gfm parsing oddity with links in link text

[TripleCamera]

Explain the problem.

In gfm, links in link text should not be parsed.

Input:

[https://bilibili.com/](https://bilibili.com/)

Actual output:

<p>[<a
href="https://bilibili.com/](https://bilibili.com/)">https://bilibili.com/](https://bilibili.com/)</a></p>

Expected output:

<p><a href="https://bilibili.com/">https://bilibili.com/</a></p>

Try pandoc!

The bug is in the autolink_bare_uris extension:

C:\Users\EricQiu>pandoc -f gfm-autolink_bare_uris
[https://bilibili.com/](https://bilibili.com/)
^Z
<p><a href="https://bilibili.com/">https://bilibili.com/</a></p>

C:\Users\EricQiu>pandoc -f gfm
[https://bilibili.com/](https://bilibili.com/)
^Z
<p>[<a
href="https://bilibili.com/](https://bilibili.com/)">https://bilibili.com/](https://bilibili.com/)</a></p>

Pandoc version?

pandoc 3.3 (the latest version)

[jgm]

Test of gfm behavior:

[[hello](url)](there)

[hello](there)

Here the inner link takes precedence. So, with autolink_bare_uris, it's just the same. The inner link (now an automatically created one) takes precedence. It's a bit hard to see how to avoid this with our current modular structure, where the core (which handles regular links) doesn't know about the autolink_bare_uris extension. I suppose we could do something ad hoc, which is probably what GFM does...

[jgm]

There are two separate issues here:

1. Ideally, the URL wouldn't be autolinked when it occurs inside the link label, since this leads to bad HTML (nested <a> inside <a>).
2. The entire rest of the link is gobbled up as part of the URL, with the result that the outer link isn't recognized as a link at all.

With 7950d58 I have fixed (2). (1) remains, but it isn't as serious as problem; I believe browsers will still display this as a single link.

It is difficult to see how to fix (1), given the architecture of this project. We would need to disable the autolink parser when parsing a link label. But we don't actually know if it's a link label (as opposed to, say, an image label or a span) until after it has been parsed (when we get to the (...) part). We could, perhaps, retokenize and reparse the label in this case, with something in state that the autolink parser can look at to tell itself to do nothing.