Re-factor Codable.decode functions to reduce the standard library code size #10126
Conversation
|
@swift-ci please smoke test |
|
@itaiferber Itai, do you mind having a look? I just re-factored the slow path. No semantic changes are intended. |
There was a problem hiding this comment.
Thanks for putting this together! 😄 I think we can make this even simpler, though.
However: given that this error production is never inlined, what's the tradeoff here we'll be making in terms of size for speed? Is there any sort of noticeable slowdown because we're not inlining?
I also have to admit, I'm not familiar with the @_semantics being applied here. What do they optimize for?
stdlib/public/core/Codable.swift
Outdated
| @_semantics("optimize.sil.specialize.generic.never") | ||
| @_semantics("optimize.sil.specialize.generic.partial.never") | ||
| @inline(never) | ||
| internal func decodeError<T: Decodable>(_ type: T.Type, forKey key: Key) throws -> T { |
There was a problem hiding this comment.
What's the point of this throwing the error instead of just returning a new error?
Can't this just be internal func _errorInDecoding(_ type: T.Type, forKey key: Key) -> DecodingError { ... }?
There was a problem hiding this comment.
Good idea! I just tried to be as similar as possible to the original code.
stdlib/public/core/Codable.swift
Outdated
| @_semantics("optimize.sil.specialize.generic.never") | ||
| @_semantics("optimize.sil.specialize.generic.partial.never") | ||
| @inline(never) | ||
| internal func decodeError<T: Decodable>(_ type: T.Type) throws -> T { |
There was a problem hiding this comment.
Same comment here. Why pretend to return a T when we can just return the error?
|
@itaiferber We are not inlining the slow path now. So, we pay a price of a generic function call. But since it is a slow path, it doesn't matter. The @_semantics are used to prevent the generic specializer from (partially) specializing the generic code for concrete types, because this would result in the code bloat in the object file. |
…e size Move the error reporting slow-path into a separate function, which is not inlined or specialized. This reduced the stdlib code size by almost 1%.
swiftix
force-pushed
the
inliner-improvements2
branch
from
8605699 to
890c2d5
Compare
|
@itaiferber I address your review comments. Looks better now? |
|
@swift-ci please smoke test |
1 similar comment
|
@swift-ci please smoke test |
|
@itaiferber Let me know if it is OK to merge or if you have any further improvements suggestions. |
|
LGTM but I'd like to be sure with a full test, if you don't mind. |
|
@swift-ci Please test and merge |
Move the error reporting slow-path into a separate function, which is not inlined or specialized.
This reduced the stdlib code size by almost 1%.