emacs/doc/misc/use-package.texi

\input texinfo    @c -*- texinfo -*-
@c %**start of header
@setfilename ../../use-package.info
@settitle use-package User Manual
@include docstyle.texi
@syncodeindex vr cp
@syncodeindex fn cp
@c %**end of header

@copying
This manual is for use-package, a configuration macro for simplifying
your init file.

Copyright @copyright{} 2022 Free Software Foundation, Inc.

@quotation
Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License, Version 1.3 or
any later version published by the Free Software Foundation; with no
Invariant Sections, with the Front-Cover Texts being ``A GNU Manual'',
and with the Back-Cover Texts as in (a) below.  A copy of the license
is included in the section entitled ``GNU Free Documentation License''.

(a) The FSF's Back-Cover Text is: ``You have the freedom to copy and
modify this GNU manual.''
@end quotation
@end copying

@dircategory Emacs misc features
@direntry
* use-package: (use-package). Declarative package configuration for Emacs.
@end direntry

@finalout
@titlepage
@title use-package User Manual
@subtitle for version 2.4.5
@author John Wiegley & Stefan Kangas
@page
@vskip 0pt plus 1filll
@insertcopying
@end titlepage

@contents

@ifnottex
@node Top
@top use-package User Manual

The @code{use-package} macro allows you to isolate package
customization in your init file in a declarative way.  It takes care
of a lot of things for you that would otherwise require a lot of
repetitive boilerplate code.  It can help with common customization,
such as binding keys, setting up hooks, customizing user options and
faces, autoloading, and more.  It also helps you keep Emacs startup
fast, even when you use many (even hundreds) of packages.

Note that use-package is not a package manager.  Although use-package
does have the useful capability to interface with the Emacs package
manager, its primary purpose is for the configuration and loading of
packages.

@insertcopying

@menu
* Basic Concepts::              Basic concepts of use-package.
* Getting Started::             A gentle introduction to use-package.
* Loading Packages::            How and when packages are loaded.
* Configuring Packages::        Package configuration keywords.
* Installing packages::         Ensuring packages are available.
* Byte-compiling::              Byte-compiling your init file.
* Troubleshooting::             What to do when there's trouble.

Appendices
* Keyword extensions::          Adding new use-package keywords.
* History::                     History and acknowledgments.
* GNU Free Documentation License::  The license for this manual.
* Index::
@end menu
@end ifnottex

@c ----------------------------------------------------------------------------
@node Basic Concepts
@chapter Basic Concepts

use-package provides the @code{use-package} macro, that simplifies the
customization and use of packages in Emacs.  It was created for a few
basic reasons, each of which drove the design.  Understanding these
reasons may help make some of those decisions clearer:

@enumerate
@item
To gather all configuration details of a package into one place,
making it easier to copy, disable, or move it elsewhere in the init
file.

@item
To reduce duplication and boilerplate, capturing several common
practices as mere keywords both easy and intuitive to use.

@item
To make startup time of Emacs as quick as possible, without
sacrificing the quantity of add-on packages used.

@item
To make it so errors encountered during startup disable only the
package raising the error, and as little else as possible, leaving as
close to a functional Emacs as possible.

@item
To allow byte-compilation of one's init file so that any warnings or
errors seen are meaningful.  In this way, even if byte-compilation is
not used for speed (reason 3), it can still be used as a sanity check.
@end enumerate

It is worth noting that use-package is not intended to replace the
standard @w{@code{M-x customize}}.  On the contrary, it is designed to
work together with it, for things that customize cannot do.

@c ----------------------------------------------------------------------------
@node Getting Started
@chapter Getting Started

This chapter provides instructions and examples for quickly getting
started with use-package.  The first thing you need to do is make sure
that @samp{use-package} itself is loaded.  To do that, put this at the
top of your init file:

@lisp
(require 'use-package)
(require 'bind-key)            ; if you use any :bind variant
@end lisp

The above makes the @code{use-macro} for in the rest of your init
file.  In this manual, we call each call to @code{use-macro} a
@dfn{declaration}, to highlight the declarative nature of its
semantic.

