[SR-16088] Parameterized result builder breaks type inference #58347
Description
| Previous ID | SR-16088 |
| Radar | None |
| Original Reporter | @stephencelis |
| Type | Bug |
Environment
Xcode 13.3 / Swift 5.6
(Also fails with 2022-03-22 snapshot)
Additional Detail from JIRA
| Votes | 4 |
| Component/s | Compiler |
| Labels | Bug |
| Assignee | None |
| Priority | Medium |
md5: cfd91f8a83f3d37d1340e06d4b9d8b10
Issue Description:
The following gist has 2 versions of similar code. The main difference (shown in a diff) is that one result builder is parameterized with a generic that is used to constrain its "buildExpression" to help with type inference, the other is not parameterized.
https://gist.github.com/stephencelis/e6fa9caed1daa054e24e9d23c186bf2d
The code is a simplified parsing DSL that should automatically skip "Void" outputs (much like the Regex DSL being currently worked on). Unfortunately, parameterizing the result builder allows things to compile that shouldn't: the unparameterized version ensures "Void" outputs never make their way into the final parser, while the parameterized version allows these values to make their way through.