Shortly after I started using Neovim, I learned about the concept of text objects. For example, when the cursor is inside a pair of [], we can use ci] to change text inside them. This is definitely one of the greatest moments on my road of learning Vim. However, I haven’t really thought about how does text object really work.

Since I use Markdown frequently, I often need to change URLs quickly. I thought it might be a good idea to define a text object for URL. Actually, there is some plugins for it. However, they all rely on another plugin. Besides, I also want to learn how to write my own text objects from scratch. So I read some documentation and online posts. Now, I am able to create my own text objects.

How to define new text objects

We can think of a text object as a piece of text. It is visually selected when we press the shortcut that defines it. When we want to define a new text objects, all we need to figure out is: how do we select it when we press the shortcut?

To select the desired text, it boils down to understanding two things:

  • The start and end position of this text object
  • Is it character-wise or line-wise or even block-wise (I haven’t found block-wise useful, though)?

Now, it is easy to define new text objects. All we do is to make sure that the desired text is selected when we press the shortcut. It does not really matter how do you achieve that with Vim script.

For example, if we want to define a text object for inner line (excluding blank space at head and tail of current line). Here is one such implementation:

xnoremap il ^og_
onoremap il :normal vil<CR>

In the above xnoremap, o is used to move cursor from on end of visual selection to another end, see :h v_o.

Text object for URL

To define a text object for URL, we need to first decide the complete URL text under cursor, find its start and end position in current line, then select the entire URL text in visual mode (char-wise not line-wise).

vnoremap <silent> iu :<C-U>call <SID>URLTextObj()<CR>
onoremap <silent> iu :<C-U>call <SID>URLTextObj()<CR>

function! s:URLTextObj() abort
  if match(&runtimepath, 'vim-highlighturl') != -1
    " Note that we use to get the URL pattern.
    let url_pattern = highlighturl#default_pattern()
    let url_pattern = expand('<cfile>')
    " Since expand('<cfile>') also works for normal words, we need to check if
    " this is really URL using heuristics, e.g., URL length.
    if len(url_pattern) <= 10

  " We need to find all possible URL on this line and their start, end idx.
  " Then find where current cursor is, and decide if cursor is on one of the
  " URLs.
  let line_text = getline('.')
  let url_infos = []

  let [_url, _idx_start, _idx_end] = matchstrpos(line_text, url_pattern)
  while _url !=# ''
    let url_infos += [[_url, _idx_start+1, _idx_end]]
    let [_url, _idx_start, _idx_end] = matchstrpos(line_text, url_pattern, _idx_end)

  " echo url_infos
  " If no URL is found, do nothing.
  if len(url_infos) == 0

  let [start_col, end_col] = [-1, -1]
  " If URL is found, find if cursor is on it.
  let [buf_num, cur_row, cur_col] = getcurpos()[0:2]
  for url_info in url_infos
    " echo url_info
    let [_url, _idx_start, _idx_end] = url_info
    if cur_col >= _idx_start && cur_col <= _idx_end
      let start_col = _idx_start
      let end_col = _idx_end

  " Cursor is not on a URL, do nothing.
  if start_col == -1

  " " now set the '< and '> mark
  call setpos("'<", [buf_num, cur_row, start_col, 0])
  call setpos("'>", [buf_num, cur_row, end_col, 0])
  normal! gv

In the above code, we first find the end and start index of all URLs in current line. We then check if current cursor position is on one of those URLs. If that is true, we then set the '< and '> mark to the start and end index of this URL. Finally, we use gv to select this URL (see :h gv) .

In the above code, we use the URL pattern from a plugin called vim-highlighturl if it is installed or use expand("<cfile>") to get the URL under cursor. Note that expand("<cfile>") also works for non-URL text, the text object iu will also work for non-URL text. It is better to use the URL pattern provided by vim-highlighturl.

There is another shorter, but probably hacky way to define the URL text object:

function! s:URLTextObj() abort
  " We need to find the start and end of URL, and select the entire URL string.
  " the following does not work well, since it also shows results even when
  " cursor is not on a valid URL.
  " let url = expand('<cfile>')
  " a better way to get url.
  let url = matchstr(getline('.'), highlighturl#default_pattern())
  normal! ^
  " We should escape special patterns in url, see,
  " see also
  let url = escape(url, '/')
  " I got the inspiration here:
  silent! execute "normal! /\\V" . url . "/s\<CR>v//e\<CR>"

Note that you are not required to define URL text object as iu, any valid LHS for mappings is acceptable. We use iu because it conforms to the naming style of native text objects.

A text object for Markdown code blocks

For Markdown code blocks, it is necessary to define both inner and around text objects. We will use ic and ac to denote inner codeblock and around codeblock, respectively.

Here is a sample implementation1:

vnoremap <silent> ic :<C-U>call <SID>MdCodeBlockTextObj('i')<CR>
onoremap <silent> ic :<C-U>call <SID>MdCodeBlockTextObj('i')<CR>

vnoremap <silent> ac :<C-U>call <SID>MdCodeBlockTextObj('a')<CR>
onoremap <silent> ac :<C-U>call <SID>MdCodeBlockTextObj('a')<CR>

function! s:MdCodeBlockTextObj(type) abort
  " the parameter type specify whether it is inner text objects or around
  " text objects.

  " Move the cursor to the end of line in case that cursor is on the opening
  " of a code block. Actually, there are still issues if the cursor is on the
  " closing of a code block. In this case, the start row of code blocks would
  " be wrong. Unless we can match code blocks, it is not easy to fix this.
  normal! $
  let start_row = searchpos('\s*```', 'bnW')[0]
  let end_row = searchpos('\s*```', 'nW')[0]

  let buf_num = bufnr()
  " For inner code blocks, remove the start and end line containing backticks.
  if a:type ==# 'i'
    let start_row += 1
    let end_row -= 1
  " echo a:type start_row end_row

  call setpos("'<", [buf_num, start_row, 1, 0])
  call setpos("'>", [buf_num, end_row, 1, 0])
  execute 'normal! `<V`>'

Since code blocks are line-wise inherently, we only need to decide its start and end row. With this info, we can set the '< and '> mark. Then we can easily select the desire text. This time, we use visual-line mode, since it makes more sense to select code blocks in visual-line mode.


In this post, I have introduced how does text object work and how to define your own text objects. The most important part is to know your text object well and find ways to visually select it when the corresponding shortcut is pressed.


  1. Other implementations can be found in this issue. ↩︎