To unconditionally load a package named @samp{foo}, add the following
declaration to your init file:

@lisp
(use-package foo)
@end lisp

@noindent
This declaration is equivalent to using @code{require}, with some
use-package specific error handling added in.  Just like require, it
needs the package @samp{foo} to be installed and available in your
@code{load-path} (@pxref{Installing packages}).

To evaluate Lisp code @emph{before} the @samp{foo} package is loaded,
use the @code{:init} keyword:

@lisp
(use-package foo
  :init
  (setq foo-variable t))
@end lisp

Similarly, @code{:config} can be used to execute code @emph{after} a
package is loaded.  In cases where loading is done lazily
(@pxref{Loading Packages}), this execution is deferred until after the
autoload occurs.  As you might expect, you can use @code{:init} and
@code{:config} together:

@lisp
(use-package foo
  :init
  (setq foo-variable t)
  :config
  (foo-mode 1))
@end lisp

The above declarations will all load the @samp{foo} package
immediately.  In most cases, this is not necessary or desirable, as
that will slow down Emacs startup.  Instead, you should try to set
things up so that packages are only loaded when they are actually
needed (autoloading).  If you have installed a package from
@acronym{GNU ELPA} that provides it's own autoloads, it is often
enough to say:

@lisp
(use-package foo
  :defer t)
@end lisp

@noindent
This will avoid loading the package.  Now, when you run any autoloaded
command, the package @samp{foo} is loaded automatically.  Package
authors will make their own decisions about which commands are marked
to autoload by default.

In some cases, you might need or want to provide your own autoloads.
The below more complex example autoloads the commands
@code{isearch-moccur} and @code{isearch-all} from
@file{color-moccur.el}, and binds keys both globally and in
@code{isearch-mode-map}.  When one of these commands are used, the
package is loaded.  At that point, @code{moccur-edit} is also loaded,
to allow editing of the @code{moccur} buffer.

@lisp
(use-package color-moccur
  :commands (isearch-moccur isearch-all)
  :bind (("M-s O" . moccur)
         :map isearch-mode-map
         ("M-o" . isearch-moccur)
         ("M-O" . isearch-moccur-all))
  :init
  (setq isearch-lazy-highlight t)
  :config
  (use-package moccur-edit))
@end lisp

Some packages will suggest ready-made @code{use-package} declarations
that you can use.  Where possible, it is a good idea to copy them, and
use that as a starting point.

That should be enough to get you started!

@c ----------------------------------------------------------------------------
@node Loading Packages
@chapter Loading Packages

@cindex loading packages
Before use-package can load an Emacs Lisp package, it must be
available in a directory on your @code{load-path}.  When you install
packages using the built-in @code{install-package} command, it will do
this automatically for you.  Packages shipped with Emacs (built-in
packages) are always available.

If you install packages manually, you must make sure they are
available on your @code{load-path}.  @xref{Lisp Libraries,,, emacs,
GNU Emacs Manual} for details.

Some packages have more than one library.  In those cases, you might
need more than one @code{use-package} declaration to make sure it is
properly loaded.  For complex configurations, you might also need more
than one declaration for a package with the same name.

use-package can interface with @samp{package.el} to install packages
on Emacs start.  @xref{Installing packages} for details.

@menu
* Loading basics::              How and when packages are loaded.
* Deferring loading::           Loading packages later.
* Forcing loading::             Loading packages immediately.
* Conditional loading::         Loading packages conditionally.
* Loading sequentially::        Loading packages in sequence.
* Load dependencies::           Don't load without dependencies.
* Load path::                   Using a custom @code{load-path}.
* Manual autoloads::            Setting up autoloads manually.
@end menu

@node Loading basics
@section How and when use-package loads packages

The @code{use-package} macro either will either load a package
immediately, or when they are first used (autoloading).  In the
simplest case, a @code{use-package} declaration loads a package when
it is evaluated.@footnote{This happens both at run-time and at
compile-time.  @xref{Byte-compiling}.}  If the declaration is in your
init file, this happens automatically each time Emacs is started.

