On Thu, Sep 06, 2018 at 11:32:41PM -0700, Jonathan Nieder wrote:

quoted

I think Stefan pointed out a "case 4" in the other part of the thread:
ones where we really care not just about fast lookup, but actual
iteration order.

I had assumed that that was the whole point of this data structure.
Anything else that is using it for lookups should indeed use a hash
map instead, and I can take my share of blame for missing this kind of
thing in review.

Keep in mind we didn't have a decent generic hashmap for many years. So
I think string-list got used in its place.

quoted

I think I like the hashmap way, if the conversion isn't too painful.

If we don't have any callers that actually need the sort-and-lookup
thing, then yay, let's get rid of it.  But I don't actually think of
this as the hashmap way.  It's the get-rid-of-the-unneeded-feature
way.

In other words, *regardless* of what else we should do, we should
update any callers that want a hashmap to use a hashmap.  Please go
ahead, even if it doesn't let us simplify the string list API at all.

Great, I think we're on the same page. Thanks!

-Peff