Very long rendering time for `_a` and `*a`

[rekado]

The following two snippets take about 11 seconds to be rendered to HTML on my machine:

First:

_a _a _a _a _a _a _a _a _a _a _a _a _a _a aaaaa

Second:

*a *a *a *a *a *a *a *a *a *a *a *a *a *a aaaaa

I think a similar issue was solved by defining StarLine and UlLine in markdown_parser.leg.

[jgm]

Ouch, you're right. This is really a terrible problem, and StarLine and UlLine won't help here. The way I deal with this kind of issue in pandoc is by keeping track of nesting level in state. But I don't think we can do that in peg-markdown.

An even worse case, and harder to fix:

*a **a *a **a *a **a *a **a *a **a *a
**a *a **a aaaaa

Any ideas?

[rekado]

Well, the specs don't say anything about nesting asterisks / underlines. Thus, the parser may only have to change state once whenever it encounters *, _, **, or __ and switch back when it encounters the same character group again.

I know that with lex one can define states and have rules match only when the lexer is in a certain state. Could we do the same here?

[jgm]

I think the idea I implement in this patch is simpler than what was there before, and it passes all my tests for nested emphasis.