[stdlib] Make String.makeContiguousUTF8() strictly true #82851
Conversation
glessard
force-pushed
the
rdar154331399-makeReallyContiguous
branch
from
0999641 to
268eb09
Compare
|
@swift-ci please test linux platform |
glessard
force-pushed
the
rdar154331399-makeReallyContiguous
branch
from
268eb09 to
d0154c2
Compare
glessard
changed the title
|
@swift-ci please test linux platform |
glessard
force-pushed
the
rdar154331399-makeReallyContiguous
branch
from
d0154c2 to
6c9b0b0
Compare
|
@swift-ci please test linux platform |
glessard
changed the title
|
@swift-ci please smoke test linux platform |
1 similar comment
|
@swift-ci please smoke test linux platform |
glessard
force-pushed
the
rdar154331399-makeReallyContiguous
branch
from
bf4ada7 to
3dfdd87
Compare
|
@swift-ci please test |
|
@swift-ci please test |
| @@ -312,9 +312,19 @@ extension String { | |||
| @inline(never) // slow-path | |||
| internal static func _copying(_ str: Substring) -> String { | |||
| if _fastPath(str._wholeGuts.isFastUTF8) { | |||
| return unsafe str._wholeGuts.withFastUTF8(range: str._offsetRange) { | |||
| var new = unsafe str._wholeGuts.withFastUTF8(range: str._offsetRange) { | |||
There was a problem hiding this comment.
In non-watchOS or 64-bit environments, will the compiler not complain about this var being unmutated?
There was a problem hiding this comment.
I didn't notice it complaining when building on macOS. Apparently it knows to consider statements that are conditionally skipped: https://swift.godbolt.org/z/aT1naWfce
| /// Contiguous strings always operate in O(1) time for withUTF8 and always | ||
| /// give a result for String.UTF8View.withContiguousStorageIfAvailable. | ||
| /// Contiguous strings always operate in O(1) time for withUTF8, always give | ||
| /// a result for String.UTF8View.withContiguousStorageIfAvailable, and always |
There was a problem hiding this comment.
Was this comment actually inaccurate before? Did String.UTF8View.withContiguousStorageIfAvailable actually NOT provide a result for these exceptional small strings?
There was a problem hiding this comment.
It was not lying: UTF8View.withContiguousStorageIfAvailable will rearrange the bytes of a small string in order to present contiguous storage to the closure.
There was a problem hiding this comment.
OK. With this returning false now for these strings, some might find it surprising that the former conditions are still true. However, seems like we're within our rights to do so, as this is implication ("isContiguousUTF8 -> the above guarantees hold") and not equivalence ("isContiguousUTF8 <-> the above guarantees hold").
| /// Contiguous strings always operate in O(1) time for withUTF8 and always | ||
| /// give a result for String.UTF8View.withContiguousStorageIfAvailable. | ||
| /// Contiguous strings always operate in O(1) time for withUTF8, always give | ||
| /// a result for String.UTF8View.withContiguousStorageIfAvailable, and always |
There was a problem hiding this comment.
OK. With this returning false now for these strings, some might find it surprising that the former conditions are still true. However, seems like we're within our rights to do so, as this is implication ("isContiguousUTF8 -> the above guarantees hold") and not equivalence ("isContiguousUTF8 <-> the above guarantees hold").
|
@swift-ci please test linux platform |
Until now,
StringProtocol.makeContiguousUTF8()only considered whether or not aStringorSubstringheap allocation contained contiguous UTF8, disregarding whether a stack-allocated instance was discontiguous. This is no longer tenable when these types vendUTF8SpanandSpanproperties, so we now also consider whether the small string form contains contiguous code units. If the small form contains discontiguously-stored code units, then it is replaced by a new heap-allocated String in contiguous storage. This new code path is specific to ABI-stable, 32-bit platforms.Addresses rdar://154331399