[antlr-interest] Debugger performance indicative of actual performance?

[Steve Bennett]

On 2/3/08, Mark Volkmann <r.mark.volkmann at gmail.com> wrote:
> I can't answer the question. However, I've never found it useful to
> process that much input in the debugger. Typically I'm using it to
> verify that my grammar is producing the AST that I want for certain
> kinds of input. The amount of input required to verify that for me is
> typically 10 lines or less. Looking through the AST for 7 KB of input
> seems overwhelming.

Ah. The main reason is that my project is basically attempting to
define a grammar for a language with a massive corpus but no real
specification. So I'm feeding it lots of medium sized files just to
find out what real-world language features have yet to be added to the
grammar.

Jim Idle wrote:
>Extremely unlikely that the debug version is giving you any performance
>information. The biggest problem with that java stuff is that Java takes
>a hundred years to initialize, then the parser executes in 0.003
>seconds. You are probably seeing the startup time for the debugger and
>so on and not the actual parsing.

Ok, that's promising, though I notice that from the time the first
output appears to the final output is still a few seconds. I guess
there are ways to render the grammar more efficient though.

Oh, and I took the obvious (silly me) advice to just try hte damn
thing in a test harness: 0.687 seconds for one 7kb file. That seems
relatively promising.

Steve