From https://blog.chmouel.com/posts/emacs-isearch/:
- Bind `avy-isearch` to directly jump to currently visible matchings
- Use `consult-line` instead of `swiper` (this also removes a package dependency)
This avoids setting those variables in Emacs 29.x, which is still supported by this
configuration (it won't hurt much to set those variables in this case as well, though).
To this end, do not use :custom in use-package declaration, because
according to the documentation[1], `:custom` declarations can conflict
with settings done via `customize-option`. However, values set directly
in the configuration files should always be allowed to be overwritten by
custom settings done by the user – at least that's the general approach
of this Emacs configuration.
[1]: https://www.gnu.org/software/emacs/manual/html_mono/use-package.html#User-options
Give buffers running ephemeral `eshell` or `shell` sessions a name that
is different from the default name chosen by these commands, and only
restrict those ephemeral shell buffers to be displayed at the side.
This way, running the commands `eshell` and `shell` will yield windows
can be handled like any others and can be used for long-running
sessions.
This is mostly copied from the current implementation in `bookmark.el`,
adapted to non-filename bookmarks.
Note that we are now using the `location` slot in bookmarks to store the
URLs, and relocating bookmarks will set the `filename` slot to `nil`.
I keep a tree of Git repositories organized as submodules in my home
directories and want each of those repositories to be considered a
separated project. The repositories that are indeed a proper part of a
superproject I can ignore.
Shrink fist column, because my repositories are usually not that long.
Version information is also less important than upstream divergence to
me, so let's move this behind.
I haven't really used `company` in the recent past, so removing this
package does not hurt much. Maybe `corfu` will be more practical for
me, as it also only relies on builtin in-buffer completion and does not
require additional magic. However, I usually do not use popup
completion dialogs much, because I find them obstructive.
`projectile` is nice, but I rarely use a fraction of its functionality
and not having to use an external package for project management feels
more robust.
Seeing active minor modes directly might be informative. Try to reduce
the list of shown minor modes to only interesting ones using the (still
existing) diminish configuration. Remove now obsolete configuration of
the minions package.
`globel-company-mode` has not been used since quite some time, and
completion in the individual programming modes should be handeled by
LSP. Disabling `company-mode` also allows to use standard completion
features, which is nice.
This is to ease the configuration, as eglot is built in. lsp-mode might
provide a better programming experience, but I do not do much Python
programming nowadays.
Just learned about `use-short-answers` (from a comment [here][]).
However, when long answers are expected for confirmation, it's often an
indication that the question at hand might be serious enough to take the
time to type these out (as recommended by the docstring of
`use-short-answers`).
[here]: http://irreal.org/blog/?p=12595
Pandoc supports grid tables, which are supported in Emacs via `table.el`.
By default, the alignment markers for those tables (:) are not
recognized in table.el, so let's add those. Furthermore, editing grid
tables in `markdown-mode` does not work, since `jit-lock-mode`
overwrites the `keymap` text property that `table.el` relies on to
enable table editing. Using `edit-indirect` to edit those tables in an
indirect buffer where everything works nicely. Indeed, editing
`table.el` tables in Org also uses an indirect buffer, for the exact
same reason.
Whenever I try to check what I actually in the past couple of days, I
try to use a clocktable for this. However, those clocktables are
usually littered with small entries, i.e. entries where I only spent a
couple of minutes on (example: a quick chat with a colleague). Those
entries clutter the view for the relevant entries, making it hard to
make sense of past activities.
Restricting the clocktable to a certain depth somehow helps, but major
single tasks are hidden this way, and I usually want to see those.
The custom formatter added in this commit tries to alleviate this
situation by providing a mechanism to filter out those entries using a
minimum clocktime threshold. Let's see how this feature plays out.
Previously, links which end in a closing brackets got a non-breaking
space appended via a separate advice to `org-link-make-string`.
However, this did not work well with the other advice that removes
statistics cookies from links, mostly because I messed up the order in
which the advices were applied. To remedy this, the advice to remove
statistics cookies now also adds a non-breaking spaces as described
above.
When a headline ends on a closing bracket, Org adds an escape character
to the link text to distinguish the end of the link from the link
description. This escape character is a zero-width space, which is
counted for Org table alignment as one character, but the link itself is
displayed shorter, because the zero-width character is displayed as a
single pixel by emacs.
