For some reason molten-nvim fails sometimes when lazy loading. This
simply turns lazy load off for the time being but I should instead
investigate a little more what is actually causing it.
Fixed issue when pynvim did not exist in the environment but we tried to
invoke a molten-nvim command. Now we first fire up python to check for
pynvim existence. This takes a moment so we do it asynchronously and
cache the result for the rest of the program runtime.
Also we only show the molten icon if we are actually connected to a
running kernel.
Switch out mason-tool-installer for mason-conform.nvim and
mason-nvim-lint as respective wrappers for automatically installing
formatters and linters.
Follows the same principle as mason-lspconfig.nvim (in fact, the repos
are mostly based on the same code) and apply the concept to the other
tools: Whatever is enabled in the respective plugins (lspconfig,
nvim-lint and conform.nvim) will automatically be installed by mason.
This is really neat and basically takes care of me ever having to
interact much with Mason itself or manually set up the tools to be
installed. All I have to make sure is that they're updated once in a
while.
Enabled snippet completion from the first letter (allowing single letter
snippet prefixes).
Extended the luasnip usage and added a telescope extension to show all
enabled snippets for the current buffer.
As a first step to disentangle the ide.lua module into completion,
formatting, lsp (and maybe lint?), this separates the completion module
into its own file.
Moved rainbow delimiters into dependencies so that it runs with TS
instead of loading TS on its own. Made TS itself more lazy (only loading
on VeryLazy by default).
Run TSUpdate on each run instead of only each build step.
With the new `img-clip.nvim` extension we can copy any image anywhere
and simply paste it into a markdown/quarto/latex/typst/... document with
the right markup already.
Those can be from the web/locally. Also allows drag and drop from e.g.
web pages.
Mapped to `<leader>pp` currently, though we will have to find a better
mapping. Or invoked with `:ImagePaste`.
We can now open any ipynb file as if it were a quarto document by
leveraging the `jupytext.nvim` extension which works in the background
to convert everything. Will only work if `jupytext` program is actually
installed on the machine.
Allows normal editing of the file and will write any changes back to the
original notebook on save and exit!
Added mapping to quickly edit any code cell in FeMaco, an extra buffer
which brings lsp functionality and similar with it. Can be used as a
fallback if otter.nvim is not working for the buffer, or when working on
a buffer which otter.nvim does not understand.
Set it up with extended options and keymaps. Will use an image provider
if it has been set externally (usually set by the 'image.nvim' setup
routine) or fall back to use the `wezterm.nvim` plugin to display it in
a wezterm pane. This display can be a little buggy so we should prefer
others.
Moved the mini bracketed comment following to use the ]k [k suffix. I am
not sure if this is completely taking since every now and again it seems
to still be overriding my own configured mapping.
Now we can either fetch the python venv basefolder
(`require('core.util').get_python_venv_basefolder()`) or specifically
the python binary contained within (`.get_python_bin()`).
Additionally, the venv result should be cached for the duration of using
nvim so that we only have to fetch it once, regardless of how many
modules require it.
Removes them from being active in vim as well as removing them from
being displayed in which-key.
Unfortunately the which-key implementation still seems broken, sometimes
removing them sometimes leaving them as-is.
Since we do not make use of my own zettelkasten plugin anymore, this
commit is a quick fix to reinstate the functionality of the `vn` alias
in the shell opening the note index.
An issue is that there is now no dedicated function to show or open the
index, so we have to make do with a hardcoded path. Perhaps at some
point I will get around to fixing that issue but until then I am fine
with it just working again.
Hopefully definitively disabled the search mode for the flash.nvim
plugin. Previously only disabled the key mapping but it would still run
every now and again.
While we can still reach vifm through the `<leader>E` mapping, the basic
`<leader>e` file editor mapping has been switched over to the mini.files
browser.
It works somewhat like oil.nvim, in that you simply get a buffer to work
with that you can delete/move/copy/rename lines in and after doing a
synchronization (with `=`) all changes are actually applied to the
files.
Since I mostly need the quick file opening to do some quick renaming
operation or want to open an adjacent file this seems like the quickest
and most painless option to do so. For larger operations I still have
access to the full vifm experience.
Mini.nvim now also comes with an align module which perfectly mimics
the behavior of my beloved junegunn vim-easy-align plugin. It is almost
sad to see such an old friend go, but ultimately it removes a redundant
plugin and switches the setup ever so slightly more towards being lua
based.
