Add new options Darwin options:
- `targets.darwin.defaults`
This adds options for configuring macOS through the `defaults(1)` system.
This option can be used to manipulate a vast majority of user settings for macOS
and its applications.
This is implemented using freeform modules and includes additional descriptions
and type information for some useful options.
- `targets.darwin.keybindings`
This adds options for configuring the default keybindings for macOS text fields.
- `targets.darwin.search`
This adds options for configuring the default search engine for macOS.
If this commit now it is possible to define a custom theme directly
using Nix, like this:
```nix
{
programs.rofi.theme = {
"*" = {
background-color = "#000000";
border-color = "FFFFFF";
width = 512;
};
listview = {
cycle = true;
};
};
}
```
And this will be converted to the proper rasi format to be used in
rofi.
Nowadays services.{redshift,gammastep} modules are really similar. They
should, since Gammastep is a fork of Redshift with the main objective is
to support Wayland.
So instead of trying to maintain two separate modules, this commit unify
the options in lib/options.nix file, making the implementation of the
module itself ends up being really simple (just calling the common
options with the necessary parameters to differentiate between them).
* rofi: migrate to rasi configuration format
The Xresources configuration format is deprecated in Rofi. For example,
using Rofi from unstable (1.6.1 as of now) you get the following
warnings when starting the application:
```
(process:9272): Rofi-WARNING **: 01:38:48.596: The old Xresources based configuration format is deprecated.
(process:9272): Rofi-WARNING **: 01:38:48.596: Please upgrade: rofi -upgrade-config.
``````
So this commit migrates it for its new configuration format, called rasi
instead.
This new implementation uses attrsets manipulation instead of using
strings, making the code clearer and also fixing some bugs found during
the way. To make sure everything is right, I also created some tests.
If someone wants to validate if the generated config is correct, just
run in terminal:
```
$ rofi -dump-config
```
And rofi will dump the current configuration file, including all
unsetted options.
* docs: document programs.rofi.extraConfig changes
* rofi: add thiagokokada as maintainer
* rofi: add toRasi function
Closes issue #1725.
This allows mpv module to be customized with support for more advanced
features than the `programs.mpv.scripts` current support. For example,
with this change now this is possible:
```nix
{
programs.mpv.package = (pkgs.wrapMpv (pkgs.mpv-unwrapped.override {
vapoursynthSupport = true;
}) {
extraMakeWrapperArgs = [
"--prefix" "LD_LIBRARY_PATH" ":" "${pkgs.vapoursynth-mvtools}/lib/vapoursynth"
];
});
}
```
Since `programs.mpv.package` doesn't necessary reflect the final
derivation anymore (see #1524), we introduce `programs.mpv.finalPackage`
that has the resulting derivation.
This includes 2 tests:
- One to check if everything is alright with mpv
- Other to validate our assertion that package and scripts can't be
passed both at the same time
* docs: document recent mpv module changes
* mpv: add thiagokokada as maintainer
We currently check `isPath` and `isString` on crxPath and version
respectively, which is
1. pointless because the module system already does such checks, and
2. wrong because isPath means path literal; a derivation therefore is
not a path.
mu-cfind is meant to search for contacts within your contacts database and the emails that you have sent/received. The use of the --personal flag in that command is meant to filter for only emails that use your email addresses (which are all the ones you specify with the ${myAddresses} variable. Disregard what I said in #1623 (comment).
--my-address=<my-email-address>
specifies that some e-mail addresses are 'my-address' (--my-address can be used multiple times).
This is used by mu cfind -- any e-mail address found in the address fields of a message which also
has <my-email-address> in one of its address fields is considered a personal e-mail address. This
allows you, for example, to filter out (mu cfind --personal) addresses which were merely seen in
mailing list messages.
To initialize the database with mu init, the ${myAddresses} is not required to be passed to successfully initialize the database, but it is heavily recommended to do so.
To see the difference, in a safe location, run mu init --maildir=<path>, then mu index. You'll notice that "personal addresses" returns <none>, although the database will still work. However, mu cfind --personal will fail (as the personal contacts don't exist). Then run mu init --maildir=<path> --my-address=<address>, then mu index. Then you'll be able to search for contacts using mu cfind --personal.
When setting `...sway.package = null`, the default bar configuration
would throw an error trying to use the bar from the null package.
