The main fixes here are:
- we weren't looking through open-existentials in the l-value
- we weren't handling mutating gets correctly unless CSApply wrapped
the base in an InOutExpr, which seems to be multifile-sensitive
- we were missing diagnostics in some cases involving subscripts
A better fix would be to re-introduce LValueAccessKind, but I wanted
a workable short-term fix that I could try to get into 5.1.
Fixes rdar://49482742, which is specific to the lazy-getter problem.
If we computed captures before completing a lazy getter body, we would fail to
consider the 'self' capture properly. Instead make it resilient to such ordering
issues by checking in capture computation if the lazy property has a getter yet
or not.
We were trying to do this when synthesizing the getter prototype, but
we don't do that immediately when we're just type-checking a reference
to the storage, which could lead to the reference thinking that the
getter was non-mutating.
Fixes rdar://45712204.