The mini module also allows more advanced lua-based modifier pattern
shenanigans but so far I don't need any of the advanced stuff: I just
need a quick way to align markdown tables and similar, and this one does
exactly the same as before, only with a couple nice additional options
and previews. It still uses the same key chord `ga<where>` but also
offers `gA<where>` which apparently comes with a nicer preview. My
fingers will probably stick to the former for the time being.
One thing that changes is that, by default, it will align *all*
separators not just the first instance on each line. To mimic the old
behavior you can, in alignment mode, hit `f` to enter a filter and enter
`n==2` to get the same result as default vim-easy-align (the first pair
of delimiters are aligned).
Also it comes with some nice extra functionality like trimming
whitespace by just hitting `t` when in alignment mode
Fixed for new invocation usage and removed deprecated commands.
Added telescope functionality (to push results into trouble) though I am
not quite happy with the close coupling yet. Have not found an easy way
to only have this mapping be created if trouble exists.
On machines that have deno installed, we use peek instead of
markdown-preview for html-based previews of md files. The preview is
more responsive and in a neater package, as well as just not relying on
any vim plugin stuff to the same degree.
We still fall back to the old markdown-preview if no deno executable is
available.
We also change the key maps slightly to prepare for future 'prose' or
'preview' based mappings: All mappings are registered under the
`<leader>p` layer, with md preview being `<leader>pp` and various
mindmap operations moved to `<leader>pm`.
This fixes the somewhat confusing issue of trying to use any of the
surround operations before entering insert mode at least once in a file
(so, probably as the first operation). Before, the plugin would just
silently fail.
Now it simply loads very lazily, which does not affect load times
strongly.
Switched pyright to basedpyright as it adds a couple noteworthy
rules and some functions that are otherwise exlusive to pylance.
Especially useful for me are semantic highlighting as well as
inlay hints (now that nvim supports it from 0.10 onwards).
LSP attachment would complain if it did not find an existing dict to
access in the `settings.python` section of the python lsp dictionaries.
This fixes that by creating an empty dict if nothing exists yet,
essentially functioning as null-check.
Since it in fact works wonderfully now, we re-enable FeMaco, allowing to
edit codeblocks in markdown-like environments (but really anywhere) by
simply invoking `<localleader>ce` ('codeblock-edit').
The mapping is only active in markdown and quarto files for the time
being, though more can probably be added.
The command itself works anywhere (`:FeMaco`), so can be used in racket
or norg or whereever.
flash.nvim was activating on searches which, while perhaps useful
for some people, was not for me. It stops the (incremental) search
as soon as no valid target exists which drops you back into normal
mode while still typing out the search and thus doing
who-knows-what.
The operator pending mode for the original jump mapping has been
removed to allow the 'surround' mappings from mini plugin to
always work correctly (`ysiw`, `csaw`, ...).
Since treesitter is deprecating modules we switch to the new fork of the
plugin which works without any use of treesitter modules.
A painless switch since I only use the default settings of the plugin
anyway.
Add 'glances.nvim' plugin which shows the LSP references, definitions and
implementations in a very nice floating preview window. (Previously we
used Telescope which is still called if the glances plugin does not
exist.)
Can preview, scroll in the floating window, go to the destination, open
the destination in vert/horiz splits.
If the cursor was on the very first letter of a word it would fix
the misspelled word instead - now it should stay and fix the
current word under cursor instead.
Flash.nvim provides a very tasty remote editing functionality which lightspeed
does not. Otherwise it behaves *mostly* the same. For now, I have the flash
search label functionality activated (default) but if it is annoying I will
turn it off (It injects labels into the normal neovim search. So if you search
a word and want to go directly there, you just press the label key as part
of the search and it jumps there. This might be problematic if I quickly type
something that does not exist but it picks up a non-existent letter as label
instead.)
Otherwise, the remote editing functionality is activated by
<operator>r<label><movement>, such as `yr<label>iw` to yank inside a word
somewhere else and stay at the current position. Similarly for example
`cr<label>$` to change from the label to the end of the line and then jump
back to the original position and so on.
Slightly adapt lualine to not have arrow section/component dividers,
but simple slanted lines. Gets rid of a tiny bit of noise while still
clearly subdividing the sections for me which I need.
Also, removed displaying the hostname. I know which host I am on
generally, and rarely make use of it so it's just unnecessary info
for me.
Since I would like to keep my plugin configurations as modular as possible
I think it is a good first step to move the mini configuration directly
into the core plugin file where mini is loaded. Since this is the plugin
spec I want to take to basically any nvim installation I have, having it
in a single file makes it much easier to be portable.
Toggleterm, as nice as it is, is not one of my core dependencies.
