Where we have rethrowing versions of functions that have typed-throws
counterparts that are only retained for ABI compatibility, wrap them
in `#if !$Embedded` so they aren't compiled into the Embedded version
of the standard library. This eliminates warnings about this code,
which cannot actually be used with arbitrary errors anyway.
https://github.com/swiftlang/swift/pull/72612 can be reverted because it is no
longer necessary for the interface of the stdlib to be compatible with
compilers without `$TypedThrows` support.
Replace the hackish use of `@_disfavoredOverload` with the more principled
use of `@_silgen_name` for the entrypoint we are maintaining, then rename
these functions in source to `__rethrows_map` to indicate what they're for.
While here, make them `throws` instead of `rethrows`. The ABI is the
same, and `throws` allows us do avoid to do/catch tricks with rethrows
functions.
Adopt typed throws for the `map` operation to propagate thrown error
types through the `map` API. This is done in a manner that is backward
compatible for almost all cases:
* The new typed-throws entrypoint is `@_alwaysEmitIntoClient` so it
can back-deploy all the way back.
* The old `rethrows` entrypoint is left in place, using
`@usableFromInline` with `internal` so it is part of the ABI but not
the public interface, and `@_disfavoredOverload` so the type checker
avoids it while building the standard library itself. The old
entrypoint is implemented in terms of the new one.
Note that the implementation details for the existential collection
"box" classes rely on method overriding of `_map` operations, and the
types are frozen, so we don't get to change their signatures. However,
these are only implementations, not API: the actual API `map`
functions can be upgraded to typed throws.
Note that this code makes use of a known hole in `rethrows` checking
to allow calling between `rethrows` and typed throws. We'll need to do
something about this for source-compatibility reasons, but I'll follow
up with that separately.
Replace the hackish use of `@_disfavoredOverload` with the more principled
use of `@_silgen_name` for the entrypoint we are maintaining, then rename
these functions in source to `__rethrows_map` to indicate what they're for.
While here, make them `throws` instead of `rethrows`. The ABI is the
same, and `throws` allows us do avoid to do/catch tricks with rethrows
functions.
Adopt typed throws for the `map` operation to propagate thrown error
types through the `map` API. This is done in a manner that is backward
compatible for almost all cases:
* The new typed-throws entrypoint is `@_alwaysEmitIntoClient` so it
can back-deploy all the way back.
* The old `rethrows` entrypoint is left in place, using
`@usableFromInline` with `internal` so it is part of the ABI but not
the public interface, and `@_disfavoredOverload` so the type checker
avoids it while building the standard library itself. The old
entrypoint is implemented in terms of the new one.
Note that the implementation details for the existential collection
"box" classes rely on method overriding of `_map` operations, and the
types are frozen, so we don't get to change their signatures. However,
these are only implementations, not API: the actual API `map`
functions can be upgraded to typed throws.
Note that this code makes use of a known hole in `rethrows` checking
to allow calling between `rethrows` and typed throws. We'll need to do
something about this for source-compatibility reasons, but I'll follow
up with that separately.
This isn't a "complete" port of the standard library for embedded Swift, but
something that should serve as a starting point for further iterations on the
stdlib.
- General CMake logic for building a library as ".swiftmodule only" (ONLY_SWIFTMODULE).
- CMake logic in stdlib/public/core/CMakeLists.txt to start building the embedded stdlib for a handful of hardcoded target triples.
- Lots of annotations throughout the standard library to make types, functions, protocols unavailable in embedded Swift (@_unavailableInEmbedded).
- Mainly this is about stdlib functionality that relies on existentials, type erasure, metatypes, reflection, string interpolations.
- We rely on function body removal of unavailable functions to eliminate the actual problematic SIL code (existentials).
- Many .swift files are not included in the compilation of embedded stdlib at all, to simplify the scope of the annotations.
- EmbeddedStubs.swift is used to stub out (as unavailable and fatalError'd) the missing functionality.
Currently, the default implementations of the various `_failEarlyRangeCheck` forms contain several _precondition invocations, like this:
```
@inlinable
public func _failEarlyRangeCheck(_ index: Index, bounds: Range<Index>) {
_precondition(
bounds.lowerBound <= index,
"Out of bounds: index < startIndex")
_precondition(
index < bounds.upperBound,
"Out of bounds: index >= endIndex")
}
```
Each such precondition call generates a separate trap instruction, which seems like a waste — theoretically it would be helpful to know which condition was violated, but in practice, that information tends not to be surfaced anyway. Combining these will lead to a minuscule code size improvement.
(The separate messages do surface these in debug builds, but only if these generic defaults actually execute, which isn’t often. These are not public entry points, so concrete collection types tend ignore these and roll their own custom index validation instead. From what I’ve seen, custom code tends to combine the upper/lower checks into a single precondition check. These underscored requirements tend to be only called by the standard `Slice`; and it seems reasonable for that to follow suit.)
More interestingly, the range-in-range version of `_failEarlyRangeCheck` performs two times as many checks than are necessary:
```
@inlinable
public func _failEarlyRangeCheck(_ range: Range<Index>, bounds: Range<Index>) {
_precondition(
bounds.lowerBound <= range.lowerBound,
"Out of bounds: range begins before startIndex")
_precondition(
range.lowerBound <= bounds.upperBound,
"Out of bounds: range ends after endIndex")
_precondition(
bounds.lowerBound <= range.upperBound,
"Out of bounds: range ends before bounds.lowerBound")
_precondition(
range.upperBound <= bounds.upperBound,
"Out of bounds: range begins after bounds.upperBound")
}
```
It is safe to assume that `range.lowerBound <= range.upperBound`, so it’s enough to test that `bounds.lowerBound <= range.lowerBound && range.upperBound <= bounds.upperBound` — we don’t need to check each combination separately.
