BREAKING CHANGE: this changes Doom's CLI framework in subtle ways, which
is listed in greater detail below. If you've never extended Doom's CLI,
then this won't affect you, but otherwise it'd be recommended you read
on below.
This commit focuses on the CLI framework itself and backports some
foundational changes to its DSL and how it resolves command line
arguments to CLIs, validates input, displays documentation, and persists
state across sessions -- and more. This is done in preparation for the
final stretch towarding completing the CLI rewrite (see #4273).
This is also an effort to generalize Doom's CLI (both its framework and
bin/doom), to increase it versatility and make it a viable dev tool for
other Doom projects (on our Github org) and beyond.
However, there is a *lot* to cover so I'll try to be brief:
- Refactor: generalize Doom's CLI framework by moving all bin/doom
specific configuration/commands out of core-cli into bin/doom. This
makes it easier to use bin/doom as a project-agnostic development
tool (or for users to write their own).
- Refactor: change the namespace for CLI variables/functions from
doom-cli-X to doom-X.
- Fix: subcommands being mistaken as arguments. "doom make index" will
resolve to (defcli! (doom make index)) if it exists,
otherwise (defcli! (doom make)) with "index" as an argument. Before
this, it would resolve to the latter no matter what. &rest can
override this; with (defcli! (doom make) (&rest args)), (defcli! (doom
make index)) will never be invoked.
- Refactor!: redesign our output library (was core/autoload/output.el,
is now core/autoload/print.el), and how our CLI framework buffers and
logs output, and now merges logs across (exit! ...) restarts.
- Feat: add support for :before and :after pseudo commands. E.g.
(defcli! (:before doom help) () ...)
(defcli! (:after doom sync) () ...)
Caveat: unlike advice, only one of each can be defined per-command.
- Feat: option arguments now have rudimentary type validation (see
`doom-cli-option-arg-types`). E.g.
(defcli! (doom foo) ((foo ("--foo" num))) ...)
If NUM is not a numeric, it will throw a validation error.
Any type that isn't in `doom-cli-option-arg-types` will be treated as a
wildcard string type. `num` can also be replaced with a specification,
e.g. "HOST[:PORT]", and can be formatted by using symbol quotes:
"`HOST'[:`PORT']".
- Feat: it is no longer required that options *immediately* follow the command
that defines them (but it must be somewhere after it, not before). E.g.
With:
(defcli! (:before doom foo) ((foo ("--foo"))) ...)
(defcli! (doom foo baz) () ...)
Before:
FAIL: doom --foo foo baz
GOOD: doom foo --foo baz
FAIL: doom foo baz --foo
After:
FAIL: doom --foo foo baz
GOOD: doom foo --foo baz
GOOD: doom foo baz --foo
- Refactor: CLI session state is now kept in a doom-cli-context struct (which
can be bound to a CLI-local variable with &context in the arglist):
(defcli! (doom sync) (&context context)
(print! "Command: " (doom-cli-context-command context)))
These contexts are persisted across sessions (when restarted). This is
necessary to support seamless script restarting (i.e. execve
emulation) in post-3.0.
- Feat: Doom's CLI framework now understands "--". Everything after it will be
treated as regular arguments, instead of sub-commands or options.
- Refactor!: the semantics of &rest for CLIs has changed. It used to be "all
extra literal, non-option arguments". It now means *all* unprocessed
arguments, and its use will suppress "unrecognized option" errors, and
tells the framework not to process any further subcommands. Use &args
if you just want "all literal arguments following this command".
- Feat: add new auxiliary keywords for CLI arglists: &context, &multiple,
&flags, &args, &stdin, &whole, and &cli.
- &context SYM: binds the currently running context to SYM (a
`doom-cli-context` struct). Helpful for introspection or passing
along state when calling subcommands by hand (with `call!`).
- &stdin SYM: SYM will be bound to a string containing any input piped
into the running script, or nil if none. Use
`doom-cli-context-pipe-p` to detect whether the script has been
piped into or out of.
- &multiple OPTIONS...: allows all following OPTIONS to be repeated. E.g. "foo
-x a -x b -x c" will pass (list ("-x" . "a") ("-x" . "b") ("-x" .
"c")) as -x's value.
- &flags OPTIONS...: All options after "&flags" get an implicit --no-* switch
and cannot accept arguments. Will be set to :yes or :no depending on which flag is
provided, and nil if the flag isn't provided. Otherwise, a default
value can be specified in that options' arglist. E.g.
