PR 79186 (https://github.com/swiftlang/swift/pull/79186) moved one of
the mandatory passes from the C++ implementation to the Swift
implementation resulting in a compiler that is unable to build the
standard library. The pass used to ensure that inaccessible control-flow
positions after an infinite loop was marked with `unreachable` in SIL.
Since the pass is no longer running, any function that returns a value
that also has an infinite loop internally must place a fatalError after
the infinite loop or it will fail to compile as the compiler will
determine that the function does not return from all control flow paths
even though some of the paths are unreachable.
This is a wild guess at what might be causing our persistent, random
String failures on the main branch:
```
Swift(macosx-x86_64) :: Prototypes/CollectionTransformers.swift
Swift(macosx-x86_64) :: stdlib/NSSlowString.swift
Swift(macosx-x86_64) :: stdlib/NSStringAPI.swift
Swift(macosx-x86_64) :: stdlib/StringIndex.swift
Swift-validation(macosx-x86_64) :: stdlib/String.swift
Swift-validation(macosx-x86_64) :: stdlib/StringBreadcrumbs.swift
Swift-validation(macosx-x86_64) :: stdlib/StringUTF8.swift
```
FWIW, it appears this is *not* caused by https://github.com/apple/swift/pull/62717:
that change has also landed on release/5.8, and I haven’t seen these
issues on that branch.
Our atomic breadcrumbs initialization vs its non-atomic loading
gives me an uneasy feeling that this may in fact be a long standing
synchronization issue that is only now causing problems (for whatever
reason). I am unable to reproduce these issues locally, so this guess
may be (and probably is) wildly off the mark, but this PR is likely
to be a good idea anyway, if only to rule out this possibility.
rdar://104751936
[Bidirectional]Collection’s default index manipulation methods (as
well as _utf16Distance) do not expect to be given unreachable
indices, and they tend to fail when operating on them. Round indices
down to the nearest scalar boundary before calling these.
Speed up conversion between UTF-16 offset ranges
and string index ranges, by carefully switching
between absolute and relative index calculations,
depending on the distance we need to go.
It is a surprisingly tricky puzzle to do this
correctly while avoiding redundant calculations.
Offset ranges within substrings add the additional
complication of having to bias offset values with
the absolute offset of the substring’s start index.
We commonly start from the `startIndex`, in which case
`_nativeGetOffset` is essentially free. Consider this
case when calculating the threshold for using breadcrumbs.
Previously we insisted on using breadcrumbs even if we only needed to
travel a very short way. This could be as much as ten times slower
than the naive algorithm of simply visiting all the Unicode scalars
in between the start and the end.
(Using breadcrumbs generally means that we need to walk to both
endpoints from their nearest breadcrumb, which on average requires
walking half the distance between breadcrumbs — and this can mean
visiting vastly more Unicode scalars than the ones that are simply
lying in between the endpoints themselves.)
Swift 5.7 added stronger index validation for `String`, so some illegal cases that previously triggered inconsistently diagnosed out of bounds accesses now result in reliable runtime errors. Similarly, attempts at applying an index originally vended by a UTF-8 string on a UTF-16 string now result in a reliable runtime error.
As is usually the case, new traps to the stdlib exposes code that contains previously undiagnosed / unreliably diagnosed coding issues.
Allow invalid code in binaries built with earlier versions of the stdlib to continue running with the 5.7 library by disabling some of the new traps based on the version of Swift the binary was built with.
In the case of an index encoding mismatch, allow transcoding of string storage regardless of the direction of the mismatch. (Previously we only allowed transcoding a UTF-8 string to UTF-16.)
rdar://93379333
Fix a long-standing issue where the UTF16View overload of
`String.Index.init(_:within:)` used to return nil for valid indices
that happened to point to a trailing surrogate in a UTF-8-encoded
string.
rdar://91935537
In Swift 5.6 and below, (broken) code that acquired indices from a
UTF-16-encoded string bridged from Cocoa and kept using them after a
`makeContiguousUTF8` call (or other mutation) may have appeared to be
working correctly as long as the string was ASCII.
Since https://github.com/apple/swift/pull/41417, the
`String(_:within:)` initializers recognize miscoded indices and reject
them by returning nil. This is technically correct, but it
unfortunately may be a binary compatibility issue, as these used to
return non-nil in previous versions.
Mitigate this issue by accepting UTF-16 indices on a UTF-8 string,
transcoding their offset as needed. (Attempting to use an UTF-8 index
on a UTF-16 string is still rejected — we do not implicitly convert
strings in that direction.)
rdar://89369680
Also, in UTF-16 slices, forward collection methods to the base view
instead of `Slice`, to make behavior a bit easier to understand.
(There is no need to force readers to page in `Slice`
implementations _in addition to_ whatever the base view is doing.)
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.
* Don't allocate breadrumbs pointer if under threshold
* Increase breadrumbs threshold
* Linear 16-byte bucketing until 128 bytes, malloc_size after
* Allow cap less than _SmallString.capacity (bridging non-ASCII)
This change decreases the amount of heap usage for moderate-length
strings (< 64 UTF-8 code units in length) and increases the amount of
spare code unit capacity available (less growth needed).
