in parser, allow spec to to read child info about prosemirror-markdown

[marijnh]

The approach to tables that uses a width attribute is likely to be deprecated in the future, in favor of using a plugin with an appendTransaction function to ensure rectangular shape instead of schema-level restrictions (which makes things like colspan and rowspan possible), and such an approach wouldn't have this problem. Maybe you can build your tables like that?

[mhuebert]

Hmm. Markdown tables don't support colspan, so maybe it makes sense to enforce it at the schema level, but I haven't written any plugins yet so I will have to learn more about transactions to know how that would work. Do you know offhand any plugins that have a similar appendTransaction function that would be worth look at?

[marijnh]

The new-style table implementation hasn't been open-sourced yet, but in effect, what it does is scan the document for tables, and for those that aren't rectangular, append a transaction that inserts cells to make them rectangular.

[mhuebert]

OK. Does that mean on every transaction it is scanning the whole doc to find and validate all the tables? (or is there a way to only scan things that were edited in the last transaction?)

[marijnh]

You can write a helper (but it's not part of the library yet) that recursively descends two documents (the one before the transaction and the one after) and skips nodes that are the same in both, in order to make scanning huge document for this kind of consistency issues cheap.

[mhuebert]

OK, that makes sense.

I do like the behavior that comes with schema constraints; when deleting the contents of a cell, the cell remains in place, and the cursor jumps to the previous cell. Without schema constraints, the cell itself is deleted, meaning the neighbouring cell gets shifted to a different column. Example: https://www.youtube.com/watch?v=JBVVCt_C7tw

[marijnh]

Deleting the whole cell when you delete its contents shouldn't happen. How exactly are you deleting it?

I agree that the schema-constraint formulation of table shape is attractive, but it's a pain to parse cleanly (as you found), doesn't generalize well at all, and basically feels like a hack, so I'm not happy with it.

[mhuebert]

I was deleting the cell contents via backspace. However, I just realized the isolating key was only added in 0.21.0, and I hadn't seen the update - now the behavior is great.

Even without a plugin / appendTransaction function, I think this should be ok for now - I don't see any obvious ways that a user will end up with an imbalanced table.

Table row-count was only issue so far where I needed lower-level access in the parser so I'll close this now.