Replace babel require hook with transformFileSync

[floatdrop]

Start of discussion - 52b0ccb#commitcomment-13971327

If you have babel and want to use require hook in application code and want to run tests with ava - you can get syntax errors (SyntaxError: Unexpected reserved word) because babel will get options from ava.

[sindresorhus]

Runtime should still be included, even if hasGenerators is true.

[sindresorhus]

Does the caching still work with this solution? Since Babel is relatively slow, the caching is very important.

[floatdrop]

@sindresorhus I'm afraid, cache should be implemented by hands.

co must be replaced (or patched to use Promise polyfill, since babel now will not do this for us).

[vadimdemedes]

@floatdrop in order to support the stuff that "generator community" is used to, co is required. Babel's built-in implementation is very limited.

[sindresorhus]

@floatdrop Maybe you could either extract this into a separate module we can depend on or do a PR on Babel to expose it?

[floatdrop]

@vdemedes yeah, I know, but you can't use co in Node.js 0.10 without some transpiler (and I don't think, that co will adapt to it soon tj/co#244). So we need to come up with something. I believe, that transpiling only test code will improve performance compared to patched require function.

@sindresorhus okay, but need to do some performance tests - I wonder, how fast cache file will fill up and parsing time eat all the benefit.

[vadimdemedes]

@floatdrop generators are only being used in node 0.11, iojs and node v4.x projects, so we don't have to worry about that.

[floatdrop]

@vdemedes co uses global Promise object - so you can't use it right away in Node.js 0.10.

[vadimdemedes]

@floatdrop Hmm, did not understand the point. I am just saying that nobody uses generators on the old node.js versions.

[floatdrop]

@vdemedes generators are not a problem, yes. But ava do require co at the top of test.js and this leads to errors in 0.10 - https://travis-ci.org/sindresorhus/ava/jobs/87312158

[floatdrop]

JFYI: seems like babel require hook have some polyfill side-effect - https://github.com/floatdrop/ava-in-node-0.10

[floatdrop]

Just to keep track - waiting for tj/co#250

[sindresorhus]

Doesn't look like co has much activity, so I don't think we should wait on that tbh.

[sindresorhus]

Why do we even depend on co again? Can't Babel+Regenerator do this for us?

[vadimdemedes]

Because co implements some very nice features, which babel's regenerator does not implement. Babel's implementation only makes Promises yieldable, but nothing else.

Why don't we use @floatdrop's fixed fork for now, until they merge his fix into master?

[sindresorhus]

Why don't we use @floatdrop's fixed fork for now, until they merge his fix into master?

👍

[sindresorhus]

"co": "floatdrop/co#343xsv324",

use the shorthand format and use a specific commit hash (first 6 chars of the hash is enough), otherwise the npm cache will become stale.

[floatdrop]

🆗

[sindresorhus]

okay, but need to do some performance tests - I wonder, how fast cache file will fill up and parsing time eat all the benefit.

@floatdrop I'm ok with merging this now if you commit to looking into caching later? Compilation is a lot larger overhead than just reading in a cached file from disk. I do think we can do caching better than Babel itself, though. Mainly using separate cache files for each test file (Babel uses one large file) and a LRU cache.

[sindresorhus]

lol, that was a fake hash, i don't know which hash you want to use.

[floatdrop]

@sindresorhus sure. I'm started to use ava in some work related projects, so I will hang around for some time :)

[sindresorhus]

I'm started to use ava in some work related projects

😍

[sindresorhus]

This looks good to me. @vdemedes Can you take a final look?

[vadimdemedes]

Good to go! 👍

[sindresorhus]

Thanks again @floatdrop :)