[antlr-interest] followup to v3 speed
Terence Parr
parrt at cs.usfca.edu
Sat May 27 11:56:34 PDT 2006
Ha! I knew I was forgetting something. v3 does much better error
recovery, but at the cost of a push/pop on a software stack for each
method call. When I comment those out, I get
parsing+lexing 3696 ms
lexing: 1503 ms
parsing+lexing+ast 7514 ms
So that means the parsing time is about 2193ms and tree construction
is about 3818ms. The relative percentages are now:
lexing 20%
parsing 30%
ast 50%
So, without this extra push/pop, v3 is already 25% faster than v2 for
java parsing and tree building combined. If I can get the tree
building down further, we'll be even faster. v2 is 2805ms for parsing
+ast versus 6011ms in v3. Gotta get that faster.
Now the big question: do I leave in the good error handling by
default or is that an option?
Ter
More information about the antlr-interest
mailing list