minify_html = true causes Major page formatting issues to pages with numbered code blocks.

[Jieiku]

Bug Report

minify_html = true causes Major page formatting issues to pages with numbered code blocks.

This was first reported to the abridge theme: Jieiku/abridge#84

I implemented this in the after-dark theme and seen the same effect (pay particular attention to the footer at the bottom of the pages)

with minify_html = false it looks normal:

Environment

Zola version: 0.16.0

I also tested 0.15.3 with the same result.

Expected Behavior

The page should look the same with or without minify_html = true

Current Behavior

No Errors, just errors in page formatting when using minify_html = true

Step to reproduce

git clone https://github.com/Jieiku/zola-minify-test
cd zola-minify-test
zola serve

open this page: http://127.0.0.1:1111/overview-code-blocks/

Notice the page section headings and page footer are all broken following the numbered code blocks.

To further confirm the issue you can removed the line numbered code blocks, then minify_html = true will no longer cause an issue.

[Keats]

The non minified HTML output looks sane, it might be a bug in https://github.com/wilsonzlin/minify-html that should be reported there

[Jieiku]

Thanks, will open an issue report now.

[skewballfox]

The non minified HTML output looks sane, it might be a bug in https://github.com/wilsonzlin/minify-html that should be reported there

It turns out the generated html was attempting to attempting to close nonexistent spans

[Keats]

Hm weird I didn't notice those invalid span 🤔

[jhpratt]

So it turns out this is going to be a fair bit more complicated than I had hoped.

<span> tags for marking the code as a given language, inside a function, etc. are opened in the first row of the table and not closed until the last row of the table. This is not allowed, as tags have to be closed in the reverse order they are opened. The only solution that would match the intended behavior is to close all tags for each row and re-open them at the start of the next row. Unfortunately this requires knowing which tags they were, which is not something that appears to be readily available.

[Keats]

Can someone submit a failing test so we have that at least?

[Jieiku]

I already did in the opening post....

git clone https://github.com/Jieiku/zola-minify-test
cd zola-minify-test
zola serve

I can try and reduce the size of the test to something more basic though.... (let me know what your looking for)

[Keats]

Woops sorry I missed the link to the repo in the initial post when I commented.

[skewballfox]

hey, was poking around the codebase. Where is the code which is causing the bug, or the likely source of the bug.

Also what's the components dir separate from src?

[Jieiku]

I'de say find the section that deals with numbered code blocks, but I do not know where that is, it sounds like @jhpratt might have investigated it at one point, before it ended up looking like a difficult fix, so maybe they know.

[jhpratt]

I don't recall offhand what file or chunk of code it was that was relevant. My previous comment effectively sums up my recollection. It's not an easy task, unfortunately.

[Keats]

It will somewhere in https://github.com/getzola/zola/tree/master/components/markdown/src/codeblock

Also what's the components dir separate from src?

This is a Cargo workspace, Zola is split in a few sub crates to speed up compilation.

[TheOnlyMrCat]

The issue to me seems to be that syntect's HTML ouptut is very inflexible.

The most flexible way (and the method we're currently using) is with line_tokens_to_passed_spans, which requires us to store a &mut ScopeStack parameter for it. We can interact with and inspect the Scopes in the scope stack, but we're missing the scope_to_classes function (used internally by line_tokens_to_passed_spans) that would properly allow us to close-then-reopen the spans.

It is possible to copy the implementation of scope_to_classes from syntect directly into zola's own code, but that's not necessarily good for maintainability.

Incidentally, this does not seem to be an issue for the InlineHighlighter since styled_line_to_highlighted_html always closes its spans.