Average improvements for moderate-length strings:
* 64-bit: on average, 8 bytes saved and 4 bytes of extra capacity
* 32-bit: on average, 4 bytes saved and 6 bytes of extra capacity
Additionally, on 32-bit, large-length strings also gain an average of
6 bytes of extra spare capacity.
Details:
On 64-bit, half of moderate-length allocations will save 16 bytes
while the other half get an extra 8 bytes of spare capacity.
On 32-bit, a quarter of moderate-length allocations will save 16
bytes, and the rest get an extra 4 bytes of spare
capacity. Additionally, 32-bit string's storage class now claims its
full allocation, which is its birthright. Prior to this change, we'd
have on average 1.5 bytes of spare capacity, and now we have 7.5 bytes
of spare capacity.
Breadcrumbs threshold is increased from the super-conservative 32 to
the pretty-conservative 64. Some speed improvements are incorporated
in this change, but more are in flight. Even without those eventual
improvements, this is a worthwhile change (ASCII is still fast-pathed
and irrelevant to breadcrumbing).
For a complex real-world workload, this amounts to around a 5%
improvement to transient heap usage due to all strings and a 4%
improvement to peak heap usage due to all strings. For moderate-length
strings specifically, this gives around 11% improvement to both.
Since scalar-alignment is set in inlinable code, switch the alignment
bit to one of the previously-reserved bits rather than a grapheme
cache bit. Setting a grapheme cache bit in inlinable would break
backward deployment, as older versions would interpret it as a cached
value.
Also adjust the name to "scalar-aligned", which is clearer, and
removed assertion (which should be a real precondition).
Fixes a general category (pun intended) of scalar-alignment bugs
surrounding exchanging non-scalar-aligned indices between views and
for slicing.
SE-0180 unifies the Index type of String and all its views and allows
non-scalar-aligned indices to be used across views. In order to
guarantee behavior, we often have to check and perform scalar
alignment. To speed up these checks, we allocate a bit denoting
known-to-be-aligned, so that the alignment check can skip the
load. The below shows what views need to check for alignment before
they can operate, and whether the indices they produce are aligned.
┌───────────────╥────────────────────┬──────────────────────────┐
│ View ║ Requires Alignment │ Produces Aligned Indices │
╞═══════════════╬════════════════════╪══════════════════════════╡
│ Native UTF8 ║ no │ no │
├───────────────╫────────────────────┼──────────────────────────┤
│ Native UTF16 ║ yes │ no │
╞═══════════════╬════════════════════╪══════════════════════════╡
│ Foreign UTF8 ║ yes │ no │
├───────────────╫────────────────────┼──────────────────────────┤
│ Foreign UTF16 ║ no │ no │
╞═══════════════╬════════════════════╪══════════════════════════╡
│ UnicodeScalar ║ yes │ yes │
├───────────────╫────────────────────┼──────────────────────────┤
│ Character ║ yes │ yes │
└───────────────╨────────────────────┴──────────────────────────┘
The "requires alignment" applies to any operation taking a
String.Index that's not defined entirely in terms of other operations
taking a String.Index. These include:
* index(after:)
* index(before:)
* subscript
* distance(from:to:) (since `to` is compared against directly)
* UTF16View._nativeGetOffset(for:)
String.Index has an encodedOffset-based initializer and computed
property that exists for serialization purposes. It was documented as
UTF-16 in the SE proposal introducing it, which was String's
underlying encoding at the time, but the dream of String even then was
to abstract away whatever encoding happend to be used.
Serialization needs an explicit encoding for serialized indices to
make sense: the offsets need to align with the view. With String
utilizing UTF-8 encoding for native contents in Swift 5, serialization
isn't necessarily the most efficient in UTF-16.
Furthermore, the majority of usage of encodedOffset in the wild is
buggy and operates under the assumption that a UTF-16 code unit was a
Swift Character, which isn't even valid if the String is known to be
all-ASCII (because CR-LF).
This change introduces a pair of semantics-preserving alternatives to
encodedOffset that explicitly call out the UTF-16 assumption. These
serve as a gentle off-ramp for current mis-uses of encodedOffset.
Defining a custom iterator for the UTF16View avoid some redundant
computation over the indexing model. This speeds up iteration by
around 40% on non-ASCII strings.
Add an isASCII fast-path for many UTF16View operations. These are
heavily utilized in random-access scenarios, allowing us to both be
more efficient and skip generating breadcrumbs for ASCII strings.
• Convert _AbstractStringStorage to a protocol, and the free functions used to deduplicate implementations to extensions on that protocol.
• Move 'start' into the abstract type and use that to simplify some code
• Move the ASCII fast path for length into UTF16View.
• Add a weirder but faster way to check which (if any) of our NSString subclasses a given object is, and adopt it
Add a use an unchecked subscript on UnsafeBufferPointer, which skips
debugPrecondition checks (in case we're not inlined) as well as a
force-unwrap check.
Rather than rely on the UTF16View, scan between breadcrumbs by hand
for a decent 20% speedup. This code will also make it more obvious how
to slot in a vectorized solution later.
