Rustdoc generates illegal "<div>" inside "<pre>" with traits

[kentfredric]

Similar to the previous bugs I've filed in this regard with illegal html:

In: 1.39.0-nightly (34e82a7 2019-09-10)

The following trait definition:

#[cfg_attr(
    feature = "external_doc",
    doc(include = "CacheFieldCollection.md")
)]
#[cfg_attr(
    not(feature = "external_doc"),
    doc = "An abstact representation of 'a thing' that can expose data \
           about its unit."
)]
pub trait CacheFieldCollection {
    /// Returns a struct containing data recorded for this thing
    fn fields(&self) -> &CacheFieldData;
    /// Returns a timestamp indicating time of last modification/update
    fn mtime(&self) -> &chrono::DateTime<Utc>;
}

Emits invalid html ( hand formatted ):

<div class="docblock type-decl hidden-by-usual-hider">
<pre class='rust trait'>pub trait CacheFieldCollection {
    fn <a href='#tymethod.fields' class='fnname'>fields</a>(&amp;self) -&gt; &amp;<a class="struct" href="../ccache_stats_reader/struct.CacheFieldData.html" title="struct ccache_stats_reader::CacheFieldData">CacheFieldData</a>;
<div class='item-spacer'></div>    fn <a href='#tymethod.mtime' class='fnname'>mtime</a>(&amp;self) -&gt; &amp;<a class="struct" href="https://docs.rs/chrono/latest/chrono/datetime/struct.DateTime.html" title="struct chrono::datetime::DateTime">DateTime</a>&lt;<a class="struct" href="https://docs.rs/chrono/latest/chrono/offset/utc/struct.Utc.html" title="struct chrono::offset::utc::Utc">Utc</a>&gt;;
}</pre></div>

"pre", like headers, only permits child nodes that are "phrasing content", of which, div is excluded.

So the error is from <div class='item-spacer'></div>

htmltidy handles this by prematurely ending the <pre> before the <div>, and then reinserting a <pre> afterwards, breaking layout:

[hirschenberger]

So, I assume a <span class="item-spacer"></span> would be legal as it is phrasing content.

I can change that?

[nox]

Why is there even an element to do spacing, of all things?

[hirschenberger]

I think it makes sense as the non spacing list looks more confusing.

[nox]

That’s not the issue here. The issue is that empty DOM elements aren’t how you are supposed to do spacing. That’s a CSS concern. If you have 1000 methods to space, that’s 999 spacing elements, those things add up and then the page is really hard on browsers.

[kentfredric]

Its also rather redundant in the context of a <pre> where all you need to add a line feed is ...:

 print!("\n")

Or equivalent.

I mean, you're already doing horizontal alignment with literal whitespace.

[hirschenberger]

Its also rather redundant in the context of a <pre> where all you need to add a line feed is ...:

 print!("\n")

Or equivalent.

I mean, you're already doing horizontal alignment with literal whitespace.

Yes, adding an extra '\n' instead if the spacer-div adds a slightly bigger vertical space, but looks ok.

[nox]

If you really want markup, then IMO each trait item should be in its own li tag. A trait contents are a list of trait items, after all.