Fix assertion from empty switch over uninhabited enum #19979
Conversation
Fixes SR-8933. Swift's uninhabited checking is conservative. An enum might not be found to be uninhabited, but all of its cases might. In that case, the switch can be empty even though the type isn't known to be uninhabited. Fix an assertion that assumed there would always be at least one case for types that aren't known to be uninhabited.
|
@swift-ci Please test |
|
@swift-ci Please test source compatibility |
|
Looks like those failures aren't due to my changes here. 😅 (Side note: Is there an easy way to reproduce and debug those?) Hopefully this is good to go. 😄 |
|
Yes, I might have some time later today to do this. |
|
Looked into this a bit further, and the source compatibility failures appear to be duplicates of SR-9084, which has a fix in #20091. So far, testing looks good. Unless things take a turn for the worse, this should be resolved upon merge. I'm going to hold off on adding any xfails at this point, since the fix is imminent. Neither of the source compatibility project failures appear to be related to changes in this pull request, so unless getting a passing build from source compatibility (also note that it's only failing in debug... the release configuration has passed), feel free to proceed. |
Fixes SR-8933 | rdar://problem/45216708.
Swift's uninhabited checking is conservative. An enum might not be found to be uninhabited, but all of its cases might. In that case, the switch can be empty even though the type isn't known to be uninhabited.
Fix an assertion that assumed there would always be at least one case for types that aren't known to be uninhabited.
This is the approach that was suggested by @jckarter in the ticket. (I hope I've done it correctly.)