I do not 'need' toggleterm on every nvim installation I have. Instead,
it can go into ui-related plugins since that is what it does, extend
nvim's UI functionality with new terminal dropdowns/floats.
Removed vim-exchange which, while an amazing functionality, is also
exactly replicated in mini.nvim (along with even more operators).
Uses slightly different mapping - not `cx<movement>` anymore but
`gx<movement>`. But this actually makes sense and fits in well with
my other g-prefixed operators (`gc` for comment and `ga` for align).
It is also the prefix for the other additional operators supplied
by mini.operators: `gs` for sorting text, `gm` for multiplying,
`gr` for replacing, `g=` for evaluating.
Since we already have the mini library installed in our setup, we don't
need to make use of external base16 plugins. It provides the same exact
functionality, and seems slim and bug-free. Nothing changes for the
user, but we have 1 plugin less to take care of (and it was
mis-behaving in new versions anyway).
Also set lualine to be reloaded on theme switch so it takes on the
colorscheme as well.
Added mapping to insert a link to an existing Zettel with
`<localleader>ni` (note insert), either from normal mode which creates
the complete link, or from visual selection mode which surrounds the
current selection (as visible link text) with the link.
New zettel links still use my own implementation so I have full control
over their naming scheme.
When switching the current buffer to the Zettelkasten index page
(`<leader>ni`), we now also switch the working directory to the
corresponding notes directory.
Setting `vim.g.disable_autoformat` to true will disable automatic linting globally,
though I have not set up commands or mappings for manual linting. It is also
missing buffer-local linting which could be copied from buffer-local formatting.
Moved loading and setting up file watching to the loading of the plugin itself.
Will make it a little more self-contained and not crash or complain if we are
missing the plugin.
HACK
At the same time, reverted back to pinned base16 commit since the new one did
not work again. Gotta figure it out at some point but no time now.
Since I am already using fidget.nvim for lsp notifications, might as well use it
for other, more explicit ones. Removed nvim-notify in favor of this since I like
how unobtrusive the notifications are (and let's be honest, most you don't need
to pay attention to, instead they just take you out of flow).
Using mason-tool-installer we ensure everything is installed correctly.
Need to improve the collection of things to install. Currently we just
do everything in lsp configuration file, even the non-lsp things
(formatters/linters) which should be sourced where they belong not
in that file.
Moved the mapping to show lsp info window from `<localleader>li` to
`<leader>vs`.
Local leader +l mappings should be reserved for lsp functionality,
while we have a whole +v layer to grab (meta) information about our
vim installation. It fits in there much better.
In the process of moving away from null-ls, added formatting with the help
of conform.nvim. Brings one new command, :ConformInfo which can also be
reached via `<leader>vc`.
AutoFormat on saving remains disabled by default but can be enabled with
:FormatEnable (and disabled again with :FormatDisable) or quickly through
`<localleader>lL`.
Manual formatting works like before with `<localleader>ll`. Uses the
formatters set in the plugin (similar setup to null-ls before) but
automatically falls back to lsp formatters if it does not have its own
and lsp has one enabled.
Mappings preceded by <leader>s 'show' something so removed a lot of the 'toggle'
wording from their descriptions.
Subsumed the toggleterm toggles under this menu since they 'show' a term window
(lazygit or ipython).
Changed Aerial mappings to show navigator by default (`<leader>so`) and the
sidebar outline only on capital version (`<leader>sO`) since this mapping is
used less often.
Removed broken molten-image setup.
Image nvim works mostly well (slow on wezterm but that will always be the case
with kitty protocol for now as far as I know).
Would love to be able to toggle images on/off dynamically but I don't see a
way to accomplish that now. (or really, get to any option of the plugin).
Molten itself also works well - the output is displayed more nicely than for
the Magma plugins and everything continues working mostly well (or rather,
just as wonky as I had it set up on my older magma install :)
For now, the molten - image.nvim integration seems to not work at all -
it simply errors out when it would produce an image as output. No clue why
and it also complains about the wrong image provider (which I have taken from
the molten readme). No time to bugfix now but maybe at some point.
To do - find a much better way of installing the image.nvim required
luarock magick - done manually with hardcoded path in setup now
Also extended the old `py` alias to a full-blown script which will in
addition to detecting the python repl also find any running molten
session for the current directory (i.e. any running jupy kernel) and let
the user choose the right one if there is multiple. Will then default to
starting a kernel-aware repl environment (euporia or jupyter-console).
Added a very simple `-c` option which lets you choose python command to
run manually.
