Optimize NSString.NSRangeToByteRange(start:length:)
#122
Conversation
|
This is that I mentioned in #120. |
|
I think the performance issue that related to realm/SwiftLint#247 is almost resolved, and another performance issues that made SwiftLint slower than 0.4.0 are caused by rules. |
|
Today, I worried about that this PR would make leaks String instance. |
| cache[location] = entry | ||
|
|
||
| // keep largest 5 entry | ||
| if cache.keys.count > 5 { |
There was a problem hiding this comment.
is it really necessary to evict cache entries? An Int/Int keypair is minuscule in memory compared to the rest of the strings we're working with.
There was a problem hiding this comment.
Now I tested again.
I confirmed that evicting cache does not affect performance. 👍
I'll remove that.
|
Thank you for reviews. |
I also found that using Should we avoid using this approach? |
norio-nomura
changed the title
NSString.NSRangeToByteRange(start:length:)NSString.NSRangeToByteRange(start:length:)
|
It seems I need to using two workarounds for resolving memory leaks.
|
norio-nomura
changed the title
NSString.NSRangeToByteRange(start:length:)NSString.NSRangeToByteRange(start:length:)
|
I added workarounds for memory leaks. |
Introduce `ByteOffsetCache` for calculating byte offset. By applying this, linting KeychainAccess by SwiftLint-0.5.1 is reduced from: ``` swiftlint 21.87s user 0.24s system 97% cpu 22.611 total ``` to: ``` swiftlint 16.62s user 0.24s system 97% cpu 17.362 total ``` On Instruments, `NSRangeToByteRange()` is reduced from 6579ms to 315ms on that test.
Rename `entry` to `byteOffsetIndexPair` Remove evicting `ByteOffsetIndexPair.cache`
There are two reasons for: 1. Avoid using associatedObject on NSTaggedPointerString (< 7 bytes) because that does not free associatedObject. 2. Using cache is overkill for short string. There is no verified reason for the threshold "50", but I
If string is `Swift.String`, holding that into `ByteOffsetCache` does not cause Circular reference, because Casting `String` to `NSString` makes new `NSString` instance. If string is native `NSString` instance, Circular reference happens on following: ``` self.utf8View = (string as String).utf8 ``` Because the reference to `NSString` is holded by every casted `String`, their Views and Indices.
norio-nomura
force-pushed
the
byte-offset-cache
branch
from
9019d20 to
6656224
Compare
|
rebased to master |
Optimize `NSString.NSRangeToByteRange(start:length:)`
|
Awesome! Thanks for yet another solid contribution. |
|
😄 |
Introduce
ByteOffsetCachefor calculating byte offset.By applying this, linting KeychainAccess by SwiftLint-0.5.1 is reduced from:
to:
On Instruments,
NSRangeToByteRange()is reduced from 6579ms:to 315ms:
on that test.