I believe I had the problem in the past that the completion wouldn't
automatically select the first entry but simply display the menu. Thus
it would simply insert a new line when pressing return to accept it.
Unfortunately I didn't commit it back then and I can't reproduce it now.
But it doesn't not hurt so I commit it to be safe.
CoC disables backup files due to a problem with a typescript
language-server (it becoming slow when the backup file is written,
triggering a `structureChanged`). But:
1. I want backup files
2. I do not use that language server
3. If I would:
1. I could disable backup files in a filetype plugin
2. The linked issue is from 2019 - no idea if this is still
present today
4. I don't think this issue is even a thing with a `&backupdir` that
does not include relative directories (which is the case for me)
So remove the lines disabling backups and go back to vim's defaults:
`nobackup` and `writebackup`.
Link: https://github.com/neoclide/coc.nvim/issues/649
I act very rarely on the whole file but quite often on hunks, so it
makes sense to have the mappings for the latter shorter.
The new mappings I wanted to use for the whole file are conflicting with
`<leader>gf` (i.e. `:GFiles`), so I've disabled them for now.
Use git's preferred way of referencing commits (`--pretty=reference`)
instead of the kernel's (`--pretty='%h ("%s")'`). I can see that
specifying the date makes sense and that the quotes are at best
unnecessary in the parentheses and at worst confusing when the subject
contains quotes itself.
Link: https://git-scm.com/docs/SubmittingPatches#describe-changes
Commit 0369218114 ("vim:settings: Include `-` in `&iskeyword`") added
the dash to `&iskeyword` but I didn't give an explanation for the change
and don't remember what led me to it. Today I'm annoyed by it since it
breaks some stuff (e.g. previously the C pointer arrow and syntax
highlighting of negative numbers in Python). The C ftplugin is obsolete
now, so delete it.
This reverts commit 0369218114.
Instead of changing the functionality of `]]` and swapping `][` and
`]]` in one mapping, do it in two separate steps. This way both changes
can be used independently of each other.
Fixes: 459e0b7113 ("vim:keys:TODO: Modularize changes to ]]/][ maps")
When substituting in a visual block selection, the automatically
inserted range `'<,'>` is not enough as the pattern will also match
outside of the selection on the same lines. To only match inside the
selection, insert the `\%V` atom.
Theoretically the pattern will also need the atom before the last
character to make sure that everything is inside the selection, but this
can be inserted manually when needed (see `:h /\%V`).
These were commented out already for longer. For `(` I have delimitMate
and the `<CR>` mapping conflicts with the one from coc.nvim (which
includes the `<C-G>u`)
When switching to <Cmd> for mappings in fc9874bf97 ("vim:keys: Use
<Cmd> for plain command mappings") I forgot that the previously inserted
selection range marks (i.e. '<,'>) are missing now. This broke vmaps
that used :substitute without having added them manually.
Since <Cmd> has some performance benefits, I want to keep it and add the
missing marks where needed.
Use tmux popup windows for fzf in vim, so that it spans (almost) the
full tmux window. My use-case for this is, that sometimes I open fzf in
a split pane and notice then that the lines are too long for the width
of the pane. Since the fzf window won't resize if I resize the pane
(TODO: issue in fzf.vim), I have to first close it, resize and then
reopen fzf. I hope to fix this by using always the big tmux popup and
eliminating the need to resize at all. The issue remains when vim is in
a split window outside of tmux. I can imagine that this will become
annoying, but we'll see...
I sometimes work in files where I cannot remove trailing whitespaces (at
least not permanently). In these cases the background highlighting of
them is quite strong and can be a bit annoying. To make this more
pleasing, use the foreground to highlight trailing characters, as long
as tabs and trailing spaces are displayed as non-space characters.
For this I also set `trail` in `&listchars`, and while at it, merged the
two lines setting `&listchars` and cleaned up some comments.
Use the `s` flag, so that the position before the last jump is updated
automatically.
Use `<Cmd>` so that both normal and visual mode mappings become
identical and can be merged. This has the added benefit that it now also
maps in operator-pending mode.
Complete the TODOs.
Apparently on older neovim versions, passing the name of script-local
functions to `timer_start` is not enough (in v0.10.2 this works, but
v0.7.2 not). Fix this by wrapping it in `function(...)` to pass a
Funcref.
See :h timer_start:
> timer_start({time}, {callback} [, {options}])
> [...]
> {callback} is the function to call. It can be the name of a function
> or a |Funcref|. [...]
Because of the deferring of the highlight, the old selection highlight
was deleted only when a new one was created. Introduce a second timer
that makes sure that the old highlight is cleared earlier.
Sometimes `_highlight_cword` is called without the existence of
w:cword_timer_id but with a w:cword_match_id. No idea why.
I thought that it might be a race condition, but does not seem like it.
(Also not sure if this could even run in parallel)
TODO: Investigate how _highlight_cword can be called without
w:cword_timer_id
Defer the clearing of the old highlight as well. This means that the old
highlight will remain a bit longer but prevents the highlight from
flickering because of the continuous `matchdelete`.
`highlight_selection()` is supposed to abort if the visual mode was
started on an empty line. This was done correctly for charwise-visual
but failed to do so in linewise-visual.
- Combine one `exists` call with an implicit `get` into a `get` with an
empty default
- Do not stop any timer in the `Highlight*` functions as this done
already by `ClearHighlights`
- Get rid of the check for existence of `w:*_timer_id` since I believe
that there is no way of it being unset in the current state. The
`_Highlight*` functions are only called via the timer so there should
always be the variable. I could also not reproduce the description in
the comment - this might have been true in a previous revision.
- Make sure that the visual selection itself is not matched. This
actually increased the performance a lot while also looking like
normally (there were some subtle differences between the highlight by
visual mode and `CursorColumn`)