The requirement machine does not allow 'where' clauses to reference
typealiases defined in protocol extensions.
These should probably be removed entirely since they're not useful
anymore, but I'd rather that decision was made by the stdlib folks.
- adds a default implementation of Collection’s subscript(bounds: Range<_>)
with the most general signature possible
- it is marked unavailable in order to prevent the
infinite recursion bug reported in SR-14848
- Collections whose SubSequence is Slice<Self> still get the proper default, as intended.
Introduce checking of ConcurrentValue conformances:
- For structs, check that each stored property conforms to ConcurrentValue
- For enums, check that each associated value conforms to ConcurrentValue
- For classes, check that each stored property is immutable and conforms
to ConcurrentValue
Because all of the stored properties / associated values need to be
visible for this check to work, limit ConcurrentValue conformances to
be in the same source file as the type definition.
This checking can be disabled by conforming to a new marker protocol,
UnsafeConcurrentValue, that refines ConcurrentValue.
UnsafeConcurrentValue otherwise his no specific meaning. This allows
both "I know what I'm doing" for types that manage concurrent access
themselves as well as enabling retroactive conformance, both of which
are fundamentally unsafe but also quite necessary.
The bulk of this change ended up being to the standard library, because
all conformances of standard library types to the ConcurrentValue
protocol needed to be sunk down into the standard library so they
would benefit from the checking above. There were numerous little
mistakes in the initial pass through the stsandard library types that
have now been corrected.
Without this change, SILGen will crash when compiling a use of the
derived protocol's requirement: it will instead attempt to use
the base protocol's requirement, but the code will have been
type-checked incorrectly for that.
This has a potential for source-compatibility impact if anyone's
using explicit override checking for their protocol requirements:
reasonable idioms like overriding a mutating requirement with a
non-mutating one will no longer count as an override. However,
this is arguably a bug-fix, because the current designed intent
of protocol override checking is to not allow any differences in
type, even "covariant" changes like making a mutating requirement
non-mutating. Moreover, we believe explicit override checking in
protocols is quite uncommon, so the overall compatibility impact
will be low.
This also has a potential for ABI impact whenever something that
was once an override becomes a non-override and thus requires a
new entry. It might require a contrived test case to demonstrate
that while using the derived entry, but it's quite possible to
imagine a situation where the derived entry is not used directly
but nonetheless has ABI impact.
Furthermore, as part of developing this patch (earlier versions of
which used stricter rules in places), I discovered a number of
places where the standard library was unintentionally introducing
a new requirement in a derived protocol when it intended only to
guide associated type deduction. Fixing that (as I have in this
patch) *definitely* has ABI impact.
Most of the stdlib's properties don't need @_borrowed because they're
@inlinable, but I did find one place in an overlay where it's probably
sensible to make the operation use generalized accessors even if
they're resilient.
* [stdlib] Minor documentation revisions
* [docs] Convert 'nonoptional' to 'non-optional'
We're switching to 'non-optional' across the board, as the unhyphenated
form is too easy to read as 'no-noptional'.
Now that we have removed overriding protocol requirements from witness
tables, they no longer have any effect on the ABI. Replace the FIXME
(ABI) comments with normal FIXMEs: there is no more ABI work to do
here.
Add the `-warn-implicit-overrides` flag when building the standard library
and overlays, so that each protocol member that overrides a member of an
inherited protocol will produce a warning unless annotated with either
‘override’ or ‘@_nonoverride’.
An annotation of `override` will mean that the overriding requirement will be treated identically to the overridden declaration. If for some reason a concrete type’s conformance to the inheriting protocol provides a different witness for the overriding requirement than the conformance to the inherited protocol’s witness for the overridden requirement, the witness for the inheriting (more-specialized) protocol will be ignored. A protocol requirement marked ‘override’ only makes sense when the declaration is needed to help associated type inference, which is why the ‘override’ annotations correlate so closely with ABI FIXMEs.
An annotation of `@_nonoverride` means that the two protocol requirements will be treated independently, and may be bound to different witnesses. Use `@_nonoverride` when we might need different witnesses, e.g., because the semantics of the potentially-overriding declaration differ from that of the potentially-overridden declaration. `BidirectionalCollection.index(_:offsetBy:)` is the most obvious example, because the `BidirectionalCollection` ’s version of `index(_:offsetBy:)` allows negative indices. `RandomAccessCollection` ’s version is also marked `@_nonoverride` because it is required to be asymptotically faster than the `Collection` or `BidirectionalCollection` versions.
- Fix error in `last(where:)` example
- Improve MemoryLayout, UnsafePointer, and integer operator discussions
- Clean up ranges and random APIs
- Revisions to overflow operators and the SignedNumeric requirements
- Standardize on 'nonoptional' in remaining uses
* [stdlib] Update complexity docs for seq/collection algorithms
This corrects and standardizes the complexity documentation for Sequence
and Collection methods. The use of constants is more consistent, with `n`
equal to the length of the target collection, `m` equal to the length of
a collection passed in as a parameter, and `k` equal to any other passed
or calculated constant.
* Apply notes from @brentdax about complexity nomenclature
* Change `n` to `distance` in `index(_:offsetBy:)`
* Use equivalency language more places; sync across array types
* Use k instead of n for parameter names
* Slight changes to index(_:offsetBy:) discussion.
* Update tests with new parameter names
* Remove case destructuring to _
* Remove some Iterator.Element
* Which idiot wrote this? Oh.
* Switch NibbleSort to just use default impls... shouldn't change perf