Logic is added to use the bar from `pkgs.sway` instead of `cfg.package`
if it is null.
Fixes#1714
* neovim: write config in $XDG_CONFIG_HOME/init.vim
instead of wrapping the configuration, which has sideeffects
https://github.com/NixOS/nixpkgs/issues/55376
* fix: update test accordingly
I also made some modifications to the systemd service to match the [AUR version](https://aur.archlinux.org/cgit/aur.git/tree/goimapnotify@.service?h=goimapnotify) of `goimapnotify`. In particular, restarting is useful in case a network failure causes `imapnotify` to exit - that shouldn't mean that it stops trying when the network comes back up.
Previously, it was not possible to set an arbitrary tmux prefix since
CTRL was hardcoded in the module.
To avoid breaking existing configs, a new option was implemented that
conveniently uses the tmux terminology but defaults to null and does
not affect previous behavior when set to null.
The behavior for the shortcut option was not completely replicated,
i.e., it does not bind "b" to send-prefix but stick to the default of
the prefix binding sending prefix (C-b C-b instead of C-b b) and it
does not bind repetition of the prefix (C-b C-b) to `last-window`,
both of these bring the option closer to the default tmux
configuration.
Fixes#1237
Brave Browser is a chromium-based browser, too.
+ it use the same web store with Chromium and Google Chrome.
+ the machanism of installing extensions works, and it's verified on
my macOS box.
The current definition makes waybar wait for dbus.service, but that
never happens because dbus.service is started on demand by
dbus.socket.
Per systemd docs:
https://www.freedesktop.org/software/systemd/man/systemd.service.html#Implicit%20Dependencies
- Services with Type=dbus set automatically acquire dependencies of
type Requires= and After= on dbus.socket.
- Socket activated services are automatically ordered after their
activating .socket units via an automatic After= dependency.
Services also pull in all .socket units listed in Sockets= via
automatic Wants= and After= dependencies.
Removing Requisite/After makes the service properly start for me,
simply specifying Type=dbus is enough.
See #1370
- Change the `attrsOf unspecified` to `pkgs.formats.json`
- Add missing default modules
- Expand the `with lib` with every function used
- Add inline documentation about the generated warnings
Using the final package in the `onChange` block broke some use cases.
This restores the old behavior and instead solves the test
dependencies in a different way.
Fixes#1611
This reverts commit 7c3c64208e.
The `invocation` is an optional attribute, so it doesn't make sense to
use it as the key in an attribute set. See
https://dystroy.org/broot/documentation/configuration/#verb-definition-attributes
Actually, `invocation` should not be defined when one wants to rebind
a built-in verb to a different key.
Also added documentation for the `key` attribute.
There exist mpv configurations which cannot be expressed in
`programs.mpv.config` currently. For example, it is impossible to use
multiple 'profile' attributes. This commit changes the way config and
profiles are parsed, using `lib.generators.toKeyValue` and
`lib.generators.toINI`, to allow for these kinds of configurations
through the use of `listsAsDuplicateKeys`.
This causes list values to be emitted as a list of key-value pairs
instead of a single key-value pair where the value is space separated.
This is useful, e.g., for socket units that would like to specify more
than one `ListenStream=` address.
The `accounts.email.accounts.<name>.neomutt.extraConfig` option is
included twice in the resulting config file for the account. One time as
part of the `mraSection`, one time as part of `accountStr` (`accountStr`
includes the `mraSection`). This removes that duplication. I opted to
keep the one in `accounStr`, since `extraConfig` doesn't necessarily
have anything to do with the `mraSection`.
In feh you can bind multiple keys to the same action, but Home Manager
only let you set a single key to an action. You can cheat and pass a
string with space-separated keys, but with this change you can pass a
list for each action to bind multiple keys to it.
Also adds a couple of tests.
Fixes#1366
* neovim: allow setting init.vim config alongside plugins
* neovim: add test for neovim plugins
* neovim: make pluginWithConfigType a have type submodule
`pgrep -x somecommand` exits with a non-zero status if it finds no
process running with the given name. When using home-manager as a
NixOS module, on boot (when sway isn't running) this script would
fail and then fail the unit since it seems the onChange scripts
are running with the -e switch.
This change ensures we're always returning a 0 exit status where we
attempt to get the pid of sway - we're only interested in either the
pid or an empty string, the exit status isn't important.
