[IRGen] Fix emitPrimitiveLoadPayloadAndExtraTag for CVW #75270
Conversation
|
@swift-ci smoke test |
rdar://129627898 When casting the projectedBits to Int8, we accidetnally passed the base addr instead of the projectedBits. This was causing the wrong bits to be read.
|
@swift-ci smoke test |
| case x(AnyObject) | ||
| case y(Int) | ||
| case x(Int) | ||
| case y(AnyObject) |
There was a problem hiding this comment.
The reason for this change is that the bug was masked by the fact that we were using tag 0 in the test. The bug caused the first byte of the object to be read, instead of the tag byte. When this is a reference, it will always cause the tag to be read as 0, causing the test to succeed. So now we are using 1, which should make the test more reliable.
There was a problem hiding this comment.
Should we duplicate this test and verify both declaration orders? (For the main PR only; the 6.0 cherry-pick doesn't need to wait for that.)
There was a problem hiding this comment.
For that matter, what happens with a 257-element enum? Does that end up with a 9-bit tag?
There was a problem hiding this comment.
No, we only treat 1-bit tags like that, everything else uses full bytes. Yes, I can add the other order as well.
rdar://129627898
When casting the projectedBits to Int8, we accidetnally passed the base addr instead of the projectedBits. This was causing the wrong bits to be read.