespace
explicit. This should make it fairly mechanical to go through and
replace llvm::Optional with std::optional, and llvm::None with
std::nullopt. It's also a change that can be brought onto the
release/5.9 with minimal impact. This should be an NFC change.
Only return macros that are valid in their current position, ie. an
attached macro is not valid on a nominal.
Also return freestanding expression macros in code block item position
and handle the new freestanding code item macros.
Resolves rdar://105563583.
`getValue` -> `value`
`getValueOr` -> `value_or`
`hasValue` -> `has_value`
`map` -> `transform`
The old API will be deprecated in the rebranch.
To avoid merge conflicts, use the new API already in the main branch.
rdar://102362022
When an function has an async alternative, that should be preferred when we are completing in an async context. Thus, the sync method should be marked as not recommended if the current context can handle async methods.
rdar://88354910
Store whether a result is async in the `ContextFreeCodeCompletionResult` and determine whether an async method is used in a sync context when promoting the context free result to a contextual result.
rdar://78317170
When completing after `@`, record what kind of attributes are applicable here (property wrapper, result builder, global actor), mark types that are marked as property wrapper etc. as having a 'Convertible' type relation and mark all other types as having an invalid type relation.
rdar://78239501
Computing the type relation for every item in the code completion cache is way to expensive (~4x slowdown for global completion that imports `SwiftUI`). Instead, compute a type’s supertypes (protocol conformances and superclasses) once and write their USRs to the cache. To compute a type relation we can then check if the contextual type is in the completion item’s supertypes.
This reduces the overhead of computing the type relations (again global completion that imports `SwiftUI`) to ~6% – measured by instructions executed.
Technically, we might miss some conversions like
- retroactive conformances inside another module (because we can’t cache them if that other module isn’t imported)
- complex generic conversions (just too complicated to model using USRs)
Because of this, we never report an `unrelated` type relation for global items but always default to `unknown`.
But I believe this change covers the most common cases and is a good tradeoff between accuracy and performance.
rdar://83846531
Computing the type relation for every item in the code completion cache is way to expensive (~4x slowdown for global completion that imports `SwiftUI`). Instead, compute a type’s supertypes (protocol conformances and superclasses) once and write their USRs to the cache. To compute a type relation we can then check if the contextual type is in the completion item’s supertypes.
This reduces the overhead of computing the type relations (again global completion that imports `SwiftUI`) to ~6% – measured by instructions executed.
Technically, we might miss some conversions like
- retroactive conformances inside another module (because we can’t cache them if that other module isn’t imported)
- complex generic conversions (just too complicated to model using USRs)
Because of this, we never report an `unrelated` type relation for global items but always default to `unknown`.
But I believe this change covers the most common cases and is a good tradeoff between accuracy and performance.
rdar://83846531
Filter name for completion item is always used. Also, for cached items,
they are used multiple times for filtering. So precomputing and caching
it improves performance.
rdar://84036006
[CodeCompletion] Make ExpectedTypeContext a class with explicit getters/setters
This simplifies debugging because you can break when the possible types are set and you can also search for references to `setPossibleType` to figure out where the expected types are being set.
This allows makes the distinction between cachable and non-cachable properties cleaner and allows us to more easily compute contextual information (like type relations) for cached items later.
Now that arguments are marked up with whether they have a default or
not, clients may not need the extra call (that has no default
arguments). Add an option to allow not adding this item.
Resolves rdar://85526214.
"add inits to toplevel" and "call pattern heuristics" are only used in
code completion. Move them from LangOptions to CodeCompletionContext so
that they don't affect compiler arguments.
The various keyword/recommended/etc fields were parsed and added to
completions, but never actually plumbed through SourceKit. Since they're
still unused and the implementation is not particularly lightweight,
just remove for now.
Resolves rdar://82464220
* Implement 'getDiagnosticSeverity()' and 'getDiagnosticMessage()' on
'CodeCompletionResult'
* Differentiate 'RedundantImportIndirect' from 'RedundantImport'
* Make non-Sendable check respects '-warn-concurrency'
rdar://76129658
* 'CodeCompletionContext' now has 'requiresSourceFileInfo()' flag
* When 'true', 'SourceFiles' is populated.
* 'SourceFiles' is a list of pairs of a known module source
file path and its up-to-date-ness
* Clients (i.e. 'CodeCompletionConsumer') can retrieve it from
'CodeCompletionContext' in 'handleResults'
* Each completion item now has 'SourceFilePath' property
* C-APIs to get those informations
To describe fine grained priorities.
Introduce 'CodeCompletionFlair' that is a set of more descriptive flags for
prioritizing completion items. This aims to replace '
SemanticContextKind::ExpressionSpecific' which was a "catch all"
prioritization flag.
'InvalidAsyncContext' depends on the decl context. That may case
"sticky" not-recommended If it's cached for a non-async context.
To workaround this, stop checking 'InvalidAsyncContext' when collecting
completion items for caching. Also consistently use the 'SourceFile' as
the decl context to avoid decl context specific behavior.
rdar://78315441
struct Foo {
init(_ arg1: String, arg2: Int) {}
init(label: Int) {}
}
func test(strVal: String) {
_ = Foo(<HERE>)
}
In this case, 'strVal' was prioritized because it can use as an argument
for 'init(_:arg2:)'. However, argument labels are almost always
preferable, and if the user actually want 'strVal', they can input a few
characters to get it at the top. So we should always prioritize call
argument patterns.
rdar://77188260
Combine IsNotRecommended with NotRecommendedReason and improve the names
of the existing cases to more clearly identify the cases they apply to.
Now all not-recommended completions are required to have a reason.
Thanks to Ben Barham for spotting this:
sizeof(~static_cast<uint8_t>(...)) is 4, but we need to write a single
byte here. Thankfully this code was probably not being hit in the
current caching scheme, which only caches declarations.
Looking at the history, this code was broken by d8fbaa01eb, which was
fixing an MSVC warning in this code. Unfortunately I do not have access
to the version of MSVC to check if there is still a warning here or for
what.
func foo() {}
let a: Int = #^HERE^#
Previously, we marked 'foo()' as 'NotRecommented' because 'Void' doesn't
have any member hence it cannot be 'Int'. But it wass confusing with
'deprecated'.
Now that we output 'typerelation' which is 'invalid' in this case. So clients
can deprioritize results, or even filter them out.
rdar://problem/57726512
