Budgets for DataTable performance (highlighting rows under cursor)

[bryophyta]

I've been adapting a version of the built-in DataTable widget for my own use, so that it will highlight a full row or column under the cursor, as opposed to just an individual cell.

It looks like this might have been planned functionality at one point? (i.e. when the CursorType type is defined) So I was thinking about trying to rework my version and open a PR to add this functionality into the core DataTable widget, if there's any interest in this feature.

However, my understanding is that the built-in DataTable widget is intended to render smoothly for pretty huge datasets out of the box, and I'm wondering if this is why only cell highlighting is currently enabled. That is, the simplest implementation of the row highlight functionality would just slightly tweak the _render_line logic and set repaint=True on the reactive cursor_cell property:

textual/src/textual/widgets/_data_table.py

Line 198 in 3f76131

cursor_cell: Reactive[Coord] = Reactive(Coord(0, 0), repaint=False)

While this works fine for small tables (which is my current use-case), I'm guessing it wouldn't be performant for larger tables as it will force a repaint of the whole thing each time. I think something more performant could be done by calculating visible regions for update and using the refresh(region) function to trigger a more targeted repaint. But there are more and less complicated ways of implementing this, so it would be good to understand what level of optimisation is important for this widget. I was wondering if Textual has any budgets for performance when rendering large DataTables, that I could take a look at?

Thanks!

[willmcgugan]

We do plan on having row and column cursors. And also per row / column styles.

refreshing regions will make it more performant, so we can repaint only the parts of the widget that change.

We would consider a PR, but bear in mind we have plans for this and will get to it within a few weeks. Let us know if you need any pointers!

[bryophyta]

thanks for the quick response! If it's something that you're likely to address soon then you'd probably spend more time reviewing my PR than implementing it yourselves, given that trying to refresh optimally will probably involve extending some fairly Textual-specific code. So I'm happy to hold off unless I find myself struck with an inspired solution!

That said, if you've got anything written up for how you'd plan to approach this in a way that fits the framework then I'd be very happy to have a go at implementing the plan. Otherwise, I look forward to seeing how you all implement it when you get the chance!