The present configuration is supposed to start the server if it's not already
running. Previously we checked this using `server-running-p', but this is not
really realiable. Instead, we now checking the `server-process' variable
Furthermore, if it turns out during startup that the currently configured server
file is already present, we warn the user about this and don't do anything
else. We let the user to fix it manually because it's (i) easy for the
user (easier than doing it automatically) and (ii) only done once, namly during
startup (the burden on the user is thus tenable).
The current implementation may not be accurate, though, as my understanding of
the implementation around Emacs' server functionality is only at the beginning.
Previously, we only copied the last element in the subtree, assuming that this
encompasses all of the content of the subtree. However, this is not true, and
thus we have to do something more elaborate. Now, starting from the end of the
subtree, we go up all elements in the subtree until we reach either the headline
or a drawer. Everything in between is copied as template to the current
This is some relic from former configurations, and might have been wrong quite
some time now. However, in Emacs 27+, package initialization is done before
loading user-init-file, and thus we don't have to do it by hand. Before that,
package initialization was done after reading user-init-file, but before calling
after-init-hook, and since we needed to load some packages in the init file, we
had to initialize package.el ourselves.
Usually, only one of them is used. Maybe one day I have to replace the two
files (or, more precisely, the custom variables pointing to them) by a single
one. But then, having two files, and also two shortcuts, also remindes me of
whether I am at home or at work, and that's quite significant, isn't it?
Configuring custom key bindings via use-package's :map keyword does seem to
install autoloads for the bound functions into the main helm package. I.e.,
when binding `db/play-radio-stations' to # in `helm-command-map' via :map,
use-package seems to install an autoload for `db/play-radio-stations' that
requires `helm', instead of the correct `db-music' package. Additionally,
defining key bindings somewhere in the init file is hard to manage, and they are
thus now collected with other key bindings in `db/run-init'.
`helm-mode' should actually not be activated, since we are still using ivy for
`completing-read' and friends. That being said, when we want to enable
`helm-mode' in the future, we should also not call `ivy-mode' anymore.
Autoloading helm does not work well with custom keybindings. Binding our
default "C-c h" to either `helm-command-map' or `helm-command-prefix' gives
errors, as both are not commands. In the previous configuration, the prefix
"C-c h" was initially undefined and only defined when helm was loaded. This led
to irritating behavior.
All this can be fixed by eagerly loading helm. This may slow down startup, in
particular on Windows, but it should be worth it.
It was actually only used for playing EMMS streams, but since the implementation
has been rewritten in EMMS, helm-emms does not work anymore. Replaced the radio
playing functionality by a custom function, obsoleting helm-emms.
If `db/mail-accounts' specifies accounts multiple times, or a definition for
some of these accounts is already present in `db/other-gnus-accounts', accounts
will be added multiple times with the current implementation. Apparently, Gnus
does not seem to care, but it would be better to have this fixed properly.
It's a bit clearer now what the function is doing when ARG is given, I hope. It
also turns out that switching to the current working directory does not make
much sense when we are in the shell buffer, because CWD is then just the, well,
current directory. The original logic used to CWD of the previous buffer (by
closing the shell buffer and immediately reopening it), but that's actually not
what the function is supposed to be doing, is it?