Re: [PATCH 21/22] teach the merge algorithm about cache iterators

Daniel Barkalow [off-list ref] writes:

Another thing to try would be the original dynamic table implementation, 
plus a hashtable for name lookups, generated the first time a lookup is 
attempted (since some programs don't do any lookups by name). This has the 
advantage of skipping the O(n) startup.

How about just the original dynamic table implementation with
the original binary search name lookups?  Am I missing
something?