Validating html-style table data with XSLT - xslt

Question

I need to be able to check xml with html-style table data to ensure that it's "rectangular". For example this is rectangular (2x2)
<table>
<tr>
<td>Foo</td>
<td>Bar</td>
</tr>
<tr>
<td>Baz</td>
<td>Qux</td>
</tr>
</table>
This is not
<table>
<tr>
<td>Foo</td>
<td>Bar</td>
</tr>
<tr>
<td>Baz</td>
</tr>
</table>
This is complicated by row and column spans and the fact that I need to accept two styles of markup, either where spanned cells are included as empty td or where span cells are omitted.
<!-- good (3x2), spanned cells included -->
<table>
<tr>
<td colspan="2">Foo</td>
<td/>
<td rowspan="2">Bar</td>
</tr>
<tr>
<td>Baz</td>
<td>Qux</td>
<td/>
</tr>
</table>
<!-- also good (3x2), spanned cells omitted -->
<table>
<tr>
<td colspan="2">Foo</td>
<td rowspan="2">Bar</td>
</tr>
<tr>
<td>Baz</td>
<td>Qux</td>
</tr>
</table>
Here are a bunch of examples of bad tables where it's ambiguous how to deal with them
<!-- bad, looks like spanned cells are included but more cells in row 1 than 2 -->
<table>
<tr>
<td colspan="2">Foo</td>
<td/>
<td rowspan="2">Bar</td>
<td>BAD</td>
</tr>
<tr>
<td>Baz</td>
<td>Qux</td>
<td/>
</tr>
</table>
<!-- bad, looks like spanned cells are omitted but more cells in row 1 than 2 -->
<table>
<tr>
<td colspan="2">Foo</td>
<td rowspan="2">Bar</td>
<td>BAD</td>
</tr>
<tr>
<td>Baz</td>
<td>Qux</td>
</tr>
</table>
<!-- bad, can't tell if spanned cells are included or omitted -->
<table>
<tr>
<td colspan="2">Foo</td>
<td rowspan="2">Bar</td>
</tr>
<tr>
<td>Baz</td>
<td>Qux</td>
<td/>
</tr>
</table>
<!-- bad, looks like spanned cells are omitted but a non-emtpy cell is overspanned -->
<table>
<tr>
<td colspan="2">Foo</td>
<td rowspan="2">Bar</td>
</tr>
<tr>
<td>Baz</td>
<td>Qux</td>
<td>BAD</td>
</tr>
</table>
I already have a working XSLT 2.0 solution for this problem that involves normalizing the data to the "spanned cells included" style then validating, however, my solution is cumbersome and starts to perform poorly for tables with an area of greater than 1000 cells. My normalization and validation routines involve iterating sequentially over the cells and passing along a param of cells that should be created by spans and inserting them when I pass their coordinates in the table. I'm not happy with either of them.
I'm looking for suggestions about cleverer ways in which to achieve this validation that hopefully would have better performance profiles on large tables. I need to account for th and td but omitted th from the examples for sake of simplicity, they can be included or ignored in any answers. I'm not checking to see if thead, tbody, and/or tfoot have the same width, this can also be included or omitted. I'm currently using XSLT 2.0 but I'd be interested in 3.0 solutions if they were significantly better than a solution implemented in 2.0.

Answer 1

I don't think this kind of problem is suited for XSLT - especially if you have to process very large tables.
I'd suggest to develop a solution using a procedural languge - maybe using XSLT to pre- or post- process the XML.