Non-simple tables with rowspans cannot contain links

[cwickham]

I have:

• searched the issue tracker for similar issues
• installed the latest version of Quarto CLI
• formatted my issue following the Bug Reports guide

Bug description

(EDIT: @cscheid edited this to reduce confusion)

The minimal repro is here:

---
format: html
title: Hello
---

+---------------+------------+
| Col1          | Col2       |
+===============+============+
| **Span rows** | Row1       |
|               +------------+
|               | Row2       |
+---------------+------------+

+---------------+----------------------+
| Col1          | Col2                 |
+===============+======================+
| **Span rows** | <http://example.com> |
|               +----------------------+
|               | Row2                 |
+---------------+----------------------+

The happens without list-table processing.

[mcanouil]

The issue is not specific to list-table, the same issue occurs for grid-table.
The rowspan attribute is properly kept when using Pandoc alone, so it must be something in Quarto.

• Quarto is dropping table row attributes when cells contain link #13795

I tried disabling HTML processing and related options but it does not change anything.

[cscheid]

quarto dev-call show-ast-trace points the blame squarely at src/resources/filters/quarto-pre/resolvescopedelements.lua and, indeed, if I comment out this filter, the bug goes away.

There's something more-or-less fundamentally wrong with this filter, because if it changes the table, it will always break non-simple tables (it goes through a to_simple_table/from_simple_table roundtrip).

The peculiar thing is that this is happening only in one of the tables.

I think the technical term for what's happening through this entire code is, um, "FUBAR".

1. the assignment of table colwidths will necessarily break grid tables, because of the to_simple_table/from_simple_table round trip. This is an unsalvageable feature of the current implementation.
2. the determination of whether table colwidths need to be computed depends on the presence of Link elements in the table, for no documented reason. But if tblColwidths is "auto" and there's a link in the table, then we return an array of zero widths throughout, which triggers the bug.

I, um, have no earthly idea what is going on in here.

[cscheid]

Unfortunately this is all code from really early on: 401b22f, and there's no comments.

[cscheid]

The only narrow fix I can think of is to skip colwidth processing for non-simple tables. It's an obnoxious fix because Pandoc filters don't allow you to catch Cell nodes natively (where the rowspan attributes live).