For example, the below declaration immediately loads the library
@code{foo}, just like @code{require} would.  If the library @samp{foo}
is not available in your @code{load-path}, it logs a warning to the
@samp{*Messages*} buffer:

@lisp
(use-package foo)
@end lisp

Note that a ``package'' is different from an Emacs Lisp ``library''.
The above declaration tells use-package to load the @emph{library}
@file{foo.el}, which the overwhelming majority of cases also resides
in a @emph{package} named @code{foo}.  But the @code{foo} package
might also contain a library named @file{foo-extra.el}.  If that
library is not loaded automatically, you will need a separate
@code{use-package} declaration to make sure that it is.  This manual
will often use these terms interchangeably, as this distinction does
not usually matter, but you should keep it in mind for the cases when
it does.

The details of how and when you should load a package might differ
from one package to another.  When in doubt, refer to the package
documentation for details.

@node Deferring loading
@section Deferring package loading

@cindex autoloading packages
@cindex loading lazily
In the examples we have seen so far, use-package loads packages every
time you start Emacs, even if that package is never used.  That will
make starting Emacs slower.  use-package therefore tries to set things
up in such a way that it only loads packages when a command is first
used (either with @kbd{M-x} or some key binding).  This is based on
autoloading, a full description of which is outside the scope of this
manual.  @xref{Autoload,,, elisp, GNU Emacs Lisp Reference Manual} for
the full story.

@cindex triggers, for loading packages
Some @code{use-package} keywords provide autoload @dfn{triggers} that
cause a package to be loaded when certain events occur.  For example,
the @code{:hook} keyword sets up a trigger that fires when the
specified hook is run, and then loads the package automatically.  The
other trigger keywords, all of which are described later in this
manual, are @code{:commands}, @code{:bind}, @code{:bind*},
@code{:bind-keymap}, @code{:bind-keymap*}, @code{:mode}, and
@code{:interpreter}.

@subheading The @code{:defer} keyword

@findex :defer
If you did not specify any autoloading keyword, use-package will fall
back to loading the package immediately (typically when Emacs is
starting up).  This can be overridden using the @code{:defer} keyword.
It takes one boolean argument: a non-@code{nil} value means to stop
this package from being immediately loaded.  Here is an example of
using @code{:defer} to postpone loading the package @samp{foo}:

@lisp
(use-package foo
  :defer t)
@end lisp

Using @code{:defer t} by itself like this is rarely useful.
Typically, you would only use it together with a keyword like
@code{:config} (@pxref{Lisp Configuration}), or @code{:ensure}
(@pxref{Installing packages}).

@subheading Defer loading until idle for N seconds

You can also give a numeric argument @var{N} to @w{@code{:defer}} to
specify that a package should be loaded (if it hasn't already) after
Emacs has been idle for @var{N} seconds.  For example, use this to
make use-package load @samp{foo} after 30 seconds of idle time:

@lisp
(use-package foo
  :defer 30)
@end lisp

@subheading When to use @code{:defer}

When using autoloading keywords, there is no need to also use
@code{:defer}.  It doesn't hurt anything to add it in this case,
perhaps for extra clarity, but it is redundant.

You should use @code{:defer} to force deferred loading, in cases when
use-package isn't creating any autoloads for you.  For example, you
might know that some other package will already do something to cause
your package to load at the appropriate time.  This is usually the
case when you install a package using @code{package-install}, as
packages installed in this way normally always have their own
autoloads already set up.

@subheading Making @w{@code{:defer t}} the default

@vindex use-package-always-defer
If you customize the user option @code{use-package-always-defer} to
non-@code{nil}, the @code{use-package} macro will behave as if
@w{@code{:defer t}} is always specified.  This can be overridden for
individual declarations using either @w{@code{:defer nil}} or
@w{@code{:demand t}} (@pxref{Forcing loading}).

@node Forcing loading
@section Forcing package to load immediately

@findex :demand
The presence of autoloading trigger keywords can be overridden using
@code{:demand t}, which forces the package to load immediately.  Thus,
even if you use an autoloading keyword such as @code{:bind}
(@pxref{Key bindings}), adding @code{:demand} will force loading to
occur immediately.  It will also avoid creating an autoload for the
bound key, as it would be redundant.

