tree b61858a4dd37ca6e0bb0d587c4cf74eff471d27c
parent 577bb7dba11ef2f6588664dcface13fba32bb354
author Dan Scales <danscales@google.com> 1633968902 -0700
committer Dan Scales <danscales@google.com> 1633979723 +0000

cmd/compile: deal with TODO related to generic wrappers with embedded fields

It turns out there is no real TODO here - things are working fine. If we
are generating a wrapper for a method on a generic type that is actually
the method on embedded type, then we should just just generate the
normal embedded wrapper, which calls the wrapper for the real receiver
type on that method. There is no need to do the generic path where we
add in the dictionary argument. So, just updated that TODO comment with
this explanation.

Added a new test case embedded.go, which specifically tests various
situations involving converting to empty and non-empty interfaces.
issue44688.go already tests a bunch of these situations as well.

Also made some other cleanups in reflect.go:

 - The shape test (that I had added) at the top of imethods is useless
   (never true), since it is always an interface type, so removed it.

 - Added usual helper function deref() to make code clearer in several
   places.

 - The shape test in methodWrapper() doesn't have to check HasShape() on
   each targ - it can just check HasShape() on the whole receiver.

 - The comment about disabling the tail call optimization for RegABI is
   no longer true.

 - Simplified code in several places by using the value of existing
   variable 'methodrcvr'.

Change-Id: I8b1a5dc518dad37585824a1f73ceebb7627a9f82
Reviewed-on: https://go-review.googlesource.com/c/go/+/355129
Trust: Dan Scales <danscales@google.com>
Reviewed-by: Keith Randall <khr@golang.org>