querystring: improve parse() performance

[mscdex]

Here are the results with existing benchmarks (and newly added input/type) for the changes in this PR:

                                                                  improvement confidence      p.value
 querystring/querystring-parse.js n=5000000 type="encodefake"          7.60 %        *** 5.588287e-48
 querystring/querystring-parse.js n=5000000 type="encodelast"         23.85 %        *** 3.772582e-43
 querystring/querystring-parse.js n=5000000 type="encodemany"         13.75 %        *** 7.027748e-27
 querystring/querystring-parse.js n=5000000 type="manypairs"           7.09 %        *** 2.271675e-49
 querystring/querystring-parse.js n=5000000 type="multicharsep"       19.71 %        *** 6.757215e-33
 querystring/querystring-parse.js n=5000000 type="multivalue"         37.01 %        *** 4.461087e-58
 querystring/querystring-parse.js n=5000000 type="multivaluemany"     33.38 %        *** 4.410104e-59
 querystring/querystring-parse.js n=5000000 type="noencode"           27.66 %        *** 8.234363e-41

CI: https://ci.nodejs.org/job/node-test-pull-request/5932/

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• commit message follows commit guidelines

Affected core subsystem(s)

• querystring

[targos]

Have you tried to replace the uses of array[array.length] = value with array.push(value)?

[mscdex]

Have you tried to replace the uses of array[array.length] = value with array.push(value)?

I've looked into it before. The former is still slightly faster in master.

[jasnell]

is there any reasonable perf impact moving to a switch?
e.g.

switch (typeof v) {
  case 'string':
    return v;
  case 'boolean':
    return String(v);
  case 'number':
    if (isFinite(v)) return String(v);
  default:
    return '';
}

[mscdex]

I dunno, I didn't do anything with this particular function. However, I remember we switched away from that type of usage (no pun intended) because V8 optimizes typeof xx === yyy statements. Before we switched to separate if statements, the code was storing the result of typeof and that was causing a performance degredation. I suspect switch(typeof v) might have the same effect.

[jasnell]

fair enough :-)

[mscdex]

I just noticed that this time last year I was making performance improvements to the same querystring functions. Totally coincidental!

[caub]

Just a question, how does it compare with the new searchParams in performance?

[jasnell]

This would need a backport PR to land in v4. Landed in v6