eat too much cpu on 3.0.5

[fengmk2]

Parse blew markdown, 3.0.5 use 0m20.459s, and 3.0.4 use 0m2.981s

[@addyosmani](https://github.com/addyosmani),
[@borismus](https://github.com/borismus),
[@carsonmcdonald](https://github.com/carsonmcdonald),
[@chriseppstein](https://github.com/chriseppstein),
[@danwrong](https://github.com/danwrong),
[@davidmaxwaterman](https://github.com/davidmaxwaterman),
[@desandro](https://github.com/desandro),
[@hemanth](https://github.com/hemanth),
[@isaacs](https://github.com/isaacs),
[@josh](https://github.com/josh),
[@jrburke](https://github.com/jrburke),
[@marcelombc](https://github.com/marcelombc),
[@marcooliveira](https://github.com/marcooliveira),
[@mklabs](https://github.com/mklabs),
[@MrDHat](https://github.com/MrDHat),
[@necolas](https://github.com/necolas),
[@paulirish](https://github.com/paulirish),
[@richo](https://github.com/richo),
[@rvagg](https://github.com/rvagg),
[@sindresorhus](https://github.com/sindresorhus),
[@SlexAxton](https://github.com/SlexAxton),
[@sstephenson](https://github.com/sstephenson),
[@svnlto](https://github.com/svnlto),
[@tomdale](https://github.com/tomdale),
[@uzquiano](https://github.com/uzquiano),

repeat codes:

var MarkdownIt = require('markdown-it');

    var md = new MarkdownIt({
      html: true,
      linkify: true,
    });
    var readme = fs.readFileSync(path.join(fixtures, 'eat-cpu.md'), 'utf8');
    md.render(readme);

[fengmk2]

/cc @puzrin this is rush!

[puzrin]

Problem is in this line change 8ca0b5b#diff-f4f9b75e4799aa5566d488b8f480de40L37

Change looks correct, but i don't see any broken tests on revert.

/cc @rlidwka , take a look please. I don't see check for ]:. Also, we should check what happens with recursion on real references, because list can be long, and that should not fall under our nesting limit..

[puzrin]

https://github.com/markdown-it/markdown-it/blob/master/lib/parser_block.js#L18

[reference] seems to be a mistake. After playing with official dingus & digging spec, i didn't found a way to do such interrupt. This rule cause heavy recursion, trashing everything.

@fengmk2 if you can't wait at all - remove nested array from that line. Will publish fix a bit later, when @rlidwka appears online.

[rlidwka]

Fixed in refs branch.

It was an optimization attempt gone south. If we have a long list of references (like 533b5c8), this parser would call getLines for each reference until the end of the paragraph. So first call would be getLines(1, 50), second call would be getLines(2, 50), etc. So I thought, it would be nice to optimize it a bit. It didn't go so well.

Anyway, the immediate solution would be to remove references from there.

And we have a benchmark that performs 3x times slower than CommonMark (in theory, their parser is linear on that input, and ours isn't), but it might be left for future discussions.

[fengmk2]

I use 3.0.4 right now.

[puzrin]

Fixed in published 3.0.6

[fengmk2]

cool! Thanks @puzrin @rlidwka