fix(dash): Fix performance regression

[avelad]

See: #4062 (comment)

[philippe-elsass-deltatre]

@avelad can I suggest another improvement in xml_utils.js?

The following is MUCH faster (and causes less GC) than Array.from(elem.childNodes).filter:

  /**
   * Finds child XML elements.
   * @param {!Node} elem The parent XML element.
   * @param {string} name The child XML element's tag name.
   * @return {!Array.<!Element>} The child XML elements.
   */
  static findChildren(elem, name) {
    const found = [];
    for (const child of elem.childNodes) {
      if (child instanceof Element && child.tagName == name) {
        found.push(child);
      }
    }
    return found;
  }


  /**
   * Finds namespace-qualified child XML elements.
   * @param {!Node} elem The parent XML element.
   * @param {string} ns The child XML element's namespace URI.
   * @param {string} name The child XML element's local name.
   * @return {!Array.<!Element>} The child XML elements.
   */
  static findChildrenNS(elem, ns, name) {
    const found = [];
    for (const child of elem.childNodes) {
      if (child instanceof Element && child.localName == name &&
          child.namespaceURI == ns) {
        found.push(child);
      }
    }
    return found;
  }

[avelad]

Done @philippe-elsass-deltatre

[philippe-elsass-deltatre]

Still too slow though - I'm investigating.

[avelad]

@philippe-elsass-deltatre do you have any update?

[philippe-elsass-deltatre]

I'm profiling Shaka 2.5 VS 3 (debug versions), and Shaka 2.5 is still literally twice as fast. Not sure why.

[joeyparrish]

For historical context, we used to have frequent issues with mistakes in code like this. It is easy to mix up i and j for example, and easy to overlook in code review. So Iterables was a way to avoid the need for those.

Clearly the performance penalty is too great, though. We should figure out what other things transpile into poorly-performing ES5 and eliminate them. We could even write a test that would check the binary for poorly-performing polyfills, to catch regressions.

[avelad]

From Monday I will continue working on other improvements, but I will create other PRs.

[avelad]

@joeyparrish can you cherry-pick it to 3.3? Thanks!