make api more generator style about co-webdriver-runner

[kesla]

Oh, thanks. That's great feedback! I wasn't that happy with the mix generator style / callback style, so thanks for pointing out that we should be able to solve it in another way.

I like the simplicity with the current API and what you're suggesting (with keeping track of process.on('exit')) seem a bit more complex to me, so I think that there's room for some iteration on this!

How do you think about a flow that works somewhat like this?

var each = require('co-each')
var runners =  yield require('co-webdriver-runner')({ baseConfig: '...', browsers: '...', local: false })

yield each(
    runners
  , function* (runner) {
      var test = runner.test
      var browser = runner.browser
      // this means that here the browser won't be initialized, so you need to do stuff with it in a test-scope
      test('title', function* (t) {
        // ...
      })
      // in start we initialize the browser
      // and then run the tests
      yield runner.start()
    }
)

[juliangruber]

yeah...what do you think about this?

var test = runner({ ... });

yield test('name', function*(t){
  // use t.browser or test.browser
});

yield test...

I think this would be the nicest and most consistent api

EDIT and you don't need process.on('exit')

[juliangruber]

both would work! but if a saw your last example in a readme i guess i'd be hesitant to use the module, because of the amount of code needed to get startet

[dey-dey]

+1 @juliangruber to this approach, great thinking here. seems more digestible when reading.

@kesla this is looking really good! excited to put this to use :)

[kesla]

@juliangruber How would you setup something specific for a browser with your suggestion? Let's say a http-server running on a certain port for every browser?

[juliangruber]

@kesla can you explain further please?

[kesla]

@juliangruber Sure, to speedup the running times the tests are ran in parallel when ran on sauce labs. Every browser is queued and executed as quickly as possible. Looking at runner.js from condor, we then start one separate http-server for each browser and one instance of localtunnel to make them reachable from the outside world. This is also the reason we're creating multiple tap-harnesseses here, so that the result from each browser can be buffered up and then presented in one go, when that browser has finished running.

[juliangruber]

ok i think i get it^^

wouldn't you orchestrate your tests just the same as with the current api?

really, i think we can get rid of all yield and just have it be

var test = require('runner')(opts);

test('foo', function*(t){
  // use t.browser
});

so it's mostly the same api as tap

[kesla]

Hm, I'm still having troubles understanding exactly what you mean.

In the current api each combination of a test-instance and browsers are ran in a closure, so that a specific server address can be used for that browser. How would you get that using the above api? Is opts here options for one browser or multiple browsers?

[kesla]

If opts here is for one browser, I'm having trouble understanding how one could write a simple example without a lot of setup-code.

[juliangruber]

opts has .baseConfig, .browsers and .local for example

[juliangruber]

instead of running the tests closure for every browser, the api from the comment above would behind the scenes fan out to all the browsers

so

test('foo', function*(t){
  // ...
})

really becomes

tests.push(function*(t){
  // ...
});

// after all the tests

each(browsers, function*(){
  runTests(tests)
})

[kesla]

Alright, but then I wouldn't be able to orchestrate the tests the same way as we do today in condor.

Today we're in a closure with the test-variable & browser-variable, so I'm setting up a server for each browser, a server that each browser instance then can use. Basically, I'm doing stuff in the browser where I'm then looking at events in the server. For example I'm clickin on a link, I then test that the server that the event was sent correctly.

Also, I think that there's a risk of confusing with your proposed api - we would have something looking very much like ordinary tap/gap but where the behaviour is quite different.

[juliangruber]

i'm probably not understanding probably...my understanding is that this would roughly expose the current module through my proposed api: https://gist.github.com/d680eef1a755ecbb23eb