(defcli! (doom foo) (&flags (foo ("--foo" :no))) ...)
When called, this command sets FOO to :yes if --foo, :no if --no-foo, and
defaults to :no otherwise.
- &args SYM: this replaces what &rest used to be; it binds to SYM a
list of all unprocessed (non-option) arguments.
- &rest SYM: now binds SYM to a list of all unprocessed arguments, including
options. This also suppresses "unrecognized option" errors, but will render
any sub-commands inaccessible. E.g.
(defcli! (doom make) (&rest rest) ...)
;; These are now inaccessible!
(defcli! (doom make foo) (&rest rest) ...)
(defcli! (doom make bar) (&rest rest) ...)
- &cli SYM: binds SYM to the currently running `doom-cli` struct. Can also be
obtained via `(doom-cli-get (doom-cli-context-command context))`. Possibly
useful for introspection.
- feat: add defobsolete! macro for quickly defining obsolete commands.
- feat: add defalias! macro for quickly defining alias commands.
- feat: add defautoload! macro for defining an autoloaded command (won't
be loaded until it is called for).
- refactor!: rename defcligroup! to defgroup! for consistency.
- fix: CLIs will now recursively inherit plist properties from parent
defcli-group!'s (but will stack :prefix).
- refactor!: remove obsolete 'doom update':
- refactor!: further generalize 'doom ci'
- In an effort to generalize 'doom ci' (so other Doom--or
non-doom--projects can use it), all its subcommands have been
changed to operate on the current working directory's repo instead
of $EMACSDIR.
- Doom-specific CI configuration was moved to .github/ci.el.
- All 'doom ci' commands will now preload one of \$CURRENT_REPO_ROOT/ci.el or
\$DOOMDIR/ci.el before executing.
- refactor!: changed 'doom env'
- 'doom env {-c,--clear}' is now 'doom env {clear,c}'
- -r/--reject and -a/--allow may now be specified multiple times
- refactor!: rewrote CLI help framework and error handling to be more
sophisticated and detailed.
- feat: can now initiate $PAGER on output with (exit! :pager) (or use
:pager? to only invoke pager is output is longer than the terminal is
tall).
- refactor!: changed semantics+conventions for global bin/doom options
- Single-character global options are now uppercased, to distinguish them from
local options:
- -d (for debug mode) is now -D
- -y (to suppress prompts) is now -!
- -l (to load elisp) is now -L
- -h (short for --help) is now -?
- Replace --yes/-y switches with --force/-!
- -L/--load FILE: now silently ignores file errors.
- Add --strict-load FILE: does the same as -L/--load, but throws an error if
FILE does not exist/is unreadable.
- Add -E/--eval FORM: evaluates arbitrary lisp before commands are processed.
- -L/--load, --strict-load, and -E/--eval can now be used multiple times in
one command.
- Add --pager COMMAND to specify an explicit pager. Will also obey
$DOOMPAGER envvar. Does not obey $PAGER.
- Fix#3746: which was likely caused by the generated post-script overwriting
the old mid-execution. By salting the postscript filenames (with both an
overarching session ID and a step counter).
- Docs: document websites, environment variables, and exit codes in
'doom --help'
- Feat: add imenu support for def{cli,alias,obsolete}!
Ref: #4273Fix: #3746Fix: #3844
'doom run' was launching out of Emacs' default
directories (~/.config/emacs and ~/.emacs.d) instead of the parent
directory of bin/doom. This commit corrects that, but is a temporary
measure until the CLI rewrite. I'm also not totally sure this will work
on Windows...
Fix: #6389
Post-rewrite bin/doom will accept multiple -l's, but for now will only
accept one. This change was made to prepare for the documentation
generator, which will live outside the repo.
This is more predictable, and is safe as a global default in CLI
sessions (but not in interactive ones). This indirectly fixes case
insensitivity in our commit linter rules.
Reduces the amount of "noise" included in bin/doom's output.
Also fixes an issue where warnings during autoloads generation would
sneak into Doom's autoloads file, producing weird void-variable errors,
like
(void-variable . rainbow-delimiters:)
(void-variable . diredfl:)
(void-variable . company:)
We only need this magic in CLI sessions. It's better to only use half
the CPUs in interactive sessions (if the user has enabled
comp-deferred-compilation for some reason).
