Seems that during refactorings of child cancellations we somehow missed
also cancelling the group itself. It seems we did not have good test
coverage of the addTaskUnlessCancelled somehow and thus this slipped
through.
This adds a regression test for addTaskUnlessCancelled and fixes how we
handle the cancellation effect in TaskStatus.
resolves#80789
resolves rdar://149177600
Because `TaskAllocator` is not a round multiple of the machine word
size on 64-bit platforms, I think we end up with padding before the
`TaskLocal::Storage` following it, which makes the `PrivateStorage`
structure larger than the calculation in `ABI/Task.h`.
rdar://149067144
We decided that using a magic typealias to set the executor factory was better
than using a compiler option. Remove the `-executor-factory` option, and replace
by looking up the `DefaultExecutorFactory` type, first in the main module, and
then if that fails in Concurrency.
rdar://149058236
Revert "Merge pull request #79707 from DougGregor/transparent-integer-conversions"
This reverts commit 9c2c4ea07f, reversing
changes made to 829e03c104.
We need to use lookThroughOptionalConditionalWitnessTable when relative witness tables are being used.
rdar://148687148
(cherry picked from commit f01d46ce38)
Iterating child tasks depends on knowing the size of AsyncTask, and changing the size of the task broke it. Instead of relying on mirroring the full structure in our out-of-process definitions, add a debug variable to libswift_Concurrency that contains the size of AsyncTask.
While we're there, add some more validation to child task enumeration. Check each child task's metadata pointer to make sure that it actually points to the AsyncTask metadata, and have the inner loop also increment and check ChildTaskLoopCount to stop runaway iteration in that loop.
rdar://148836760
(cherry picked from commit e3057031da)
The metadata creation system detects cycles where metadata depends on other metadata which depends on the first one again and raises a fatal error if the cycle can't be fulfilled.
Some cycles can be fulfilled. The cycle may involve a requirement for a metadata state less than full transitive completeness which can be reached without resolving the entire cycle. We only want to raise a fatal error when we detect a cycle that can't be fulfilled.
Normally this happens because the cycle checking in `blockOnMetadataDependency` only sees a cycle when it can't be fulfilled. Metadata initialization is advanced as far as it can be at each stage, so a cycle that can be fulfilled will see a fulfilling state and won't generate the dependency in the first place, since we only generate dependencies that haven't yet been met.
However, when two threads race to create types in a cycle, we can end up with such a dependency, because the dependency may be generated before another thread fulfilled yet. The cycle checker doesn't account for this and incorrectly raises a fatal error in that case.
Fix this by checking the cyclic dependency against the metadata's current state. If we have a dependency that's already been fulfilled, then there isn't really a dependency cycle. In that case, don't raise a fatal error.
rdar://135036243
(cherry picked from commit 9ad534bc4a)
To facilitate back deployment, make use of the fact that the async bit
has up to now never been set for read and modify accessors and claim
that set bit to indicate that it is a callee-allocated coroutine. This
has the virtue of being completely back deployable because like async
function pointers coro function pointers must be auth'd and signed as
data.
We need this so that older compilers can handle the .swiftinterface
files we generate. It's unnecessary for newer compilers and can be
removed later.
Fixes rdar://148529962.
(cherry picked from commit 0100104ff7)
As of the custom main/global executor changes, there is a race in
`_runAsyncMain()` to construct the main executor; if this goes the wrong way,
the IRGen async tests, which use this function, can fail.
Fix by explicitly constructing a task and enqueing it on the main executor,
instead of detaching a task and trying to hop to it.
rdar://148506256
This avoids the potential to race with the triggering coming from
task_cancel, because we first set the cancelled flag, and only THEN
take the lock and iterate over the inserted records. Because of this we
could: T1 flip the cancelled bit; T2 observes that, and triggers
"immediately" during installing the handler record. T1 then proceeds to
lock records and trigger it again, causing a double trigger of the
cancellation handler.
resolves https://github.com/swiftlang/swift/issues/80161
resolves rdar://147493150
Remove Malloc Type Descriptor cache and trivialize
`computeMallocTypeSummary()` to only provide
language. The remaining info in
`malloc_type_summary_t` are currently not used by
the allocator.
The principled, long-term solution is to have the
Swift compiler compute type descriptors for Swift
types.
rdar://137993434
This caused a hard-to-reproduce bug where a machine that had slept for
a long time would take a very long time to run some of the tests because
we'd be scheduling Dispatch executions with the wrong clock.
rdar://148337712
