This looks good. Just one comment to address.

Reviewed 2 of 2 files at r1.
Reviewable status: 1 of 2 files reviewed, 1 unresolved discussion (waiting on @apolcyn, @AspirinSJL, @markdroth, and @mhaidrygoog)

src/core/lib/gprpp/map.h, line 44 at r1 (raw file):

struct RefCountedPtrLess {
  bool operator()(const RefCountedPtr<T>& p1, const RefCountedPtr<T>& p2) {
    return *p1 < *p2;

I don't think this is the right default behavior for RefCountedPtrLess<>, because not all ref-counted types will define operator<(). I think RefCountedPtrLess<> should compare only based on addresses. This is analogous to how std::less<> provides a default specialization for pointer types like std::shared_ptr<>, which compare only by address.

I think it's fine to implement a more specific comparison for LocalityName, but that should be a custom type defined in the xds code, not something implemented here in the map library.

Reviewable status: 0 of 1 files reviewed, 1 unresolved discussion (waiting on @apolcyn, @markdroth, and @mhaidrygoog)

src/core/lib/gprpp/map.h, line 44 at r1 (raw file):

Done.

This looks good. Feel free to merge after addressing the remaining comment.

Reviewed 2 of 2 files at r3.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @apolcyn, @AspirinSJL, and @mhaidrygoog)

src/core/ext/filters/client_channel/lb_policy/xds/xds.cc, line 362 at r3 (raw file):

      bool operator()(const RefCountedPtr<LocalityName>& lhs,
                      const RefCountedPtr<LocalityName>& rhs) {
        int cmp_result = gpr_stricmp(lhs->region_.get(), rhs->region_.get());

Do we want all of these comparisons to be case-insensitive? Unless the XDS API specifically indicates that all of these names should be case-insensitive, this seems like the wrong thing to do.

Thanks!

Reviewable status: 0 of 1 files reviewed, 1 unresolved discussion (waiting on @apolcyn, @markdroth, and @mhaidrygoog)

src/core/ext/filters/client_channel/lb_policy/xds/xds.cc, line 362 at r3 (raw file):

Done.

The spec doesn't say that. I was following the previous comparing code. But I agree it should be case sensitive.