Efficiency enhancement after node profiler investigation

[ajesshope]

Search Terms

getRelativeUrl, enhancement

Problem

Not a problem... an efficiency enhancement. We ran typedoc thru the nodeJs profiler to investigate why running typedoc against our system was taking so long.

Top 6 of the node profiler output:

   ticks  total  nonlib   name
  95136    4.5%   23.0%  JS: *renderElement ~/<pkgName>/node_modules/typedoc/dist/lib/utils/jsx.js:79:23
  55925    2.7%   13.5%  JS: *getRelativeUrl ~/<pkgName>/node_modules/typedoc/dist/lib/output/components.js:62:19
  43660    2.1%   10.6%  JS: *join node:path:1170:7
  36254    1.7%    8.8%  JS: *links ~/<pkgName>/node_modules/typedoc/dist/lib/output/themes/default/partials/navigation.js:93:19
  33670    1.6%    8.1%  JS: *renderChildren ~/<pkgName>/node_modules/typedoc/dist/lib/utils/jsx.js:130:28
  26826    1.3%    6.5%  RegExp: [\s\S]*?(?:[^_-][_-](?=[^_-])|[^A-Z](?=[A-Z][^A-Z]))

There were 2 significant areas for us:

1. At 23% was renderElement in lib/utils/jsx.js. We know the large number of interfaces in one d.ts was the cause of this and have opened an enhancement for it. See open issue #2382
2. At 13.5% was getRelativeUrl in lib/output/component.js. This method converts an absolute path to a relative path and is the target of this enhancement request below.

Suggested Solution

The enhancement we suggest is to use a Map to hold the results of previous conversions and return the held result if the key matches.

    /**
     * Map to hold previous absolute path to relative path conversions
     */
    protected absoluteToRelativePathMap = new Map<string, string>();

    /**
     * Transform the given absolute path into a relative path.
     *
     * @param absolute  The absolute path to transform.
     * @returns A path relative to the document currently processed.
     */
    public getRelativeUrl(absolute: string): string {
        if (this.urlPrefix.test(absolute)) {
            return absolute;
        } else {
            const key = `${this.location}:${absolute}`;
            let path = this.absoluteToRelativePathMap.get(key);
            if (path) return path;
            path = Path.posix.relative(this.location, absolute) || ".";
            this.absoluteToRelativePathMap.set(key, path);
            return path;
        }
    }

After we implemented this locally the getRelativeUrl went from 13.5%(2nd highest @55,925 ticks) of the total time to %0(99th highest @ 25 ticks).
For our project typedoc calls getRelativeUrl a total of 87,530,384 times (yup... that is 87.5 million)... 149,731 of those are unique... which means there are 87,380,653(99.83%) duplicate calls... we currently have 9,342 pages.
Even though a Map get/set are not noops... they use significantly less processing than Path.posix.relative... therefore we make huge savings (~13.5%) on processing with this small change.
Obviously with small projects this is probably not even noticeable. However, the larger the project the bigger the impact.

Latest profiling top 6 (using change above).

   ticks  total  nonlib   name
  91250    4.3%   32.3%  JS: *renderElement ~/<pkgName>/node_modules/typedoc/dist/lib/utils/jsx.js:79:45
  33172    1.6%   11.7%  JS: *renderChildren ~/<pkgName>/node_modules/typedoc/dist/lib/utils/jsx.js:139:28
  28007    1.3%    9.9%  JS: *links ~/<pkgName>/node_modules/typedoc/dist/lib/output/themes/default/partials/navigation.js:105:19
  27063    1.3%    9.6%  RegExp: [\s\S]*?(?:[^_-][_-](?=[^_-])|[^A-Z](?=[A-Z][^A-Z]))
  14535    0.7%    5.1%  JS: *DefaultThemeRenderContext.urlTo ~/<pkgName>/node_modules/typedoc/dist/lib/output/themes/default/DefaultThemeRenderContext.js:49:22
  10872    0.5%    3.8%  RegExp: [&<>'"]

[Gerrit0]

PR welcome! Basically all of my test cases are tiny

[Gerrit0]

27063 1.3% 9.6% RegExp: [\s\S]*?(?:[^_-]_-|^A-Z)

I just noticed this... wat, that makes me want to kill wbr. Not sure it provides any real value to begin with, it's a carryover from back when the themes were in handlebars.

I'm working (hopefully tomorrow) on having navigation load dynamically, which should drastically reduce the number of calls to getRelativeUrl