Fixes#5042
Still a long way to go, but this introduces a few niceties for
debugging CLI failures:
+ The (extended) output of the last bin/doom command is now logged to
~/.emacs.d/.local/doom.log
+ If an error occurs, short backtraces are displayed whether or not you
have debug mode on. The full backtrace is written to
~/.emacs.d/.local/doom.error.log.
+ bin/doom now aborts with a warning if:
- The script itself or its parent directory is a symlink. It's fine if
~/.emacs.d is symlinked though.
- Running bin/doom as root when your DOOMDIR isn't in /root/.
- If you're sporting Emacs 26.1 (now handled in the elisp side rather
than the /bin/sh shebang preamble).
+ If a 'doom sync' was aborted prematurely, you'll be warned that Doom
was left in an inconsistent state and that you must run `doom sync`
again.
May address #3746
Some doom commands will generate a temporary script at
~/.emacs.d/.local/.doom.sh so that it can run an arbitrary shell command
after the current invocation of bin/doom ends. Very useful for, say,
restarting the currently running doom command after a destructive
operation, like updating Doom's source code, tangling your literate
config, or for launching arbitrary programs, like a new instance of
Emacs. This is necessary because elisp lacks an execv implementation.
However, for some folks, .doom.sh wasn't executing at all. This meant:
1. Some `doom upgrade`s would upgrade Doom itself but never move on to
the second step of the process: updating its packages.
2. Literate config users could tangle their configs on `doom sync`, but
the actual syncing process would never happen (#3746).
3. `doom run` would do nothing.
I hadn't realized /bin/sh runs bash in POSIX mode (at least, on systems
where /bin/sh = bash, like nixOS or macOS). In POSIX mode the script
will abort the if a builtin command (like export) returns a non-zero
exit code. Since .doom.sh is basically a bunch of exports followed by an
arbitrary command, and there are some environment variables
that can trigger validation errors (like UID triggering a "read-only
variable" error), we have a problem.
Hopefully addresses #3746
Tangling would load org libraries. If org hasn't been installed yet,
this means the older version is loaded, later interfering with the
installation and byte-compilation of the new package, causing down the
road.
- Tangling no longer adds temp files to recentf (#3685)
- If :tangle yes is used, the result is no longer tangled to
/tmp/config.org.*.el
- In interactive sessions the org buffer is no longer interfered with
when tangling (by scrolling up to the top of the page, or undoing
overlays/markers).
- Tangling no longer triggers formatters (or any save/write hooks).
- Appease byte-compiler sama, complaining about free variables.
elisp lacks an execv implementation (or mature subprocess library), so
we exploit some splenderiffic hackery to get Emacs to execute arbitrary
shell commands after a 'doom ...' command completes. This allows us to
daisy chain doom commands in distinct sessions (wonderful for reloading
doom after a 'doom upgrade', which we do). This minimizes errors when a
'doom upgrade' pulls in breaking changes to Doom's CLI.
We also bring 'doom run' into elisp, since this new functionality
enables us to.
That 'doom refresh' and 'doom sync' do the same thing has been a point
of confusion for a while. By phasing out 'doom refresh' we can finally
put that to rest.
We no longer need two separate autoloads files, so I merged them and
optimized its generation logic.
Other changes
- Doom will refuse to start up (with a helpful error) if it's in an
incomplete state. This should hopefully reduce the number of bug
reports from folks that have done something weird, e.g.
1. You've changed Emacs versions without running 'doom sync -b'.
2. You've updated Doom outside of `doom upgrade` and didn't run `doom
sync -u`.
3. You've forgotten to run 'doom sync' in the first place!
4. If a previous 'doom ...' command was aborted midway without running
'doom sync' afterwards.
- 'doom sync' will emit reminders that you need to reload/restart Emacs
- Autoloads API now uses the `doom-autoloads-` prefix, intead of
'doom-cli-autoloads-', as will be the new convention in the coming
rewrite.
- Errors from within the package autoloads should be easier to invoke
the debugger on.
- `doom-modules` is now stored in your autoloads file. Your module list
will soon be frozen between calls to 'doom sync' to allow for our new,
atomic CLI I'm working on. This will also means the `doom!` block
won't cost anything in interactive sessions.
`format` isn't an appropriate for this library, considering it (and
future additions to it) will be mainly concerned with printing or
capturing output.
