This is an attempt to solving
https://github.com/swiftlang/swift/issues/72199.
As I've already lined out in the issue, this seems to be a two-fold
problem.
The first one was a rather easy fix, as I've seen similar approaches in
different parts of the codebase. It is pretty much just un-currying the
generic function.
The second problem was, that `DependentMemberType`s were counted towards
the generic-parameter-should-only-be-mentioned-once-limit even though
you cannot infer the generic type from the dependent member type.
This mostly makes it easier to test dependency corner cases. The analysis still
doesn't recognize UnsafeRawPointer.init(), so regular users still need to use
_overrideLifetime.
Fixes rdar://137608270 ([borrows] Add Builtin.addressof() support
for @addressable arguments)
Both the syntax and relative order of the LLVM `nocapture` parameter
attribute changed upstream in
https://github.com/llvm/llvm-project/pull/123181.
To reduce conflicts with rebranch, adjust FileCheck patterns to expect
both syntaxes and orders anywhere the presence of the attribute is not
critical to the test. These changes are temporary and will be cleaned
up once rebranch is merged into main.
statusCompletePendingReadyWaiting(), offer(), and poll() did a one-off compare_exchange_strong which could fail if the group was concurrently cancelled. Put these into loops so that they are retried when needed.
DiscardingTaskGroup creation passed the group result type as the task result type. waitAll() would then use the group result type when collecting task results. Since the task result type is always Void in this case, this would overflow the result buffer if the group result type was larger. This often works as it writes into the free space of the task allocator, but can crash if it happens to be at the end of a page or the group result type is particularly large.
rdar://151663730
Ideally this would also update the `--version` output to be overridden
by `SWIFT_TOOLCHAIN_VERSION`, but unfortunately various tools rely on
the current format (eg. swift-build).
`HAVE_SYS_RESOURCE_H` was removed in
https://github.com/llvm/llvm-project/pull/123288, so this header is no
longer included at this particular location on rebranch, which breaks
the Linux build, where it is not transitively included either.
Use the same condition as in the use site
(`getChildrenMaxResidentSetSize`) instead.
Also, don't wrap `HAVE_GETRUSAGE` in `defined()` in case it does get
defined to 0.
If key or value of a literal collection expression doesn't conform
to protocol(s) expected by the contextual existential type, let's
diagnose that via a tailed collection mismatch fix instead of a
generic conformance one.
Resolves: rdar://103045274
`@autoclosure` is associated with a parameter, we use argument mismatch fix
to diagnose missing explicit calls as well as any mismatches in that position.
Resolves: rdar://110527062
If the argument has an extra `?` or `!`, let's not attempt to force
optional (because it's use is already invalid) and re-introduce
the constraint with unwrapped type instead. This would help to diagnose
the invalid chaining as well as any argument to parameter mismatches.
Resolves: rdar://126080504
https://github.com/swiftlang/swift/pull/80821 added `@isolated(any)`
which is an ABI breaking change but since we are not planning to keep
this API around, let's use `@_silgen_name` to bring the old name back
for now.
Resolves: rdar://152178221
An `IRGenFunction` may not have a debug scope--for example, a dispatch
thunk--but creating an `AritificalLocation` requires one. Only create
an `ArtificalLocation` here if the `IRGenFunction` has one.
Unfortunately, I don't have an isolated test case. Fixes a compiler
crash.
Make `getOriginalLocation` work with source ranges, and adjust the
cursor info logic to map the range into the original buffer. This
fixes the case where we were using bogus range lengths for macro
expansion decls.
rdar://151411756
The function convention for the first argument is not identified as indirect-out.
This lets alias analysis assume that the memory pointed to by argument 0 cannot be written by the called function.
The problem is that subscripting a LazyFilterCollection (with the base index, e.g. `Int`) does not work as expected, because it returns the nth element of the base collection!
The fix is to implement the subscript "manually".
Fixes a mis-compile.
rdar://152160748
... in distributed_actor_accessor_thunks_32bit.swift
Spotted a typo while skimming through the test so let's fix it. it
should be CHECK-SAME and not CHECK_SAME which would just do nothing
When issuing warnings about an import not needing to be public, we did
not account for the Distributed module MUST be imported when a
distributed actor is declared. This also actually means that a public
distributed actor effectively is a public use of the DistributedActor
protocol
resolves rdar://152129980
This PR updates the CI summary table in the README.md to summarize all
architectures for a host to be on the same line with the architecture
specified in the badge itself (rather than a generic "build" label),
thereby enabling the whole CI status to be visible on a single page.
It also adds accessibility labels to each of the CI badges for Global
Accessibility Awareness Month.
WIP to add more overloads to optimize SIMD codegen on concrete types.
Here we do:
- init(repeating:)
- init(lowHalf:highHalf:)
These are always inlined, even in debug, since LLVM knows how to lower
them to one or two instructions on the targets that we care about.
[send-non-sendable] Recurse to the full underlying value computation instead of just the object one when computing the underlying object of an address.
