Cache transformed files

[floatdrop]

I have some tests hanging with cache, but implementation will look something like this. Opening to gather suggestions.

[sindresorhus]

Can you make sure it follows XDG?

https://github.com/sindresorhus/xdg-basedir#cache

[jamestalmage]

We are going to need a way to include the settings for the transform in the hash as well.

Perhaps it is:

  MD5(MD5(source_file_contents) + MD5(
    ava_version + '\n' +
    settings_file_contents + '\n' +
    ...???...
  ));

How do we handle changing dependency versions? What happens when the babel-plugin-transform-runtime has a patch level bump (meaning the output of the transform may have changed some for the same input). Are we going to have to scan all the plugin version numbers and make those part of the hash? I can't think of an other way.

However we do it, we should compute the part of the hash that is consistent across the test run once (the ava_version + plugin_hash part), and pass that to the forked processes so they only need to compute the hash of the file being required

[floatdrop]

@sindresorhus done.

@jamestalmage you are perfectly right. Until #202 is resolved we can hash with ava version.

[sindresorhus]

And I dig the module name ;)

[jamestalmage]

https://github.com/jamestalmage/babel-plugin-detective

It is based on Browserify's algorithm for finding dependencies.

Once we make the move to Babel 6, we can use this to precompile dependencies before ever launching the test. We will need other strategies for dynamic requires (possibly the server / client option I discussed earlier), but initially we can demand static requires if they want them transpiled (should cover 90% of use cases anyways).

[sindresorhus]

Closing in favor of #352.