querystring: improve performance

[mscdex]

Results:

                                                                   confidence improvement accuracy (*)     (**)    (***)
 querystring/querystring-parse.js n=1000000 type='altspaces'             ***    985.03 %      ±27.88%  ±40.04%  ±58.85%
 querystring/querystring-parse.js n=1000000 type='encodefake'            ***   1253.69 %      ±26.56%  ±38.14%  ±56.07%
 querystring/querystring-parse.js n=1000000 type='encodelast'            ***   1012.30 %      ±26.81%  ±38.49%  ±56.56%
 querystring/querystring-parse.js n=1000000 type='encodemany'            ***    712.02 %      ±34.90%  ±50.11%  ±73.65%
 querystring/querystring-parse.js n=1000000 type='manypairs'             ***      4.42 %       ±1.35%   ±1.85%   ±2.52%
 querystring/querystring-parse.js n=1000000 type='multicharsep'          ***   1053.69 %      ±21.01%  ±30.15%  ±44.28%
 querystring/querystring-parse.js n=1000000 type='multivalue'            ***   1161.95 %      ±41.03%  ±58.93%  ±86.66%
 querystring/querystring-parse.js n=1000000 type='noencode'              ***   1094.45 %      ±96.37% ±138.44% ±203.66%

 querystring/querystring-stringify.js n=1000000 type='array'             ***      8.28 %       ±2.20%   ±2.94%   ±3.84%
 querystring/querystring-stringify.js n=1000000 type='noencode'           **      3.38 %       ±2.43%   ±3.24%   ±4.21%

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• commit message follows commit guidelines

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/25224/

[mscdex]

I believe this may be applicable to v10.x as well (in addition to v12.x of course), but I have not yet tested.

[addaleax]

Do you know why the perf difference is so large? It doesn’t look like there are any substantial changes to the code itself, which makes me curious…

[bnoordhuis]

I'm also curious. I would expect preallocating the charCodes() array to be 2-3x faster but that only explains the multicharsep benchmark and that still doesn't explain the 10x speedup. :-)

Nice work at any rate!

[mscdex]

I didn't dig that deep to find out why but I think V8 is not optimizing the parse() function? I also noticed that node v8.x is ~5x faster or so than node v10.x+ when it comes to parse() (without these changes).