As I mentioned in#25509 (comment) , I'm testing the currentvim.Pos andvim.Range implementation to find paint-points (ThePrimeagen/refactoring.nvim@fabfe7b is the commit where I refactored the code inrefactoring.nvim to use my bundled version of them). This PR includes all the problems I've found and my proposed solutions for them. Let me know if it would be better to split it into smaller PRs.

I also haven't included tests yet because I'm expecting discussion about how to better solve the issues I've found.

Problem:

There is novim.Pos norvim.Range interface for:

• mark-line indexing
• treesitter indexing
• api indexing
• vimscript indexing

Solution:

Add said interfaces. This PR still leaves outto_mark andto_vimscript.

Problem:

There is no builtin way to tell if avim.Pos is inside of avim.Range.

Solution:

Addvim.Range.has_pos.

Problem:

Not allto_xxx interfaces have wrapping objects liketo_lsp

Solution:

Return unwrapped values inallto_xxx interfaces in theto_xxx interfaces where it makes sense. Accept unwrapped values in "from" interfaces where it makes sense, this also has the benefit of creatingvim.Ranges values like

localbuf=vim.api.nvim_get_current_buf()localnode=vim.treesitter.get_node()localrange=vim.range.treesitter(buf,node:range())

Problem:

start andend positions have different semantics, so they can't be compared.vim.Range relies on comparing theend andstart of two ranges to decide which one is greater, which doesn't work as expected because of the different semantics.

For example, for the ranges

locala= {start= {row=0,col=22,  },end_= {row=0,col=24,  },}localb= {start= {row=0,col=17,  },end_= {row=0,col=22,  },}

in the following code

localfoo,bar="foo","bar"--               |---||-|--                 b  a

The rangeb is smaller than the rangea, but the current implementation comparesb._end (col = 22) anda.start (col = 22) and concludes that, sinceb.col is not smaller thana.col,b should be greater thana.

Solution:

• Add atype field tovim.Pos to distinguishend andstart positions and use it on__lt,__le and__eq to normalizeend andstart positions before comparing them. With the same example above,b._end would now be considered to havecol = 21 which is smaller thana.startscol = 22.
  • This would also avoid having separatedvim.Pos.treesitter_start/vim.Pos.treesitter_end functions (also applies forvim.Pos.mark andvim.Pos.api) and simply checking whether the position is astart or anend inside of each function.
• If this is deemed not desired, an alternative could be changing the semantics ofvim.Pos so that bothstart andend are inclusive.

Problem:

Currently,vim.Range represents a 0 based end-exclusive range, but the Neovim API uses ranges withstart ==end to represent certain operations (e.g. insert text in a position withnvim_buf_set_text).

Solution:

Special casevim.Range.api andvim.Range.to_api to avoid modifyingend (by not performing the usual translations betweenvim.Range andapi-indexing semantics) whenstart ==end to keep them in sync. For example

localbuf=vim.api.nvim_get_curent_buf()localrange=vim.range(0,0,0,0, {buf=buf})-- ...localsrow,scol,erow,ecol=range:to_api()vim.api.nvim_buf_set_text(buf,srow,scol,erow,ecol, {'this is inserted at the beggining of the buffer'})