Interactive non-login Bash shells now apply the current session
variables file before bashrcExtra and initExtra. New terminals pick up
values changed by home-manager switch.
Login shells skip the refresh because .profile already applied it.
Non-interactive shells skip it because sshd can read .bashrc for remote
commands, where startup output can corrupt protocols such as scp and
rsync.
Apply programs.bash.sessionVariables after generic values to preserve
its precedence. Expansions run on every interactive non-login shell, so
warn when direct self-references can change repeatedly. Declarative
search paths remain stable across repeated sourcing.
A profileExtra assignment that collides with a Home Manager-managed
variable is now overwritten in non-login shells. Move such overrides to
initExtra, which runs after the refresh.
targets.genericLinux no longer needs its separate initExtra source.
`literalExpression` is intended just to signify code that needs to stay
a string that gets represented exactly as-is for docs. It has been
misused heavily and people get confused repeatedly on when or not to use
it because of the rampant misuse.
Allow boolean values in sessionVariables for Home Manager shell modules and serialize them consistently in generated session files.
- modules/lib/shell: stringify boolean values using lib.boolToString in export helper
- modules/home-environment, programs.bash, programs.zsh: include lib.types.bool in sessionVariables option type
- tests: add IS_TRUE/IS_FALSE assertions in home-environment, bash, and zsh session-variables tests
Some applications consider environment variables to be “true” simply if
they are set. Previously it was not possible to have a var set and later
override (e.g. with `lib.mkForce`) to *unset* it. This makes that
possible. The filtering must be done in the `exportAll` as attempting to
do it in the option’s `apply` causing infinite recursion (likely why it
is `lazyAttrsOf`).
checkwinsize is enabled by default since bash5.
It's not much hassle if enabled, but it's not necessary anymore and it conflicts if attempting to piggy-back on .bashrc with other kinda-compatible shells (e.g. osh).
bash5 is default for a couple of years already, and there is no bash4 in there at all
I had programs.bash.sessionVariables.CDPATH = ... defined in the
different nix files. One was taken into account and the other one
ignored.
Turns out plain "types.attrs" has the following merging strategy:
===
merge = loc: foldl' (res: def: res // def.value) { };
===
Changing it to "attrsOf", the module system warned me of the conflicting
values.
Reduce maintenance burden and increase efficiency by automatically
importing modules following a specific convention.
Co-authored-by: awwpotato <awwpotato@voidq.com>
Signed-off-by: Austin Horstman <khaneliman12@gmail.com>
Can generate the config without installing application through home-manager. Helpful when a package is broken (or not provided) on a specific platform through nixpkgs and needs to be installed through other means but you still can benefit from the declarative configuration.
These (and the `*MD` functions apart from `literalMD`) are now no-ops
in nixpkgs and serve no purpose other than to add additional noise and
potentially mislead people into thinking unmarked DocBook documentation
will still be accepted.
Note that if backporting changes including documentation to 23.05,
the `mdDoc` calls will need to be re-added.
To reproduce this commit, run:
$ NIX_PATH=nixpkgs=flake:nixpkgs/e7e69199f0372364a6106a1e735f68604f4c5a25 \
nix shell nixpkgs#coreutils \
-c find . -name '*.nix' \
-exec nix run -- github:emilazy/nix-doc-munge/98dadf1f77351c2ba5dcb709a2a171d655f15099 \
--strip {} +
$ ./format
This process was automated by [my fork of `nix-doc-munge`]. All
conversions were automatically checked to produce the same DocBook
result when converted back, modulo minor typographical/formatting
differences on the acceptable-to-desirable spectrum.
To reproduce this commit, run:
$ NIX_PATH=nixpkgs=flake:nixpkgs/e7e69199f0372364a6106a1e735f68604f4c5a25 \
nix shell nixpkgs#coreutils \
-c find . -name '*.nix' \
-exec nix run -- github:emilazy/nix-doc-munge/98dadf1f77351c2ba5dcb709a2a171d655f15099 \
{} +
$ ./format
[my fork of `nix-doc-munge`]: https://github.com/emilazy/nix-doc-munge/tree/home-manager
The NixOS variant of Markdown doesn't make a distinction between
`<code>` and `<literal>` or `<quote>` and... quotes, and doesn't
support `<parameter>` or `<replaceable>`. These are infrequently used
(apart from `<code>`) and don't add much, so just convert them to
simpler forms to allow the options containing them to be converted
to Markdown automatically.
A few minor syntactic adjustments were also made to make
`nix-doc-munge`'s job easier.
Resolves#1843. Allows aliases to be expanded in initExtra, and adds a
visible bashrcExtra option for commands that should be run in ~/.bashrc
even by non-interactive shells.
The bash module always assigns a value to HISTFILE in the bashrc, even
when no value is explicitly set. This makes it impossible to tell bash
to use a different HISTFILE by setting the HISTFILE environment variable
HISTFILE=/tmp/bash_history bash
This changes the default value of programs.bash.historyFile to null, and
only writes the HISTFILE=... line to the bashrc if it is changed to
something else.
Unfortunately the document generator is not smart enough to quote the
`..` alias in the documentation which is very misleading. By making it
a literal example the quotes stay.
Unfortunately, using `attrsOf` is not possible since it results in too
eager evaluation. In particular, the
home.sessionVariables = {
FOO = "Hello";
BAR = "${config.home.sessionVariables.FOO} World!";
};
example will cause an infinite recursion.
This commit restores the option type of
- `home.sessionVariables`,
- `pam.sessionVariables`,
- `programs.bash.sessionVariables`, and
- `programs.zsh.sessionVariables`
to `attrs`. It also adds test cases for the above options to avoid
regressions.
Fixes#659
This variable adds some extra flexibility in constructing the
`~/.bashrc` file. Currently the option is hidden from public
documentation since the option name is provisional.
