Performance issues after Babel 6 upgrade

[vadimdemedes]

Since babel 6 upgrade the performance is seriously ****ed up. Using master branch and the following test file:

var test = require('./');

test('wow', t => {
    t.true(true);
});

$ time ava example.js
ava example.js  4.04s user 0.39s system 101% cpu 4.344 total

Running it for like 10th time and getting the same results.

Would be great if someone else confirmed this.

[sindresorhus]

Confirmed.

~/dev/ava master*
❯ time ava example.js

  ✔ wow

  1 test passed


real    3.46s
user    3.21s
sys 0.32s

[sindresorhus]

If I run npm dedupe and then try again, it's faster.

❯ time ava example.js

  ✔ wow

  1 test passed


real    1.40s
user    1.30s
sys 0.14s

[sindresorhus]

Babel 6 is a lot more modular, so that might be where all the overhead comes from. Using npm@3 would also be faster (since it dedupes by default).

[vadimdemedes]

Then we are stuck with initial delay on every test run. Not good.

[vadimdemedes]

I think the solution to this is to cache babel's output. There already was some development in that direction in #189, but left incomplete.

My proposal is to cache transformed code based by md5 of original source code. That way, we don't even need to require('babel'), when test files don't change. When md5 changes, load babel and re-transpile source code.

This should improve performance.

[jamestalmage]

using npm 3

That is basically a requirement of Babel 6. npm dedupe is the only other option.

caching

I just finished implementing much the same for nyc with very good results. It's part of my plan after I land #349

[vadimdemedes]

Seems like #349 will take good amount of time. I can implement cache now and send a quick PR, so that we can do a release with babel 6 within next few days.

[sindresorhus]

This can be closed now, right?

[vadimdemedes]

Yep, #352 got it covered.