#58786 (rdar://93030932) was failing because the `swift-frontend` invocations passed a `swiftExecutablePath` to `Invocation.parseArgs`. This caused the `ClangImporter` instance to point to a `clang` binary next to the `swift-frontend` executable while SourceKit used PATH to find `clang`. The clang executable next to `swift-frontend` doesn’t actually exist because `clang` lives in `llvm-linux-aarch64/bin` and `swift-frontend` lives in `swift-linux-aarch64/bin`.
So some checks for a minimum clang verison failed for the normal build (because the executable doesn’t actually exists) while they pass during the SourceKit build (which used `clang` from `PATH`). This in turn caused the `outline-atomics` to be enabled to the SourceKit clang compiler arguments but not the clang compiler arguments for a normal build and thus resulted in two separate module cache directories (which includes the enabled features in the module directory hash).
To fix this issue, also set the swift executable path for compiler invocations created from SourceKit.
Fixes#58786 (rdar://93030932)
The `SWIFT_COMPILER_VERSION` define is used to stamp a vendor’s version number into a Swift compiler binary. It can be queried from Swift code using `#if _compiler_version` and from Clang by using a preprocessor definition called `__SWIFT_COMPILER_VERSION`. These are unsupported compiler-internal features used primarily by Apple Swift.
In Swift 1.0 through 5.5, Apple Swift used a scheme for `SWIFT_COMPILER_VERSION` where the major version matched the embedded clang (e.g. 1300 for Apple Clang 13.0.0) and the minor version was ignored. Starting in Swift 5.6, Apple Swift started using major and minor version numbers that matched the Swift.org version number. This makes them easier to understand, but it means that version 1300.0.x was followed by version 5.6.x. Not only did version numbers go backwards, but also the old logic to ignore minor versions was now a liability, because it meant you would not be able to target a change to 5.7.x compilers but not 5.6.x compilers.
This commit addresses the problem by:
* Modifying the existing `#if _compiler_version(string-literal)` feature so it transforms the major version into a major and minor that will compare correctly to new version numbers. For instance, “1300.*” is transformed into “1.300”, which will compare correctly to a “5.6” or “5.7” version even if it doesn’t really capture the fact that “1300” was a Swift 5.5 compiler. As a bonus, this allows you to use the feature to backwards-compatibly test new compilers using the existing feature: “5007.*” will be seen by compilers before 5.7 as an unknown future version, but will be seen by 5.7 compilers as targeting them.
* Modifying the `__SWIFT_COMPILER_VERSION` clang define similarly so that, to preprocessor conditions written for the old scheme, a 5.7 compiler will appear to have major version 5007.
* Adding a new variant of `#if _compiler_version` with the same syntax as `#if swift` and `#if compiler`—that is, taking a comparison operator and a bare set of dotted version numbers, rather than a string literal. Going forward, this will be how version checks are written once compatibility with compilers before this change is no longer a concern.
These changes are only lightly tested because tests have to work without any compiler version defined (the default in most configurations), but I’ve tested what I can.
Fixes rdar://89841295.
I wrote out this whole analysis of why different existential types
might have the same logical content, and then I turned around and
immediately uniqued existential shapes purely by logical content
rather than the (generalized) formal type. Oh well. At least it's
not too late to make ABI changes like this.
We now store a reference to a mangling of the generalized formal
type directly in the shape. This type alone is sufficient to unique
the shape:
- By the nature of the generalization algorithm, every type parameter
in the generalization signature should be mentioned in the
generalized formal type in a deterministic order.
- By the nature of the generalization algorithm, every other
requirement in the generalization signature should be implied
by the positions in which generalization type parameters appear
(e.g. because the formal type is C<T> & P, where C constrains
its type parameter for well-formedness).
- The requirement signature and type expression are extracted from
the existential type.
As a result, we no longer rely on computing a unique hash at
compile time.
Storing this separately from the requirement signature potentially
allows runtimes with general shape support to work with future
extensions to existential types even if they cannot demangle the
generalized formal type.
Storing the generalized formal type also allows us to easily and
reliably extract the formal type of the existential. Otherwise,
it's quite a heroic endeavor to match requirements back up with
primary associated types. Doing so would also only allows us to
extract *some* matching formal type, not necessarily the *right*
formal type. So there's some good synergy here.
