fix: use byte col to call extmarks api

[mrbeardad]

Neovim use 3 kinds of column:

1. Byte column, 0-based index, this is used for most of nvim.api, include extmarks api
2. Character column, 1-based index, this is used for cursor left/right move, and is used in this plugin
3. Virtual column, 1-based index, this is used for cursor up/down move, curswant is a virtual column but not a character column

For example,

1234567
abc你好

你 has 3 bytes, 1 chracter and 2 display width. So

• its byte column is 3 (start byte index)
• its character column is 4
• its virtual column is 5 (end virtual column)

In this pr, I've convert all vc.col into byte column when call extmarks api. However, there are too many vc.curswant = vc.col, so I make the pr to discuss the solution for it.

[brenton-leighton]

There's another function, [getcurpos()](https://neovim.io/doc/user/builtin.html#getcurpos()), that gives the byte index for the cursor. Would it solve the problem to use that instead (as well as cursor() instead of setcursorcharpos())? I used getcursorcharpos() because I thought it would handle extended characters, but maybe I have it backwards.

[mrbeardad]

No. The problem about column for extmarks api has been solved with this pr, the remaining problem is, curswant is a virtual column, it is relative with character display width, and affect up/down motion. All vc.curswant = vc.col is wrong, it needs to be converted first.

local text = vim.fn.strcharpart(vim.fn.getline(lnum), 0, col)
local virtcol = vim.fn.strdisplaywidth(text)

[mrbeardad]

local text = vim.fn.strcharpart(vim.fn.getline(lnum), 0, col)
local virtcol = vim.fn.strdisplaywidth(text)

However, the behavior is slightly different from the builtin motion, becase when cursor at a character that has 2-char display width, the curswant can be either the first char or the second char, but the code above always set curswant to the second-char position

12
你

The curswant of cursor that on 你 is 1 if it has moved down from 1, or 2 from 2. so always recover vc.curswant from vc.col is not a good idea

[brenton-leighton]

Does this branch fix the position issues? I've changed all the position functions to use the byte count versions. I think there's still a problem with the move right command, because this is implemented manually. But otherwise, is it just a problem with the extmarks?

[mrbeardad]

The display width needs to be considered when set end_col, check this

[mrbeardad]

Left/Right motion use character column, up/down motion use virtual column(display width). Why don't use normal command to move cursor? Thus, user setting could be considered automatically, such as whichwrap,

[brenton-leighton]

Left/Right motion use character column, up/down motion use virtual column(display width).

What do you mean by virtual column? The curswant value?

Why don't use normal command to move cursor?

Originally I tried using virtualedit = "onemore" for insert mode but it didn't seem to work. I thought it was because setting the option wasn't "api-fast" (so it would occur in the event loop like feedkeys). I think I realise now that the issue is that the $ command doesn't work in insert mode even with virtualedit="onemore.

Thus, user setting could be considered automatically, such as whichwrap,

I'll try changing things to use "onemore" and whichwrap

[mrbeardad]

What do you mean by virtual column? The curswant value?

Virtual column is the last screen position occupied by the character at that position, different unicode character maybe has different display width. For example, the display width of a is 1, and 你 is 2, so a occupies 1 screen column, 你 occupies 2 screen column. The virtual column of a is 1 and 你 is 5.

1234567
abc你好

Assuming the cursor is at 好, the curswant of the cursor is 6, the virtual column of 好 is 7. Then you moved the cursor up and it was placed on the 6 where the virtual colmun equal to the curswant.

[:h virtcol()](https://neovim.io/doc/user/builtin.html#virtcol())