KspTypeNamesGoldenTest not enabled for KSP2 [394692559]

Assigned

Bug

Status Update

No update yet.

Description

bc...@google.com

created issue #1

Feb 6, 2025 01:53AM

This test was never enabled with KSP2 due to known issues (https://github.com/google/ksp/issues/1930) but that issue is now closed.

We should enable this test for KSP2 (initially exempting any failures) to prevent regressions and track progress towards.

Comments

bc...@google.com <bc...@google.com> #2Feb 7, 2025 06:20AM

there are some image for help

bc...@google.com <bc...@google.com> #3Mar 1, 2025 01:24AM

hello?

bc...@google.com <bc...@google.com> #4Mar 1, 2025 01:26AM

Release 2.1.10-1.0.31 contained fixes for

https://github.com/google/ksp/issues/2345

This fixed all but 2 remaining signatures in KSTypeNamesGoldenTest for KSP2 (see aosp/3517754).

bc...@google.com <bc...@google.com> #5Mar 1, 2025 01:29AM

I spent some time looking into the 2 remaining KSP2 signatures that don't match KAPT.

This is definitely due to a discrepancy between KSP1 and KSP2, but IMO KSP2 has the correct behavior.

Consider the case below:

fun method(): MyAlias
alias MyAlias = @JvmSuppressWildcards List<Foo>
interface Foo

In KSP1, the return type, MyAlias, has the @JvmSuppressWildcards annotation. In KSP2, the return type does not have the @JvmSuppressWildcards annotation. Instead, it's on the aliased type, List<Foo>.

Since KSP2's behavior aligns better with the actual annotations in source code I think KSP2's behavior is more correct. Thus, rather than filing a KSP bug for this, I'm going to update XProcessing's code to work with the new behavior in KSP2.

bc...@google.com <bc...@google.com> #6Mar 3, 2025 10:21PM

This should be fixed now that the following is submitted:

Fix KSTypeVarianceResolver for typealias with @JvmSuppressWildcards..

ap...@google.com <ap...@google.com> #7Mar 3, 2025 10:25PM

Project: platform/frameworks/support
Branch: androidx-main
Author: bcorso <bcorso@google.com>
Link: https://android-review.googlesource.com/3518076

Fix KSTypeVarianceResolver for typealias with @JvmSuppressWildcards.

Expand for full commit details

Fix KSTypeVarianceResolver for typealias with @JvmSuppressWildcards. 
 
In KSP1, @JvmSuppressWildcards annotation was copied from the type being 
aliased to the typealias itself. However, in KSP2, that does not happen. 
The KSP2 behavior actually seems more correct, so rather than file a 
KSP bug, I think we should fix this in XProcessing. In particular, 
XProcessing shouldn't rely on the incorrect KSP1 behavior to work, so I've 
updated the code to keep track of the annotations manually to make things 
work with KSP2. 
 
I've also removed a workaround for https://github.com/google/ksp/issues/1376 
which is has been fixed in KSP for a while now. 
 
Bug: 394692559 
Test: KSTypeNamesGoldenTest.kt 
Change-Id: Id1abbbac330c334a2ee6493d6e78e609e0c0c404

Files:

M room/room-compiler-processing/src/main/java/androidx/room/compiler/processing/ksp/KSTypeExt.kt
M room/room-compiler-processing/src/main/java/androidx/room/compiler/processing/ksp/KSTypeVarianceResolver.kt
M room/room-compiler-processing/src/test/java/androidx/room/compiler/processing/ksp/KspTypeNamesGoldenTest.kt

Hash: 6abd916f4b6cd63490fad17c543d8d188e03ed54
Date: Fri Feb 28 16:59:39 2025

Issue 394692559