Creating a mangle-node tree is annoying, but it's much better
than trying to reproduce the mangling logic exactly.
Also, add support for mangling some existential types. The
specifier for parameterized protocol types has been future-proofed
against the coming change to include the associated type names
in the mangling.
Queue up diagnostics when lexing, waiting until
`Lexer::lex` is called before emitting them. This
allows us to re-lex without having to deal with
previously invalid tokens.
We need to run SILGen for diagnostics (to actually get all diagnostics).
All non-completion requests share an AST and thus they too run SILGen.
Any lazy typechecking run in SILGen assumes that it succeeds.
Cancellation can cause typechecking to fail here though, since we simply
check the flag and error if it's set. This unfortunately has the ability
to cause any any number of crashes since various invariants in SILGen
are then broken.
Disable cancellation of in-flight non-completion requests for now until
we have a proper fix in place.
Resolves rdar://91251017.
Moved the _gCRAnnotations declarations to their own object module,
which will help to avoid duplicate symbol problems (at least with .a
files).
Also tweaked things to make it so that the demangler and runtime
versions of the message setting code will interoperate (and so that
they'll interoperate better with other implementations that might
creep in from somewhere, like the one in LLVMSupport).
rdar://91095592
The immediate use case is only concretely-constrained existential
types, which could use a much simpler representation, but I've
future-proofed the representation as much as I can; thus, the
requirement signature can have arbitrary parameters and
requirements, and the type can have an arbitrary type as the
sub-expression. The latter is also necessary for existential
metatypes.
The chief implementation complexity here is that we must be able
to agree on the identity of an existential type that might be
produced by substitution. Thus, for example, `any P<T>` when
`T == Int` must resolve to the same type metadata as
`any P<Int>`. To handle this, we identify the "shape" of the
existential type, consisting of those parts which cannot possibly
be the result of substitution, and then abstract the substitutable
"holes" as an application of a generalization signature. That
algorithm will come in a later patch; this patch just represents
it.
Uniquing existential shapes from the requirements would be quite
complex because of all the symbolic mangled names they use.
This is particularly true because it's not reasonable to require
translation units to agree about what portions they mangle vs.
reference symbolically. Instead, we expect the compiler to do
a cryptographic hash of a mangling of the shape, then use that
as the unique key identifying the shape.
This is just the core representation and runtime interface; other
parts of the runtime, such as dynamic casting and demangling
support, will come later.
Some parts of the type metadata system are difficult to unit-test
because they rely on structures that contain relative references,
which the C compiler cannot generate. We have traditionally just
relied on integration testing with the compiler. For constrained
existentials, I wanted to do better, so I spent a few days hacking
up this little system which can generate graphs of objects with
relative references to one another.
Currently it's missing the ability to generate a lot of things
which I didn't need in order to adequately test the metadata
system for constrained existentials.
Crash reporter integration was only enabled for iOS. Enable it for
any Darwin platform, but disable it for the minimal build.
Also fix up a couple of issues that popped up when it was enabled.
rdar://89139049
Implicit casts are allowed to provide cast type directly without
a type repr, adjust solution application logic to handle this just
like constraint generator does. Also fix a couple of places where
declaration context from constraint system was used instead
of one associated with the expression rewriter.
Implicit casts are allowed to be constructed with a type, instead
of a type repr. Constraint generation should honor that, and fallback
to using cast type when repr is was not given.
I wanted a bit vector that I could use as a compact sorted
set of enum values: an inline-allocated, fixed-size array
of bits supporting efficient and convenient set operations
and iteration.
The C++ standard library offers std::bitset, but the API
is far from ideal for this purpose. It's positioned as
an abstract bit-vector rather than as a set. To use it
as a set, you have to turn your values into indices, which
for enums means explicitly casting them all in the caller.
There's also no iteration operation, so to find the
elements of the set, you have to iterate over all possible
indices, test whether they're in the set, and (if so)
cast the current index back to the enum. Not only is that
much more awkward than normal iteration, but it's also
substantially less efficient than what you can get by
counting trailing zeroes in a mask.
