rustdoc unexpectedly formatted leading numeral as a list

[davepacheco]

I have a feeling I'm doing something wrong here so I apologize in advance if this is just noise. But the behavior I'm seeing seems surprising.

I tried this code:

//! 1. One
//!    1. Two
//!
//! This record means: user 123 has the "admin" role on the "project" with id
//! 234.  (Note.)

and I built this with cargo doc.

I expected to see this sentence together in one paragraph: "This record means: user 123 has the "admin" role on the "project" with id 234. (Note.)"

Instead, this happened:

Looking at the HTML, it's interpreted the leading "234." as a numbered list. Thinking this was some Markdown error, I tried it in the dingus but it does what I expected.

Meta

$ cargo version
cargo 1.56.0 (4ed5d137b 2021-10-04)

$ rustdoc --version --verbose
rustdoc 1.56.1 (59eed8a2a 2021-11-01)
binary: rustdoc
commit-hash: 59eed8a2aac0230a8b53e89d4e99d55912ba6b35
commit-date: 2021-11-01
host: x86_64-apple-darwin
release: 1.56.1
LLVM version: 13.0.0

On cargo 1.59.0-nightly (a359ce160 2021-12-14), I see almost the same behavior, but the "2" in 234 gets eaten:

[camelid]

I imagine this is probably an upstream bug with pulldown-cmark. cc @marcusklaas

[camelid]

I wonder if the leading 2 is actually in the HTML but is being cut off by some CSS rules. Can you post the relevant snippet of HTML for both rustdoc versions?

[davepacheco]

With cargo 1.56.0 (4ed5d137b 2021-10-04):

<div class="docblock"><ol>
<li>One
<ol>
<li>Two</li>
</ol>
</li>
</ol>
<p>This record means: user 123 has the “admin” role on the “project” with id</p>
<ol start="234">
<li>(Note.)</li>
</ol>
</div>

With cargo 1.59.0-nightly (a359ce160 2021-12-14):

<div class="docblock"><ol>
<li>One
<ol>
<li>Two</li>
</ol>
</li>
</ol>
<p>This record means: user 123 has the “admin” role on the “project” with id</p>
<ol start="234">
<li>(Note.)</li>
</ol>
</div>

In case it's useful, this is very easily reproduced -- e.g., with https://github.com/davepacheco/rustdoc-test.

[camelid]

Thanks! So, indeed, it seems the 2 is being hidden by the CSS; it is actually emitted in the HTML.

[camelid]

Assigning priority as discussed in the prioritization working group.

@rustbot label: +P-medium -I-prioritize

[marcusklaas]

Ooh, this is an interesting one. While this is certainly surprising, I'm not 100% confident this is a bug in pulldown. I played around in dingus, the CommonMark reference implementation, and found that it also produced a list for this snippet:

a
1.  d

However, if the final line is changed to 2. d instead, it does not produce a list.

I will have to check the spec, because I can't remember there being different rules for lists that do not start at one.

[marcusklaas]

It turns out there is a special rule for starting lists with the number one. From the spec:

In order to solve of unwanted lists in paragraphs with hard-wrapped numerals, we allow only lists starting with 1 to interrupt paragraphs.

And then there follow two tests that test exactly this distinction. Pulldown passes those tests, so I'm confused why it also produces a list for

1. a
    1. a

a
2. a

I will create an issue on the pulldown repo and investigate further...

[marcusklaas]

We have found the underlying issue and created a PR to address it here. Thanks to @camelid for flagging it with us!

[camelid]

Thanks for fixing it!

[camelid]

@marcusklaas what is the timeline for the next pulldown release?

[marcusklaas]

Very soon!

[marcusklaas]

Pulldown version 0.9.0 has just been released!

[camelid]

Thanks!

[camelid]

This will be fixed once #92252 has landed, which will happen shortly. Closing this issue.

EDIT: The PR has landed now.