fix(#3085): compiled lib not finding MimeUtils

[bwallberg]

There's probably some closure config somewhere that would fix this but I don't know closure what so ever, this does however align more with how the rest of the code references imports.

[joeyparrish]

TL;DR: The fix looks good. I analyzed the apparent compiler bug, and we'll try to come up with a way to hunt for and prevent this issue in the future.

The old code should have been fine. To me, this looks like a compiler bug. But we can't directly fix the compiler, so workarounds are almost the best we can do.

I picked apart the compiled code for this section, and it looks like this with formatting:

// "k" is an iterator, "l" is the current position in the iterator
for(var l = k.next(); !l.done; h = { Yb: h.Yb, dc: h.dc }, l = k.next()) {
  h.dc = l.value;  // "stream"
  h.Yb = hd;  // MimeUtils??
  h.dc.trickModeVideo = e.streams.find(function(m) {
    // Called with "h", so "m" == "h" here.
    return function(n) {  // This is the "find" callback
      return m.Yb.a(m.dc.codecs) == m.Yb.a(n.codecs);
    }
  }(h));
}

If "hd" is MimeUtils, its methods have all been stripped as dead code:

function hd(){}

No static methods are ever assigned on "hd" in the compiled bundle. I expect that other static methods of MimeUtils have been inlined where they are used.

There may be other places where this compiler bug is affecting us. We will try to use eslint plugins or Closure compiler's conformance rules to catch other instances of this, since auditing the code base by hand is very time consuming and makes regressions too easy.

[joeyparrish]

With your fix, the m.Yb.a part is replaced by a single, non-namespaced function kd. (At least, in the build I just did. Your randomly-assigned names may vary.) This kd function is MimeUtils.getCodecBase, which has been hoisted out of the class it's in.

I've filed #3088 to track efforts to find more of these problems, if there are any.

[shaka-bot]

All tests passed!