Adds a pet module without sync support as it makes no sense when
configuration is managed with Home Manager and the config would be
unwritable for pet anyway.
PR #1045
* mbsync: option for configuring a channel
A channel is a relationship between 2 directories/boxes/mailboxes
between the local machine (slave) and the remote mail server (master).
Each channel must be given at least:
* an account-unique name
* a pattern for which mailboxes to sync from master
* a pattern for what directory where that mail ends up on the
slave
Additional options can be added later.
* mbsync: option for configuring a group
A group is a grouping of channels together, so that many channels with
very different names can be handled as a single entity.
Groups are unique in mbsync because they will shadow channels that
have the same name on the command-line.
* mbsync: create groups configuration attribute
This is the end of the configuration that the end-user will use.
They will specify an attribute set that contains the name for the
group, so they can say
`accounts.email.accounts.<aname>.groups.<gname>` to access the
configuration for the group with the name `<gname>`.
* mbsync: write function to generate group-channel blocks
This function takes in a set of groups, and their consituent
channels and writes the appropriate .mbsyncrc block. The block is as
shown below:
Group groupName1
Channel channelName1
Channel channelName2
Group groupName2
Channel channelName3
Each group must have a unique name, no matter which account it is
declared under. The same holds true for channels. However, if there is
a group that shares the same name as the channel, the channel will
effectively be "shadowed" by the group, and mbsync will default to
working with the group in that case.
* mbsync: write function to generate channel configuration blocks
This function takes in a set of groups, which includes their
consituent channels and writes the appropriate .mbsyncrc block for the
channel. The block that is generated is shown below:
Channel groupName1-channelName1
Master :<accountName>-remote:<master-pattern>
Slave :<accountName>-local:<slave-pattern>
Channel groupName2-channelName2
Master :<accountName>-remote:<master-pattern>
Slave :<accountName>-local:<slave-pattern>
Each group must have a unique name, no matter which account it is
declared under. The same holds true for channels.
Using channels with the patterns set up this way allows one to specify
which maildir directories are to be synchronized FROM the master TO
the slave. In addition, it allows for these maildirs to be remapped,
between the master server and the local slave.
This is critical, because Gmail has a strange way of storing its mail
that makes using mbsync, mu, and mu4e more difficult.
There are additional channel parameters that are already present in
this codebase from the previous use of group-channel configuration,
which will be reused.
* mbsync: set the submodule's names field according to parameter
This is the same method as is used in creating an email account, named
`<name>` under `accounts.email.accounts.<name>`. This allows the user
to specify groups and channels, in a list-like format, but still gets
the "namespacing" to more easily handle the options available in each
of these locations.
* mbsync: provide examples of master/slave patterns for channels
* mbsync: create nested-let function to generate channel pattern
This pattern is required to either NOT be present, which means the
master pattern is used to match, or it has a list of patterns to use
beneath the master maildir to match against.
This function checks to ensure that if patterns is not empty, ONLY
then is the `Pattern` keyword printed. Otherwise, there are many, many
problems.
If there IS a list of patterns, then we use proper escaping methods to
ensure that the exact string is constructed.
* mbsync: per-account groups can have additional patterns
Gave the
`accounts.email.accounts.<name>.mbsync.groups.<gname>.channel.<cname>`
set a `patterns` option, which will allow for greater customization
and filtering of the master maildir to sync to the slave maildir.
* mbsync: add extraConfig option for easier-to-format options
These are options that can be handled by the `genSection` function in
the `genAccountFunction`, so they are left to the user to decide.
Most of these are made on a global basis anyways.
* mbsync: remove unneeded extraConfig.channel
This was originally placed here, seemingly, just to get this module
working. However, this field is actually more confusing now that a
separate per-channel configuration option for extra configurations has
been made available.
* mbsync: correct and improve comment in masterPattern description
* mbsync: switch channel/group generation to new functions
Changing this out is what moves us from the old system to the new one.
Instead of having a single channel manage a whole mailbox, we can now
specify an attribute set of groups that should correspond to an email
account.
Each of these groups contains an attribute set of channels that make
it up, and are grouped together for synchronization. In addition, each
of these channels can have additional IMAP4 parameters attached to
them to further refine synchronization.
Lastly, each of the channels is grouped together under the Group
section, ensuring that the channels' mailboxes synchronize as they
have been specified.
