Fix Interpreter/dynamic_self.swift test for ExistentialSpecializer

[rajbarik]

No description provided.

[rajbarik]

@swift-ci test.

[rajbarik]

@swift-ci test linux

[atrick]

I think you should comment why the attribute is necessary, at least in the commit message but preferably in test itself--the optimizer may specialize the call and attempt to inline, but inlining that would defeat the purpose of the test which is intended to exercise IRGen on these function bodies. Inlining may also cause self to be destroyed before the print statement executes.

[slavapestov]

@rajbarik Does the behavior of the test change without the @inline(never)? If so it's a bug in the specializer.

[slavapestov]

@atrick I don't understand what you're suggesting here. Optimizations should never change the behavior of any code. The test is run with both optimizations off and on (check-swift-validation and check-swift-validation-optimize) and the latter has caught optimizer bugs before.

[slavapestov]

Oh I see. How about we just remove the print statements from the deinits.

[atrick]

@slavapestov I think it's a problem in general when IRGen unit tests are written in .swift. I agree it's best not to change the RUN line. So the question is whether you still want the test to be useful at -O. My point is that if the test is actually useful, then those print statements will be in the correct order. It seems strange to me to write passing CHECKs for a test that's already broken. But I don't have a strong opinion.

[slavapestov]

@atrick The part of the test that tests when the deinit occurs is not very useful. We should remove it. The rest is still useful. It's mostly an execution test for code generated by IRGen, but it has randomly turned up optimizer crashes in the past, mostly because our coverage of DyamicSelfType elsewhere in the optimizer wasn't great.

[atrick]

@slavapestov sounds fine to me. I was just guessing at the purpose of the test, and couldn't help but notice that IRGen will never even see a dynamicself in the -O build.

@rajbarik I guess you can just remove the print statements in the deinitializers, then you don't need the @inline(never).

[rajbarik]

@swift-ci test.

[rajbarik]

@swift-ci test linux

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 1304a0a97e2874cd746aa390b2d5f8c5c923ee5e

[rajbarik]

@swift-ci test.

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 4a2344213849f9deca888607f341161cf208fdf5

[swift-ci]

Build failed
Swift Test OS X Platform
Git Sha - 1304a0a97e2874cd746aa390b2d5f8c5c923ee5e