[SR-176] Skip type checking case patterns when switch is error type.

[gregomni]

This stops emitting unhelpful diagnostics about “<>” (as in
SR-176) while still complaining about the switch itself and any other
errors in case bodies.

[lattner]

I will take a look at this in more detail this evening, but should coercePatternToType itself handle the errortype case? What behavior do other callers expect?

[gregomni]

Other callers:

For closure argument patterns, this only gets called after the closure's function type itself has been determined, and if any of the args were errortype, the closure would be errortype instead of a function type.

For exception catch patterns, this only gets called with the built in exception type.

For a for-each pattern, this only gets called during a BindingListener::appliedSolution(), which (I think) can only happen if all of the type constraints successfully bound to real types.

In short, it ought to be fine (just via inspection, haven't run tests or anything yet) to move this check in as a precondition inside coercePatternToType() if that seems like a better place for it, since no other callers will ever call it with that type.

[lattner]

If it will work, merging it into coercePatternToType sounds better to me, thanks!

[gregomni]

Turns out I was wrong in my inspection: The TypeChecker can call coercePatternToType() via typeCheckPattern() with an errortype while it is checking closure params, and it depends upon the side-effect of the types of those params being set to errortype as a result, among other things.

So to get the same result of supressing the bad diagnoses, the checks for errortype need to be in specific cases inside coercePatternToType() instead of a blanket precondition. Added a second commit that does that.

[lattner]

i'm sorry for losing track of this patch. The patch looks good to me, but I can't accept the PR because there is a silly conflict (someone renamed element to elt in TypeCheckPattern.cpp. Please update the patch and I'll be happy to accept it, thank you for fixing this!

[gregomni]

It's not like you have anything else to do or anything, sheesh! 😀 Rebased off the base branch and merged/squashed. Hopefully good now (tests all still pass, github says there are no conflicts).

[lattner]

Fantastic, thank you again for fixing this!