[css-break-3] Should broken boxes always be extended to fill the gap?

[Psychpsyo]

CSS Break 3 says, in 5.3. Splitting Boxes, that when a box gets split, it's block size is effectively increased. This leads to its backgrounds and borders being painted into the remaining empty space in that fragmentainer.

This usually seems desirable, but it gets ugly when a table cell with the rowspan attribute gets broken due to the rows that it spans falling into different fragmentainers. In this case, the empty space that comes to exist between those rows has nothing in it, apart from the rowspan-cell, which I'd expect to only render next to the rows that it spans.

I think there should either be an exception for these cells, or a CSS property that can control this behavior, similar to box-decoration-break. Alternatively, box-decoration-break could be extended to accept something like [ slice | clone ] || skip-gap instead of just slice | clone. This new behavior should, in my opinion, be applied to rowspan cells by default, so this could be added to the UA stylesheet:

td[rowspan], th[rowspan] {
    box-decoration-break: skip-gap;
}

The UA stylesheet addition would be a breaking change, though currently only Safari manages to properly break these cells at all. Both Chrome and Firefox fail to lay them out correctly in different ways. (Chrome doesn't extend the height on the rowspanned cell so it doesn't reach the end of the second row and Firefox draws the background only inside the first fragmentainer and only next to the first row.)

TL;DR:

[fantasai]

I think what you're saying is that if we're breaking between rows, then we should actually break the (visual) box of a rowspanned cell at that point, too (if it's breakable at that point). This makes sense to me. Probably the same should apply for Grid. I don't think we're at the point where we can require this behavior, but we can suggest it.

[Psychpsyo]

I think what you're saying is that if we're breaking between rows, then we should actually break the (visual) box of a rowspanned cell at that point, too (if it's breakable at that point). This makes sense to me. Probably the same should apply for Grid. I don't think we're at the point where we can require this behavior, but we can suggest it.

Yes, pretty much. Though I think "breaking the (visual) box" of something is not yet a concept that exists in the spec.
5.3. says that any break causes a box to grow taller, to fill the rest of its fragmentainer. So just forcing a regular break into rowspanned cells would still result in the "Current" case from my drawing above. The sort of break inserted here would need to not stretch the box out to fill the rest of the fragmentainer.

Also, if box-decoration-break is set to clone, the break would need to be inserted above the end of the row, to leave space for the bottom padding and margins and such.

I've written up some spec text that I hope does a good job at defining behavior that can accomplish this: #12805