[parse] Recover better from malformed subscript decls for code-completion

[benlangmuir]

Code-completion of generic types expects to get a DeclContext for the
subscript, so make sure we recover well enough to produce a
SubscriptDecl.

rdar://35619175

[benlangmuir]

@swift-ci please test

[benlangmuir]

@DougGregor any concerns about the type-checker changes or in general about subscript element type now being potentially null?

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 06703dbd80ef8419b8ce5a741cabf6ee7b932d6b

[swift-ci]

Build failed
Swift Test OS X Platform
Git Sha - 06703dbd80ef8419b8ce5a741cabf6ee7b932d6b

[slavapestov]

Can we set the element type to ErrorType in the recovery path in the parser instead, so that downstream clients don't have to handle it?

[slavapestov]

Talked to @benlangmuir in person, we're both happy with the fix now.

[benlangmuir]

The test failure is a crash in a case we used to give up earlier. We're crashing because there is no interface type on the parameter of the getter for the subscript:

frame #4: 0x0000000102a25878 swiftc`swift::ValueDecl::getInterfaceType(this=0x0000000113847f80) const at Decl.cpp:1923
frame #5: 0x0000000102bfdca1 swiftc`swift::ParameterList::getInterfaceType(this=0x0000000113848018, C=0x0000000114009a00) const at Parameter.cpp:147
frame #6: 0x0000000102249ef1 swiftc`swift::TypeChecker::configureInterfaceType(this=0x00007ffeefbf7c78, func=0x0000000113848040, sig=0x0000000113849570) at TypeCheckGeneric.cpp:910
frame #7: 0x000000010220230e swiftc`(anonymous namespace)::DeclChecker::visitFuncDecl(this=0x00007ffeefbf70f8, FD=0x0000000113848040) at TypeCheckDecl.cpp:5291

But I don't see how this was going to work. In the parser, if a subscript is not inside a container type, we immediately set its interface type:

  if (Invalid) {
    Subscript->setInterfaceType(ErrorType::get(Context));
    Subscript->setInvalid();
  }

And the DeclChecker sees that the decl has a type, so it early-exits without setting types on the parameters of the Subscript. Then we propagate the same types onto the Decl for the accessor, which then asserts because it expects a parameter type.

@slavapestov what part(s) of this chain of events are incorrect? I'm not sure how to proceed.

[benlangmuir]

@swift-ci please test

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 06703dbd80ef8419b8ce5a741cabf6ee7b932d6b

[swift-ci]

Build failed
Swift Test OS X Platform
Git Sha - 06703dbd80ef8419b8ce5a741cabf6ee7b932d6b

[benlangmuir]

Update: I talked to Doug in person about this and he suggested we not set the interface type in the parser but let the DeclChecker do it correctly. I also took another look at how parameters are handled and discovered that while they can be null, that only should happen with closures not with functions, so instead of using null I'm setting ErrorTypeRepr on the element type to be more consistent, which is also closer to what Slava originally suggested.

[benlangmuir]

@swift-ci please test