LLVM and Swift offer a number of other bit vectors, but
they're all dynamically allocated because they're meant
to track arbitrary sets. That's not a non-starter for my
use case, which is in textual serialization and so rather
slow anyway, but it's also not very hard to whip together
the optimal data structure here.
I have committed the cardinal sin of C++ data structure
design and provided the operations as ordinary methods
instead of operators.
This separates it from `libSwiftScan` and allows us to build this library without building much of the rest of the compiler.
Also refactor `utils/build-parser-lib` into `utils/build-tooling-libs` which builds both SwiftSyntaxParser and SwiftStaticMirror.
A task can be in one of 4 states over its lifetime:
(a) suspended
(b) enqueued
(c) running
(d) completed
This change provides priority inversion avoidance support if a task gets
escalated when it is in state (a), (c), (d).
Radar-Id: rdar://problem/76127624
Each trace point is declared as a function in the new `Tracing.h` header. These functions are called from the appropriate places in the concurrency runtime.
On Darwin, an implementation of these functions is provided which uses the `os/signpost.h` API to emit signpost events/intervals.
When the signpost API is not available, no-op stub implementations are provided. Implementations for other OSes can be provided by providing implementations of the trace functions for that OS.
rdar://81858487
This change has two parts to it:
1. Add in a new interface (addStatusRecordWithChecks) for adding task
status records that also takes in a function ref. This function ref will
be used to evaluate if current state of the parent task has any changes
that need to be propagated to the child task that has been created.
This is necessary to prevent the following race between task creation
and concurrent cancellation and escalation:
a. Parent task create child task. It does lazy relaxed loads on its own
state while doing so and propagates this state to the child.
b. Child task is created but has not been attached to the parent
task/task group.
c. Parent task gets cancelled by another thread.
d. Child task gets linked into the parent’s task status records but no
reevaluation has happened to account for changes that might have happened to
the parent after (a).
2. Move status record management functions from the
Runtime/Concurrency.h to TaskPrivate.h. Remove any corresponding
overrides that are no longer needed. Remove unused tryAddStatusRecord
method whose functionality is provided by addStatusRecordWithChecks.
Radar-Id: rdar://problem/86347801
When looking for a Swift module on disk, we were scanning all module search paths if they contain the module we are searching for. In a setup where each module is contained in its own framework search path, this scaled quadratically with the number of modules being imported. E.g. a setup with 100 modules being imported form 100 module search paths could cause on the order of 10,000 checks of `FileSystem::exists`. While these checks are fairly fast (~10µs), they add up to ~100ms.
To improve this, perform a first scan of all module search paths and list the files they contain. From this, create a lookup map that maps filenames to the search paths they can be found in. E.g. for
```
searchPath1/
Module1.framework
searchPath2/
Module1.framework
Module2.swiftmodule
```
we create the following lookup table
```
Module1.framework -> [searchPath1, searchPath2]
Module2.swiftmodule -> [searchPath2]
```
We remove the existing `swift_reflection_iterateAsyncTaskAllocations` API that attempts to provide all necessary information about a tasks's allocations starting from the task. Instead, we split it into two pieces: `swift_reflection_asyncTaskSlabPointer` to get the first slab for a task, and `+swift_reflection_asyncTaskSlabAllocations` to get the allocations in a slab, and a pointer to the next slab.
We also add a dummy metadata pointer to the beginning of each slab. This allows tools to identify slab allocations on the heap without needing to locate every single async task object. They can then use `swift_reflection_asyncTaskSlabAllocations` on such allocations to find out about the contents.
rdar://82549631
Previously, `SwiftASTManager` and `SlowRequestSimulator` maintained their own list of in-progress cancellation tokens. With code completion cancellation coming up, there would need to be yet another place to track in-progress requests, so let’s centralize it.
While at it, also support cancelling requests before they are scheduled, eliminating the need for a `sleep` in a test case.
The current implementaiton leaks tiny amounts of memory if a request is cancelled after if finishes. I think this is fine because it is a pretty nieche case and the leaked memory is pretty small (a `std::map` entry pointing to a `std::function` + `bool`). Alternatively, we could require the client to always dispose of the cancellation token manually.
