Improving performance by reducing useless parsing

[wooorm]

Subject of the feature

There are two main places where parsing is done that is (potentially) useless.

• content: At the end of a line in content, we parse ahead, to figure out if the paragraph should be closed
  This is double, because when the paragraph is closed, we will actually do the parsing.
  This was done in remark too. And similar to there, it is a bit optimized.
  This point in parsing markdown is rather complex, because of interplay with definitions, setext headings, paragraphs, but also lazy lines.
  Removing lookaheadConstruct improves performance by 13%. The alternative should be possible and hopefully is not too big.
• document: to figure out whether containers continue, close flow, start new flow, or have lazy lines, another throwaway inspection is done.
  Removing document completely improves performance by 28% (although lists are complex so it some time spent there is unavoidable) (SOLVED IN 939e90d)

[fazouane-marouane]

It could be helpful to have rough idea of bottlenecks. I tried chrome tools for nodejs but it doesn't provide useful reports for micromark. Maybe there's a flag to include or maybe it works better for browser code.

[wooorm]

I’ve essentially run micromark on 1mb of markdown, and then toggled a bunch of code on and off to figure out where potential bottle necks are. Hence the 13% and 28% mentioned above.
I did see that V8 in Chrome performs fast, and in Node it’s a bit slower. I’ve looked a bit at Chrome’s profiler, but I’m not too great at it 😅

[fazouane-marouane]

Oh sorry I overlooked the performance speedups. 13% and 28% look good indeed !

[wooorm]

this repo might be useful! https://github.com/wooorm/a-dump-of-markdown.

And for these two numbers: well, block quotes and lists do need to get parsed. So 28% improvement won’t be reached. But, 1/3rd of the time is spent with ’em.

[wooorm]

The document part is now fixed!