Consider prohibiting name from contents for rows in static (non-grid) tables

[joanmarie]

Take a look at https://bugzilla.mozilla.org/show_bug.cgi?id=1568360, in particular the two comments after the opening report in which I paste the accessible name of a table row. That strikes me as in no way a good thing. How can we stop it? 😄

[jcsteh]

FWIW, Chrome seems to only calculate name from contents for rows (HTML tr or role="row") if the row is focusable. This seems like a nice pragmatic solution to me, though I realise the ARIA spec doesn't really have a concept of "name from: contents only if focusable". :)

[joanmarie]

But we do have rows which descend from grids and rows which descend from tables.

[mcking65]

I see this as related to issue #1008 where the only way to distinguish between rows, columnheaders, and rowheaders that should be selectable is by context. Similarly, Name should be calculated from content for rows only if they are focusable and in a grid or treegrid.

Historically, since the 1.0 ontology included grid but not table, we ended up with a problem for table because the 1.0 ontology used row for grid rows, columnheader for grid columnheaders and rowheader for grid rowheaders. This did not give us a way of distinguishing between interactive rows and static rows. In 1.0, we had a gridcell role, so we were able to add a cell role that represents the static cell and keep the gridcell role for interactive cells.

Consider nested lists and trees. We have 100% separation in the ontology. If you want an interactive hierarchical list, you use tree, with groups of treeitem elements nested in treeitem elements. And, if you want nested static lists, you use list elements nested in listitem elements. Since we do not have a concept of control patterns, this is probably how table and grid should be. But, given the way that things are named, we don't have a good way of doing that.

Thus, we use hacks that can be confusing and make the spec complex. For instance, we opted to make separator a widget only if it is focusable. We could do something similar for row, columnheader, and rowheader, making them widgets only if they are in the grid context.

[joanmarie]

Consensus from TPAC: Do this. May or may not be done for 1.2.