Closed cowtowncoder closed 9 years ago
Given that there is the new hash-based symbol table impl in 2.6, I think I'll skip this one. Earlier attempts at using tries have suggested they can't match the performance; and profiling does not suggest that symbol table access was a major performance issue at this point either.
While existing symbol table implementations to reasonable job, it might be possible to get an even faster (and possibly slightly more multi-threading friendly!) implementation by using a trie that is not byte-based, but uses quadlets (int32 containing 4 bytes) that are already being used internally.
Benefits in this case are:
The main idea is just to see if such structure is feasible; at parser level it would only have to be as fast as symbol table, if it would open up improvements for data-binding.