Make special subclass from Parser without expectations.

[dikmax]

Some parsers shouldn't have error messages, they always should produce success results. For such parsers caring about expectations is really expensive and unnecessary. Idea is to make some subclass or some flag to exclude everything unneeded. I've tested on my library (just copied and rewrote many, skipMany and choose methods) and parser started working 30% faster.

[polux]

Sounds like a good idea, although I'm not sure I understand all the implications yet, nor what is the best way to do it. Did you profile your markdown parser using obervatory? If yes, where is this extra time spent after your change to lazy expectation? In CombinedExpectation allocations?

[dikmax]

Actually EmptyExpectation allocation was higher. CombinedExpectation allocation + EmptyExpectation allocation takes around 8-10% of time. But combined with other operations like checking expectations parameter while ParseResult.copy and so on, gives ~30% less time for parsing.

You can check version without expectations here: https://github.com/dikmax/parsers/tree/light-parser

I don't know best way to do this, too. Version above is just to testing.

BTW, after removing expectation most expensive operations become Parser.>> and ParseResult allocation.

[dikmax]

Also removing expectation will give room for additional optimization. For example, we could replace in operator >>:

return res2.copy(
            isCommitted: res.isCommitted || res2.isCommitted);

with

return res2.isCommited == res.isCommitted || res2.isCommitted ? res2 :
        return res2.copy(isCommitted: !res2.isCommited);

That would eliminate ParseResult allocation in really many cases.