Generalize accessor storage to preserve the original accessor list. #17238
Conversation
|
@swift-ci Please test. |
|
Build failed |
|
Build failed |
|
|
||
| /// Either 0, meaning there is no registered accessor of the given kind, | ||
| /// or the index+1 of the accessor in the accessors array. | ||
| AccessorIndex AccessorIndices[NumAccessorKinds]; |
There was a problem hiding this comment.
Can you use llvm::TrailingObjects instead of a supposed fixed-sized array?
There was a problem hiding this comment.
That is a fixed-size array; it's indexed by AccessorKinds. There is a trailing-objects use in that type, though.
There was a problem hiding this comment.
Oops, my brain misread NumAccessorKinds as MaxNumAccessors. Sorry for the noise!
There was a problem hiding this comment.
I've updated to use TrailingObjects for the part that's actually a trailing array.
rjmccall
force-pushed
the
full-accessor-lists
branch
2 times, most recently
from
3f1e547 to
841e69b
Compare
|
@swift-ci Please test. |
|
Build failed |
|
Build failed |
rjmccall
force-pushed
the
full-accessor-lists
branch
from
841e69b to
3c44d64
Compare
|
@swift-ci Please test. |
|
Build failed |
|
Build failed |
rjmccall
force-pushed
the
full-accessor-lists
branch
from
3c44d64 to
33014de
Compare
|
@swift-ci Please test. |
|
Build failed |
|
Build failed |
rjmccall
force-pushed
the
full-accessor-lists
branch
from
33014de to
fc64f54
Compare
|
@swift-ci Please test. |
|
Build failed |
|
Build failed |
rjmccall
force-pushed
the
full-accessor-lists
branch
from
fc64f54 to
7f259cc
Compare
|
@swift-ci Please test. |
|
Build failed |
|
Build failed |
| @@ -4367,13 +4367,14 @@ public class FooOverlayClassDerived : Foo.FooOverlayClassBase { | |||
| { | |||
| key.kind: source.lang.swift.decl.var.global, | |||
| key.accessibility: source.lang.swift.accessibility.public, | |||
| key.setter_accessibility: source.lang.swift.accessibility.public, | |||
There was a problem hiding this comment.
@benlangmuir @nkcsgexi I'm pretty sure that these changes are good — these are the global aliases we emit for C enumerators, so (1) they're not settable and (2) they do have the body {get} IIUC — but they're not intentional, so I'd feel better with a thumbs up. :)
There was a problem hiding this comment.
I understand why these changes occurred now. The generated interface file contains snippets like var x: Int {get} in all sorts of contexts. If you ask Swift to parse this, it will usually reject it because the accessor does not have a body. Previously, the parser wouldn't even make an AccessorDecl for the accessor, with two effects: first, there's no AST node for the syntax library to associate with the syntax, leading it to state that the variable has no body; and second, the parser sees a variable with no accessors and mistakenly decides it must be stored, and all stored variables are settable. My patch preserves the incomplete accessor declaration and properly considers the variable to be computed, leading to these changes. This is a clear improvement in our output.
rjmccall
force-pushed
the
full-accessor-lists
branch
from
7f259cc to
de4d56b
Compare
|
Test failure is a consequence of other changes. Rebasing and resubmitting. |
|
@swift-ci Please test. |
|
Build failed |
|
Build failed |
rjmccall
force-pushed
the
full-accessor-lists
branch
2 times, most recently
from
36ad0c3 to
5506a1a
Compare
|
@swift-ci Please test. |
|
Build failed |
|
Build failed |
Only not NFC because it's detectable by source tools.
rjmccall
force-pushed
the
full-accessor-lists
branch
from
5506a1a to
69f4dd1
Compare
|
@swift-ci Please test. |
|
Build failed |
|
Build failed |
Only not NFC because it's detectable by source tools.