* mbsync: only generate group/channel configuration if channels present
Typically, when a group is specified, channels will be specified as
well. However, if due to error or mistake, the user forgets to specify
ANY channels for a group, we should not generate that group's
information.
This means that no channels are specified (which maps the remote
master to local slave). In addition, the `Group <gName>` block (which
brings the separate channels together) is also not generated.
Another thing to consider is that a user might specify a group and a
channel, but perform no additional configuration of the channel.
In a configuration, this would be realized by
`accounts.email.accounts.<aName>.mbsync.groups.<gName>.channels.<cName>;`
This creates the channel with the name `<cName>` and the
`masterPattern`, `slavePattern`, and `patterns` fields use their defaults.
By definitions set within mbsync, these defaults actually specify that
the remote master's `INBOX` mail directory is synchronized to the
local slave's `INBOX` directory.
So, if there is a channel that has no fields specified, then we DO
want to generate its configuration. But if there is a group that has
no channels, then we do NOT generate it.
* mbsync: acc comment explaining why groups attr set is never empty
* Revert "mbsync: remove unneeded extraConfig.channel"
This reverts commit 941c4771ca.
To support backwards compatibility, I need to leave this field/option
in the module, even if it will likely be more confusing to do it this way.
* mbsync: channel compatibility with previous iteration of mbsync
The previous version of mbsync used a single channel for an entire
account. This leads to issues when trying to change the mailbox
hierarchy on the local machine. The problem with this is that some
email providers (Gmail, among others) use a slightly different maildir
hierarchy, where the standard mailboxes (Inbox, Drafts, Trash, etc.)
are stored inside another directory (`[Gmail]/` in the case of Gmail).
This new version allows the user to specify any number of groups with
any number of channels within to reorder their mail however they wish.
However, to maintain backwards compatibility, I moved the original
channel-generating code to a function that will run ONLY when
there are no groups specified for THIS account.
* Revert "mbsync: channel compatibility with previous iteration of mbsync"
This reverts commit b1a241ff9f.
This function is in the wrong location and this was wrongly committed.
* mbsync: function for backwards compatibility with previous mbsync
NOTE THAT THIS IS THE CORRECT COMMIT FOR THIS CHUNK OF CODE!!
The previous version of mbsync used a single channel for an entire
account. This leads to issues when trying to change the mailbox
hierarchy on the local machine. The problem with this is that some
email providers (Gmail, among others) use a slightly different maildir
hierarchy, where the standard mailboxes (Inbox, Drafts, Trash, etc.)
are stored inside another directory (`[Gmail]/` in the case of Gmail).
This new version allows the user to specify any number of groups with
any number of channels within to reorder their mail however they wish.
However, to maintain backwards compatibility, I moved the original
channel-generating code to a function that will run ONLY when
there are no groups specified for THIS account.
* mbsync: function to choose which style of group/channels to generate
This is a simple if-check. If the old style is used, then this
account's mbsync.groups attribute set is empty. If that is the case,
then the old-style single-channel per account is used.
If that is NOT the case, then the new style is used in preference of
the old. This means that ALL channel code that would be generated by
the old version is replaced by the new one.
* mbsync: switch per-account config generation to check channels
* mbsync: program-wide groups if no account-specific groups
At the end, we have to choose whether or not to generate the old style
of having program-wide groups to specify things, where the boxes on
the channel underneath the group specifies which mailboxes to sync.
Here, we only generate the old style of group IF there is ANY account
that does NOT have the new `accounts.mbsync.groups` defined. At that
point, it is up to the user to ensure that the accounts in
`programs.mbsync.groups.{}` align with the name chosen for the
account, as I have made no attempt to change this old code.
However, if ALL accounts have their `mbsync.groups` defined, even if
each of the groups has a single empty channel, it will generate the
groups in the new style.
* mbsync: ensure \n after hm-generated comment
This was a multi-part fix. First, the `# Generated by Home Manager.`
comment has been reworked to ensure that it will ALWAYS have a
newline, even if the program-wide extraConfiguration is empty.
Next, we switched to placing 2 newlines between every account, to
provide further visual distinction between each account, which can
have multiple channels and multiple groups defined at the same time.
Lastly, the groupsConfig was slightly reworked, so that both the old
and new version can be used, but the new one will take precedence.
Because of this, groupsConfig is now a list of strings, which will
have single newlines inserted between each element.
