undo region, multibuffer/nrrwrgn (many buffers stitched together as one virtual buffer)

[orenbenkiki]

Problem

The ability to select a region of the file (probably via visual selection would be most popular), and then undo only operations that affected that region.

That is, make it possible to do some changes in some file region, move to another part of the file, make other changes, and then undo just the changes in the first part.

A classical use case is adding debug. prints in different areas of a file. Currently, one has to manually revert any debugging changes made in each region once the problem is determined to be somewhere else.

I realize this is "out there" but I find myself missing this feature frequently enough to at least ask.

Expected behavior

Having some variant of u and/or U which would apply only to the currently selected region. The key binding is the least of the problem here. Undo would be linear (reverse order of changes) for all operations that affected the region. For operations that also had impact outside the region, it is a question whether to include them or not; I tend towards "yes" but that might need to be configured or controlled somehow.

[justinmk]

Would be cool but will be hard to implement. Not sure if the use case is strong enough.

A more tractable and flexible model, could be to design a way that multiple buffers can be stitched together (is that the idea of ctrlspace @Konfekt ?), so you can scroll and search a single "multi-buffer", and each buffer in the "quilt" is just a normal buffer with its own undo, etc.

NrrwRgn ("narrow region") is a related concept: https://github.com/chrisbra/NrrwRgn

[orenbenkiki]

Narrowing would be easier to implement (still not easy, I suspect) but I think this would be less flexible as one would have to plan in advance the set of regions for the region-specific undo, instead of "just working". That said, I fully appreciate region-specific undo would be a very non-trivial feature to implement.

Whether the use case is string enough is a good question. I find myself missing this on a monthly basis, but that's just me. One could claim this would make other, more complex forms of "undo tree" less necessary - and a general "undo tree" is even worse to implement in a way humans can actually use it...

[Konfekt]

Would be cool but will be hard to implement. Not sure if the use case is strong enough.

Persistent undo was a student's coding project, so that would be another useful one

is that the idea of ctrlspace

I depend on it foremost because tabs containerize buffers and these tabs containers can be stored in sessions, called workspaces.

NrrwRgn is what comes closest as far as I know, good call. I have it mapped to gz, but not thought of this use case.

[justinmk]

Plugin: https://github.com/SunnyTamang/select-undo.nvim looks pretty minimal.
@SunnyTamang I think this would be worth including in core, just need to decide the right default mapping.

[SunnyTamang]

So current default mapping is 'gu' and also has one experimental mapping as 'gcu' (for character wise undo, which is not being used).
So what's the plan?