Codecov Report

Merging #3841 into master will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #3841   +/-   ##
=======================================
  Coverage   97.06%   97.06%           
=======================================
  Files         184      184           
  Lines        6503     6518   +15     
  Branches     1884     1887    +3     
=======================================
+ Hits         6312     6327   +15     
  Misses        101      101           
  Partials       90       90           

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c57aede...4e6946e. Read the comment docs.

Hmm, additional miss line is for scenario where file added and immediatly deleted. Good point, codecov!

Pretty strange what "buggy" scenario is covered by tests, hmm...

All done!

Just noticed there are issues on Windows and MacOS. Will need to take a look before this can be released.

I think, it's needed to improve tests. Currently timeout-based approach is really non determenistic - failed build prove it.

Can we turn buildDelay to accept a function, which managed when re-run build? I think it both make tests determenistic and can be useful for scenarios where watcher is runned by external toold.

{
  watcher: {
    buildDelay: (build: () => void): Disposer => {
      let t = setTimeout(build, 1000)
      return () => clearTimeout(t)
    }
  }
}

That is by itself an interesting idea. Ideally, plugins should be able to hold off a rebuild until they say: Ok, now I am done, please rebuild.

But the error I am seeing is not from non-determinism, it is very deterministically broken for me, and I do not understand the error yet.

In short, it looks like the test runner is telling me that "done" was called multiple times, which is particularly odd as we are not using a "done" function.

Uh-oh, cannot help, 'cause don't have nor win nor mac. But I'll think about plugins build delay.

Maybe if we allow plugins to return a Promise from the watchChange hook to block rebuilds?

So in the end, I did not really understand the issue with the test runner, but it only seemed to occur when there was an actual error, so I rather fixed the issues. So what I found:

• Creating a file on Mac and immediately afterwards starting a watcher can trigger an "update" event for that file on that watcher. Waiting a few ms between creating and starting the watcher seems to help
• On Windows, deleting a file and recreating it causes an additional update event to be created. BEFORE the actual create event. This might really deserve some more looking into it, but for now, I decided to reorder the events in the test. I hate this.

Maybe if we allow plugins to return a Promise from the watchChange hook to block rebuilds?

Good idea! I'll make PR soon, also fix #3789 and buildDelay