📝 I'm debating whether we should add caching of MemberNames in SourceMemberContainerSymbol at this point... Thoughts?

I'm debating whether we should add caching of MemberNames in SourceMemberContainerSymbol at this point... Thoughts?

I would prefer that... but it's not necessarily a huge deal if you don't (unlesss people have tons of records).

The purpose of these collectoins is to allow the IDE to search for names very efficiently (i.e. in a non-allocating fashion). So it is important that the vast majority of source types not allocate here. And, as long as that holds we'll be ok.

But it would be a bit nicer to cache IMO in case you had someone with, say, 1k generated records, and we kept generaitng this over and over again.

Note: You don't need to do Distinct. MemberNames is explicitly doc'ed as returning dupes. exactly because we don't want to do the CPU/memory work to find and remove duplicates for htese sensitive scenarios.

MemberNames is explicitly doc'ed as returning dupes

Where do you see that?
The implementations I looked at (such as source named type symbols) seem to explicitly de-dupe.

        public ICollection<string> MemberNames
        {
            get
            {
                if (_lazyMemberNames == null)
                {
                    var names = UnionCollection<string>.Create(this.Declarations, d => d.MemberNames);
                    Interlocked.CompareExchange(ref _lazyMemberNames, names, null);
                }

                return _lazyMemberNames;
            }
        }

The merged decl for named types (i.e. for partials) just literally 'combines' (i.e. concatenates) all the names of all the individual parts. SourceMemberContainer just does this:

        public override IEnumerable<string> MemberNames
        {
            get
            {
                return IsTupleType ? GetMembers().Select(m => m.Name).Distinct() : this.declaration.MemberNames;
            }
        }

So this will return dupes. We do not do any sort of 'distinct' as we want this not to cost anything in terms of memory/CPU. This was very intentional from teh impl. Source: i wrote it :)

@CyrusNajmabadi Thanks! Removed the .Distinct :-)

is order guaranteed here? should this test sort, just to be sfe against that?

We probably could relax the order checking if we feel the need.