But if the old style is NOT used, then the groupsConfig list
contains one element, an empty string. A single element has nothing
added as a separator, and an empty string produces no output.
* mbsync: only generate new group/channels if channels present
Here, the problem was if the user created a group for an account, but
did not also include a set of channels. If no channels have been
specified, then the group should NOT have its group-channel mapping generated.
I also corrected and improved the comment regarding
`genGroupChannelString`'s function and intended behavior.
* mbsync: channel patterns generate their own newlines
This means that when a channel has extra `patterns` defined for it, it
will generate those, and a single newline will be appended to the end
of that newly constructed string.
The moving of the newline character is slightly important because
otherwise, every account would receive an extra newline after every
channel, leading to 2 newlines after every channel.
* mbsync: place newline between each channel in a group
* mbsync: ensure old group/channel has proper spacing
This ensures that if the old style of generating program-wide groups
that there is the proper spacing before the group and in between each
line within the group.
* mbsync: ensure no empty channels present
If the user specifies a group correctly, they must still specify an
attribute set of channels. However, if they do not, then we need to
ensure that a group with no channels does NOT have any channel
configurations generated for it.
If there is a channel string generated for a channel that is empty,
then the `mapAttrsToList` returns a singleton list that contains just
the empty string. Thus, we can filter out all those results, to ensure
that no empty channels are generated.
It is important to keep in mind the difference between an empty
channel and a channel that has received no configuration, but is
named.
* A named channel is technically configured to have a name.
While the `masterPattern`, `slavePattern`, and `patterns`
field have NOT been populated, mbsync assumes that if
master/slave-Pattern are empty that means match against
`INBOX`.
If `patterns` is empty, no patterns are printed.
* An empty channel set is a set that has no channels within
it, but `mbsync.groups.<gName>.channels` is defined.
* mbsync: filter empty groups and correct newlines
First thing, someone can specify that a group is empty. If this is
done, technically a group with channels would be generated at the end.
However, because they were empty and did not exist, whitespacing would
be generated, leading to a usable, but mangled config file.
The `filter` solves this problem by removing empty strings (which are
generated by groups that are empty) from the output strings to place
in the file.
Lastly, because the whitespacing was fixed elsewhere in the file, the
crazy double-newline at the end was changed to a single newline.
However, the double newline within the `concatStringsSep` is still
required, because the list that is being concatenated together is a
list of channel configurations. Each element corresponds to one of the
groups specified, whose contents are the channels specified within.
The double newline is needed because each string element is lacking a
trailing newline, because `concatStringsSep` does not add the
separator to the end of the last element in the list. So, the last
channel to be configured will not have that newline appended when the
channel-configuration list is created, thus, 2 are inserted here.
* mbsync: update test input to use per-account channels
* mbsync: comment how old/new style collision handled
This is left in the test input for now, because I think it is useful
to see why certain things are happening the way they are.
* mbsync: update test output pattern
The test output should now have the correct configuration according to
the way I have specified it in the input file.
* mbsync: use format script on new code
* mbsync: add KarlJoad as maintainer
Co-authored-by: Nick Hu <me@nickhu.co.uk>
The `ExecStart=` option of systemd must take arguments fully quoted.
That is,
"-sshargs=-i somekey"
and not
-ssargs="-i somekey"
Additionally, inside arguments passed to unison, `=` characters must
be quoted. After unquotation by systemd, one must have
-sshargs=-o Foo\=4
instead of
-sshargs=-o Foo=4
The apropos software is useful to get a list of manpages matching a
description or to get a list of all manpages. The latter feature is
used by Emacs to get manpage completion (`M-x man`).
To have apropos working, a database of all available manpages must be
built with mandb. This is what this commits does.
A similar change was done for NixOS:
edc6a76cc0
Running `zplug install` will always product output, even if there is
nothing to do.
Gating it behind a `zplug check` eliminates that output when there is
nothing to do, and is recommended in the zplug README.
Adds a new `keybindings` option to the `vscode` configuration.
It contains a list of key bindings, which will be written to
`%vscode-dir%/User/keybindings.json`.
PR #1351
The previous implementation would allow variables to sneak into the
file names. This commit makes sure the resulting target file path
exactly matches the expected path.
This removes the dependency on the `nixpkgs` channel within the
modules for state version ≥ 20.09. The default Nixpkgs source starting
from this state version is the path of the `pkgs` argument used to
bootstrap the Home Manager modeuls.
