[GSB] Always ensure that we wire up typealiases in protocol extensions.

[DougGregor]

We were only wiring up typealiases in protocol extensions under some
circumstances, which meant that we could miss some equivalences between a
typealias in one protocol and an associated type in an inherited protocol,
which could manifest as a crash in IR generation.

Fixes SR-7097 / rdar://problem/38001269.

[DougGregor]

@swift-ci please smoke test

[DougGregor]

@swift-ci please smoke test compiler performance

[huonw]

I'm confused. As far as I can tell this change just means that updateNestedTypeForConformance now gets called with ArchetypeResolutionKind::AlreadyKnown in addition to the other two kinds, but, I can't see a way in which that call actually does anything.

https://github.com/apple/swift/blob/adfead025138474a5b4a6f3c1dee585c5654cc42/lib/AST/GenericSignatureBuilder.cpp#L2796-L2934

It seems like that function does three "interesting" things, but none of them happen with this kind:

• processes delayed requirements, but this is conditional on the kind being CompleteWellFormed https://github.com/apple/swift/blob/adfead025138474a5b4a6f3c1dee585c5654cc42/lib/AST/GenericSignatureBuilder.cpp#L2813
• creates a new archetype, but this is explicitly not meant to happen for AlreadyKnown https://github.com/apple/swift/blob/adfead025138474a5b4a6f3c1dee585c5654cc42/lib/AST/GenericSignatureBuilder.cpp#L2847-L2868
• does "more processing" to an archetype, conditional on shouldUpdatePA, which is only changed to true when creating a new archetype, i.e. it doesn't happen for AlreadyKnown https://github.com/apple/swift/blob/adfead025138474a5b4a6f3c1dee585c5654cc42/lib/AST/GenericSignatureBuilder.cpp#L2872

What am I missing?

[xedin]

Same question here...

[DougGregor]

You're both right and... somewhere I messed something up locally, because the test marked as fixed doesn't seem to be fixed. Very sorry :(

[swift-ci]

Build comment file:

Summary for master smoketest

No regressions above thresholds

• Debug
  • debug brief
  • debug detailed
• Release
  • release brief
  • release detailed

Debug

debug brief

Regressed (0)

name	old	new	delta	delta_pct

Improved (0)

name	old	new	delta	delta_pct

Unchanged (delta < 1.0% or delta < 100.0ms) (2)

name	old	new	delta	delta_pct
LLVM.NumLLVMBytesOutput	23,223,548	23,223,540	-8	-0.0%
time.swift-driver.wall	39.3s	39.2s	-16.6ms	-0.04%

debug detailed

Regressed (0)

name	old	new	delta	delta_pct

Improved (0)

name	old	new	delta	delta_pct

Unchanged (delta < 1.0% or delta < 100.0ms) (23)

name	old	new	delta	delta_pct
AST.NumImportedExternalDefinitions	44,325	44,325	0	0.0%
AST.NumLoadedModules	6,161	6,161	0	0.0%
AST.NumTotalClangImportedEntities	123,466	123,466	0	0.0%
AST.NumUsedConformances	2,816	2,816	0	0.0%
IRModule.NumIRBasicBlocks	67,316	67,316	0	0.0%
IRModule.NumIRFunctions	35,960	35,960	0	0.0%
IRModule.NumIRGlobals	35,332	35,332	0	0.0%
IRModule.NumIRInsts	732,937	732,937	0	0.0%
IRModule.NumIRValueSymbols	64,658	64,658	0	0.0%
LLVM.NumLLVMBytesOutput	23,223,548	23,223,540	-8	-0.0%
SILModule.NumSILGenFunctions	27,549	27,549	0	0.0%
SILModule.NumSILOptFunctions	31,631	31,631	0	0.0%
Sema.NumConformancesDeserialized	130,598	130,598	0	0.0%
Sema.NumConstraintScopes	416,611	416,611	0	0.0%
Sema.NumDeclsDeserialized	872,519	872,519	0	0.0%
Sema.NumDeclsValidated	29,522	29,522	0	0.0%
Sema.NumFunctionsTypechecked	26,107	26,107	0	0.0%
Sema.NumGenericSignatureBuilders	38,656	38,656	0	0.0%
Sema.NumLazyGenericEnvironments	163,947	163,947	0	0.0%
Sema.NumLazyGenericEnvironmentsLoaded	17,255	17,255	0	0.0%
Sema.NumLazyIterableDeclContexts	149,976	149,976	0	0.0%
Sema.NumTypesDeserialized	917,350	917,350	0	0.0%
Sema.NumTypesValidated	108,166	108,166	0	0.0%

Release

release brief

Regressed (0)

name	old	new	delta	delta_pct

Improved (0)

name	old	new	delta	delta_pct

Unchanged (delta < 1.0% or delta < 100.0ms) (2)

name	old	new	delta	delta_pct
LLVM.NumLLVMBytesOutput	27,721,024	27,721,016	-8	-0.0%
time.swift-driver.wall	69.1s	69.2s	26.3ms	0.04%

release detailed

Regressed (0)

name	old	new	delta	delta_pct

Improved (0)

name	old	new	delta	delta_pct

Unchanged (delta < 1.0% or delta < 100.0ms) (23)

name	old	new	delta	delta_pct
AST.NumImportedExternalDefinitions	6,977	6,977	0	0.0%
AST.NumLoadedModules	223	223	0	0.0%
AST.NumTotalClangImportedEntities	20,771	20,771	0	0.0%
AST.NumUsedConformances	2,818	2,818	0	0.0%
IRModule.NumIRBasicBlocks	63,693	63,693	0	0.0%
IRModule.NumIRFunctions	27,852	27,852	0	0.0%
IRModule.NumIRGlobals	29,179	29,179	0	0.0%
IRModule.NumIRInsts	557,514	557,514	0	0.0%
IRModule.NumIRValueSymbols	53,336	53,336	0	0.0%
LLVM.NumLLVMBytesOutput	27,721,024	27,721,016	-8	-0.0%
SILModule.NumSILGenFunctions	13,972	13,972	0	0.0%
SILModule.NumSILOptFunctions	20,753	20,753	0	0.0%
Sema.NumConformancesDeserialized	71,217	71,217	0	0.0%
Sema.NumConstraintScopes	397,535	397,535	0	0.0%
Sema.NumDeclsDeserialized	133,646	133,646	0	0.0%
Sema.NumDeclsValidated	19,193	19,193	0	0.0%
Sema.NumFunctionsTypechecked	7,393	7,393	0	0.0%
Sema.NumGenericSignatureBuilders	4,634	4,634	0	0.0%
Sema.NumLazyGenericEnvironments	20,629	20,629	0	0.0%
Sema.NumLazyGenericEnvironmentsLoaded	2,513	2,513	0	0.0%
Sema.NumLazyIterableDeclContexts	13,013	13,013	0	0.0%
Sema.NumTypesDeserialized	157,277	157,277	0	0.0%
Sema.NumTypesValidated	29,559	29,559	0	0.0%

---

[DougGregor]

This is the wrong way to go; the real fix is in #15002.