Use zero-based numeric indices for much faster instrumented code

[mourner]

Closes #21. Makes instrumented code run much faster by making V8 use “fast” mode when accessing stats objects, which wasn't triggered before because of non-0-based indices.

Here are the timings from mapbox-gl-js render suite:

• no coverage: 1m 6s
• nyc current: 4m 37s (or 3m 31s overhead)
• nyc updated: 1m 20s (or 14s overhead)

So, basically the performance overhead is 15 times smaller, down from 320% to 20%.

• update tests

cc @bcoe @mramato

[yahoocla]

CLA is valid!

[mourner]

@bcoe Travis passes, so this ready to go.

[mourner]

@bcoe @gotwarlost anything I could help with to get this merged? A 15x speedup seems like something that's worth prioritizing. :)

[aaronabramov]

this is amazing! ❤️

[bcoe]

@DmitriiAbramov @mourner I see no reason not to land this; but since I haven't heard back from @gotwarlost, I will be testing quite throughly; I'll start by publishing this to a next tag.

[bcoe]

@mourner I've published these changes to the next tag on npm, if you'd like to help test a bit:

npm cache clear; npm i istanbul-lib-instrument@next

[cpojer]

Oh this is awesome! Is anyone interested in trying this out on https://github.com/facebook/jest itself? :) (run jest with --coverage).

[mramato]

Thanks @mourner, I would love to try this out to confirm my issue in gotwarlost/istanbul#556 is gone. Unfortunately I currently rely on https://github.com/karma-runner/karma-coverage which hasn't updated to this library yet (but I'm hoping it happens soon or I can open a PR to do it for them).

[aaronabramov]

@cpojer i tested it on jest itself, but we don't really instrument much code, so there isn't much we can optimize :)
although it seems to run 1 second faster, 28s on average instead of 29s

[mourner]

@bcoe thanks! The new version works perfectly in our project:

• no coverage — test.sh takes 4m53s
• nyc latest — test.sh takes 11m22s
• nyc new — test.sh takes 5m56s

[gotwarlost]

So sorry about my lack of response (this is getting old :( )

I see no reason off the top of my head why this should not be merged. 👍

[gotwarlost]

ah, it is merged already. Great.