Prefer the locally installed version of AVA.

[jamestalmage]

This hunts for AVA in the locally installed project, and runs that
CLI instead of the globally installed one.

Fixes #157.

[jamestalmage]

The second commit wraps the entire CLI in an IIFE, to avoid a ParseError from xo.

eslint/eslint#1158

[jamestalmage]

If xojs/xo#51 were to be implemented, the second commit would not be necessary.

[jamestalmage]

This needs a test suite, which is doable but will involve a number of fixtures. I would like agreement on my general approach before completing.

Right now, it outputs the following:

• Run locally, or globally without a local version: no warning issued

• Run globally, with a local version:

  Using local install of AVA (v0.6.1), which differs from global (v0.6.0)
  

  or

  Using local install of AVA (v0.6.0), which matches global (v0.6.0)
  

[sindresorhus]

Instead of wrapping it in an IIFE, just remove XO for now. We can add it back when we've sorted this out.

[sindresorhus]

Right now, it outputs the following:

I don't think it matters if they differ. All we're doing with the global CLI if local is available is to require the local one. As long as the user has AVA 0.7.0 containing this pull request installed globally they should be fine.

[sindresorhus]

Been thinking about changing resolve-from to just return null when not found instead of throwing. Thoughts?

[shinnn]

FYI try-resolve-from works as you expected. https://github.com/shinnn/try-resolve-from/blob/50cd2cb89934ec0172e43db86fe78263759908ee/index.js

[jamestalmage]

I think resolve-from should adopt that behavior. I can't think of very many scenarios where you would want resolve-from, and not want to wrap the thrown exception.

Also, it looks like someone was already assuming that behavior here: babel.js#L41. This looks like the more common use-case.

Perhaps retain the old behavior somewhere:

require('resolve-from/throws');

or something like that.

[shinnn]

OK I'll unpublish try-resolve-from. sindresorhus/resolve-from@4cb3cf0

[jamestalmage]

Instead of wrapping it in an IIFE, just remove XO for now. We can add it back when we've sorted this out.

This is what led me to the shim behavior in fallback-cli, it allows you to do this:

// cli-shim.js
require('fallback-cli')('ava/cli.js', function (opts) {
  // returns are allowed inside the callback - linters stay happy.
  if (opts.globalCli && opts.localCli) {
    console.warn('Using the local install of AVA instead of global.');
  }
  require(opts.cli); // opts.cli === opts.localCli || opts.globalCli
});

[vadimdemedes]

I don't think we need that warning, it gives no helpful information to the test output. Plus, it will not look good.

[sindresorhus]

Maybe this is something we could have in a future --verbose mode or just add it as a debug()?

[jamestalmage]

OK. Honestly, that line was there to appease you: xojs/xo#32 (comment).

I thought you brought up a good point, that silently changing standard behavior is a bad thing.

@sindresorhus, what do you think?

[sindresorhus]

@jamestalmage Bump resolve-from and it will no longer throw ;)

[sindresorhus]

Generally looks good to me.

[sindresorhus]

You don't have to check if debug is enabled. That's already done by debug

[sindresorhus]

One minor comment. Otherwise it's good. Feel free to merge when that's resolved.