Alter the value metadata layout to use an absolute pointer for the
nominal type descriptor rather than a relative offset relative to the
complete type metadata. Although this is slightly less efficient in
terms of load times, this is more portable across different
environments. For example, PE/COFF does not provide a cross-section
relative offset relocation. Other platform ports are unable to provide
a 64-bit relative offset encoding.
Given that the value witness table reference in the value metadata is
currently an absolute pointer, this page is most likely going to be
dirtied by the loader.
We can finally get rid of -sil-serialize-all when building the standard library! This option will be completely eliminated in the future commits.
Instead of serializing just everything as we did before, we now serialize only functions annotated with @_inlineable. This way we can selectively control what needs to be available to the clients. This is an important step towards building a resilient standard library.
While this is a huge change for the serialization of the stdlib, it should be virtually invisible to the clients. For example, there are no noticeable performance regressions on any of the benchmarks.
makeUnique hoisting can create a situation where we hit an assert in
_reserveCapacityAssumingUniqueBuffer that the buffer is not unique if we use
_makeMutableAndUniqueOrPinned to replace
_makeUniqueAndReserveCapacityIfNotUnique.
It is actually fine do to the replacement because we will make the buffer unique
_reserveCapacityAssumingUniqueBuffer if the capacity is zero so adjust the
assert accordingly.
do {
if (someCond) {
// array.append(...)
_makeUniqueAndReserveCapacityIfNotUnique(&array)
_reserveCapacityAssumingUniqueBuffer(&array)
_appendElementAssumeUniqueAndCapacity(&array, …)
} else {
// array[i] = …
_makeMutableAndUniqueOrPinned(&array)
addr = _getElementAddress(&array)
store 1, addr
}
} while();
to:
_makeMutableAndUniqueOrPinned(&array) // does not replace empty arrays.
do {
if (someCond) {
// array.append(...)
_reserveCapacityAssumingUniqueBuffer(&array) // hit the assert.
_appendElementAssumeUniqueAndCapacity(&array, …)
} else {
// array[i] = …
addr = _getElementAddress(&array)
store 1, addr
}
} while();
Tested by the performance test if we build the stdlib with assertions.
rdar://34149935
The various _*Indexable protocols only exist to work around the lack of
recursive protocol constraints. Eliminate all of the *_Indexable protocols,
collapsing their requirements into the corresponding Collection protocol
(e.g., _MutableIndexable —> Collection).
This introduces a number of extraneous requirements into the various
Collection protocols to work around bugs in associated type
inference. Specifically, to work around the lack of "global" inference
of associated type witnesses. These hacks were implicitly present in
the *Indexable protocols; I've made marked them as ABI FIXMEs here so
we can remove them when associated type inference improves.
Fixes rdar://problem/21935030 and a number of ABI FIXMEs in the library.
Rather than using the default slice type when slicing the collection produced
by a lazy map or filter, slice the base collection and form a new
lazy map/filter collection from it. This allows any optimizations provided by
the collection SubSequence type to kick in, as well as ensuring that slicing
a lazy collection provides the same type as producing a lazy collection of a
slice.
This is technically source-breaking, because someone could have spelled out
the types of slicing a lazy filter or map… but it seems unlikely to matter
in practice and the benefits could be significant.
Fixes ABI FIXME’s #28 and #46.
Make the Indices types conform to the appropriate Collection protocol:
* Collection.Indices: Collection
* BidirectionalCollection.Indices: BidirectionalCollection
* RandomAccessCollection.Indices: RandomAccessCollection
Introduce (recursive) constraints that make the *Collection constraint
of SubSequence match that of its enclosing *Collection, e.g.,
MutableCollection.SubSequence conforms to MutableCollection.
Fixes rdar://problem/20715031 and more of SR-3453.
Addressed ABI FIXME’s #4, #5, #104 and #105, making Sequence’s
SubSequence conform to Sequence, with the same element type, and for
which the SubSequence of a SubSequence is the same SubSequence.
Fixes SR-318 / rdar://problem/31418206.
Int is still an insufficient stride type for binary integers,
but this improves the situation for values at the extremes of the
concrete integer types.
This changes the conditional for the definition of the _VaListBuilder.
This makes it easier to read as well as to extend to other platforms
which use a non-pointer va_list type. NFC.
When a substring gets bridged to NSString, it loses the initial offset,
therefore APIs that accept or return StringIndex ranges should handle
this case explicitly by adding/subtracting the substring start offset.
Fixes <rdar://problem/33873277>
We no longer need this for anything, so remove it from metadata
altogether. This simplifies logic for emitting type metadata and
makes type metadata smaller.
We still pass the parent metadata pointer to type constructors;
removing that is a separate change.
Wean the routine that builds a mangling tree from metadata from
looking at the parent metadata pointer, instead using the nested
depth to partition the arguments into a list of arguments for
each nesting depth.