This is a prerequisite for using Home Manager withing Nix flakes.
PR #1420
Before the profile commands would not run if a single package is
installed since `buildEnv` will produce a symlink directly to that
package. By adding this dummy package we ensure that a real directory
will be generated.
Fixes#1392
The kakoune editor has a plugin mechanism and several plugins are
already packaged under `pkgs.kakounePlugins`. However, adding these
packages to `home.packages` is not enough: the `kakoune` package needs
to be configured with the list of plugins to include, so that they get
sourced on start-up.
We add a `programs.kakoune.plugins` option, analogous to
`programs.vim.plugins`.
The change is backwards compatible since `pkgs.kakoune` is defined as
wrapKakoune kakoune-unwrapped { };
and `wrapKakoune` defaults the list of plugins to empty.
PR #1356
The git-send-email [0] script uses StartTLS if `smtpEncryption` is set
to `tls`, which can break services that don't support StartTLS.
[0]: bd42bbe1a4/git-send-email.perl (L1533)
PR #1395
Before this change,
```rust
fn main() {
println!("{:?}", glib::get_user_special_dir(glib::UserDirectory::Documents));
}
```
would return `None` even though `~/Documents` is available and
`xdg.userDirs.enable = true`. Checking the differences between
`xdg-user-dirs-update` shows that the latter has quotes around each
thing.
PR #1440
We were passing the separators for the `show-whitespaces` highlighter
verbatim. This was problematic in case one wanted to use, spaces,
quotes or `%` as separators since the resulting kakoune configuration
would be invalid.
According to kakoune's docs, the separator has to be one character
long, so we can use a simple rule for escaping them. It is possible
that people has been working this around by passing, e.g. `"' '"` as
separator in order to get a space (i.e., escaped explicitly by the
user), so we just let longer strings be used verbatim.
PR #1357
This makes the systemd module use the sd-switch application to perform
the unit switch during a generation activation.
Since the closure of sd-switch is relatively lightweight we
unconditionally pull it in as a dependency. We simultaneously remove
the `systemd.user.startServices` option and perform the switch action
automatically.
PR #1388
The previous fish integration for starship erroneously used parts of
POSIX-esque test syntax. It also used `-n` instead of `-z` to check
for an unset variable.
PR #1422
When running the service start script with `DISPLAY` set, a `gi`
import error is triggered. Blanking the variable will make the script
use a different code path that does not attempt to import `gi`.
Also moves activation script up into start of script instead.
PR #1415
The option to remove the default keybindings by setting the
`programs.qutebrowser.enableDefaultKeybindings` variable to `false`
had a list wrapped around the `config.py` line. This would cause a
type coercion error.
PR #1410
This option can be used to enable optional Spotifyd features, such as
looking up the Spotify password in the system keyring or enabling
MPRIS support.
PR #1390
Before the profile directory value would point directly to the build
output in the Nix store. Unfortunately this would cause an infinite
loop if the user's configuration directly or indirectly refers to the
profile directory value.
Fixes#1188
Emacs populates 'exec-path' at launch from the 'PATH' environment
variable. Likewise, the emacs derivation from nixpkgs populates
'load-path' from the 'NIX_PROFILES' variable. As neither of these are
available by default in the systemd user manager, revert to the
previous behavior of launching the Emacs daemon from a login shell.
Fixes#1354Fixes#1340
PR #1355
Add 'services.emacs.socketActivation.enable' for generating an
'emacs.socket' systemd unit.
Emacs since version 26 has supported socket activation, whereby an
external process manager such as systemd listens on a socket and passes
it to the Emacs daemon when the manager launches it. This improves
startup time of the user session and avoids launching the daemon when not
needed, for example when launching the user session via SSH.
This implementation hard-codes the socket path to the default for the
version of 'programs.emacs.finalPackage', because systemd does not
perform shell expansion in the socket unit's 'ListenStream' parameter
and it seems like an advanced use-case to change the socket path. Shell
expansion would be desirable as the socket path usually resides in
directories such as $XDG_RUNTIME_DIR or $TMPDIR.
Tests were added to verify behavior in the following cases:
- Emacs service with socket activation disabled
- Emacs 26 with socket activation enabled
- Emacs 27 with socket activation enabled
PR #1314