To work around this issue until the upstream fix is released, let's add
a final non-breaking space to those link descriptions to avoid the need
for the zero-width escape characters.
I found that opening this output on the right side to be annoying, so
let's switch try the bottom side window instead. This also works with
having an eshell buffer open at the same time, resulting in a nice reuse
of the right space of eshell windows that is usually empty.
With vertico, using the default `yank-pop` is nice enough. Furthermore,
yanking with helm does not update the current candidate selection when
in a minibuffer prompt (like for `find-file` or `org-insert-link`),
sometimes leading to confusion.
With `common-lisp-indent-function`, the indentation of `while` was not
correct – oops.
This change will break some current indentations, though. Stay tuned.
This is to have this extra blank line after refiling.
This reintroduces 76c8717, and reverts f064bf9 and 8ed64b7. The issue
with too many blank lines at some items may reappear.
I access bookmarks far more often than the list of local important
files, so let's skip the usual `C-o` hazzle and have bookmarks right
under point when available.
This is more reasonable than inserting a link to the current location of
point, which can be quite arbitrary (at least in my workflow). When a
link to the item at point is needed, it should be inserted manually.
This change is experimental and might be reverted.
Idea: we don't want to see those items until the deadline warning date
is due anyway. Drawback: those items are not shown even when jumping to
the specific date (or past it), so planning ahread get's a bit more
complicated.
Those cookies are correct when the link is inserted, but subsequently
gets updated (usually to `[0/0]`) when the statistics cookie is updated
in the Org item where the link is placed — removing any meaning from
this cookie. It's thus better to not have it at all, I think.
`org-capture-fill-template` unconditionally adds a final newline to each
template. This caused extra empty lines with the old templates when
nothing is inserted at point (`%?`), because in this case empty lines
before and after point were present, with nothing in between.
Those extra empty lines are gone now, and point is positioned after the
final non-blank character. When extra empty lines are needed, they have
to be inserted manually.
This is to better be able to identify those empty lines and keep the
notes tidy.
The new value for org-cycle-separator-lines` is the current default
value.
Inspired by the `use-package` documentation, but the other places where
`add-hook` is called this simplification does not seem to be appropriate
from my point of view (missing `-hook` suffix, better readability when
calling `add-hook` directly).
This seems to be slow, causing proced to take more time updating a
buffer than waiting the one second until the next update – resulting in
an unresponsive proced buffer.
Note: I am not sure whether this setting is correct, as `(not
on-windows)` will not be evaluated when `use-package` calls
`custom-theme-set-variables`. However, I think it will be evaluated
eventually, and that should be all that's needed.
It's nice to be able to collapse plain lists easily, as I have some
quite large ones. The comment advises to be careful with certain
situations yields strange results, though, which is why this setting is
deemed experimental for the time being.
Using orderless by default get's in the way of “usual” code completion
and shell completion, where prefix completion is the expected
behavior (for me). So let's try to use basic completion first, and try
to rely on orderless completion to kick in as soon as spaces are
involved (which should not happen for shell and code completion).
The completion colors should also help to see whether we are still using
basic completion or orderless completion. Let's try this for now.
Since `counsel-switch-buffer` has been replaced by the plain
`switch-to-buffer`, recently used files are not shown there anymore, so
let's make `recentf` easier to access.
NB: `consult` also provides a function akin to `counsel-switch-buffer`,
so maybe this keybinding will not be that relevant anymore in the
future.
Using ivy for completing in regions somehow has problems in eshell, so
let's not do it. The main use-case for using ivy in region completion
is for ledger, so maybe it can only be activated there?
This mostly reverts commit 615b285f74.
The default completion style is now to use orderless by default, but
certain categories can benefit from additional completions (e.g., file
and partial-completion).
It's quite hard to determine the appropriate completion categories,
though. Does Org have a dedicated one?
The reason is that while filtering Org headlines, sometimes one of the
other completion styles matches, hiding other possible matches that
would have been found by the `orderless` style. This can happen when
spaces are included in Org headlines (which is usually the case) and
those spaces match the search pattern. In this case, the position of
point is also relevant, which is not something I want.
It would be better to know which completion category is use in this case
and override the completion style for this only, but so far I haven't
found it. Continue looking …
This may change back in the future.
The ivy configuration is kept on purpose, in particular because some
counsel functions are still in use. Maybe those can be also be replaced
later on, maybe with helm functions? Not quite sure whether this is a
good idea, though.