If you specify both @w{@code{:demand t}} and @w{@code{:defer t}}, the
@code{:defer} keyword will take precedence.

@node Conditional loading
@section Loading packages conditionally

@findex :if
@findex :when
@findex :unless
The @code{:if}, @code{:when}, and @code{:unless} keywords predicates
the loading and initialization of packages.  They all accept one
argument, an Emacs Lisp form that is evaluated at run-time.

If the argument of the @code{:if} keyword evaluates to non-@code{nil},
the package will be loaded and initialized.  The @code{:when} keyword
is provided as an alias for @code{:if}.  Finally, the @code{:unless}
keyword is the inverse of @code{:if}, such that @w{@code{:unless foo}}
means the same thing as @w{@code{:if (not foo)}}.

For example, if you only want @samp{foo} in graphical Emacs sessions,
you could use the following:

@lisp
(use-package foo
  :if (display-graphic-p))
@end lisp

Another common use case is to make it conditional on the operating
system:

@lisp
(use-package foo
  :if (memq window-system '(mac ns)))
@end lisp

@cindex conditional loading before @code{:preface} or @code{:ensure}
If you need to conditionalize a use-package form so that the condition
occurs before even @code{:ensure} or @code{:preface}, use @code{when}
around the use-package form itself.  For example:

@lisp
(when (memq window-system '(mac ns))
  (use-package foo
    :ensure t))
@end lisp

@node Loading sequentially
@section Loading packages in sequence

@findex :after
Sometimes it only makes sense to configure a package after another one
has been loaded, because certain variables or functions are not in
scope until that time.  This can achieved with the @code{:after}
keyword, which allows a fairly rich description of the exact
conditions when loading should occur.  It takes either a symbol
indicating the package name, a list of such symbols, or a list of
selectors (see below).

Here is an example of using the @acronym{GNU ELPA} packages hydra,
ivy, and ivy-hydra.  Note that ivy-hydra will always be loaded last:

@lisp
(use-package hydra)

(use-package ivy)

(use-package ivy-hydra
  :after (ivy hydra))
@end lisp

In this case, because the declarations are evaluated in the order they
occur, the use of @code{:after} is not strictly necessary.  However,
if @samp{hydra} and @samp{ivy} were to be autoloaded, using
@code{:after} guarantees that @samp{ivy-hydra} is not loaded until it
is actually needed.  By using @code{:after}, the above code will also
work even if the order of the declaration changes.  This means that
moving things around in your init file is less likely to break things.

@subheading Using @code{:after} selectors

@findex :all (with :after)
@findex :any (with :after)
The @code{:after} keyword also accepts a list of selectors.  By
default, @code{:after (foo bar)} is the same as @w{@code{:after (:all
foo bar)}}, meaning that loading of the given package will not happen
until both @code{foo} and @code{bar} have been loaded.  Here are some
of the other possibilities:

@verbatim
:after (foo bar)
:after (:all foo bar)
:after (:any foo bar)
:after (:all (:any foo bar) (:any baz quux))
:after (:any (:all foo bar) (:all baz quux))
@end verbatim

When you nest selectors, such as @code{(:any (:all foo bar) (:all baz
quux))}, it means that the package will be loaded when either both
@code{foo} and @code{bar} have been loaded, or when both @code{baz}
and @code{quux} have been loaded.

Pay attention when setting @code{use-package-always-defer} to a
non-@code{nil} value, and also using the @code{:after} keyword.  In
this case, you will need to specify how the declared package is to be
loaded: for example, by some @code{:bind}.  If you are not using one
of the keywords that registers autoloads, such as @code{:bind} or
@code{:hook}, and your package manager does not provide autoloads, it
is possible that your package will never be loaded if you do not add
@code{:demand t} to those declarations.

@node Load dependencies
@section Prevent loading if dependencies are missing

@findex :requires
While the @code{:after} keyword delays loading until the dependencies
are loaded, the somewhat simpler @code{:requires} keyword @emph{never}
loads the package if the dependencies are not available when the
@code{use-package} declaration is evaluated.  In this context,
``available'' means that @code{foo} is available if @w{@code{(featurep
'foo)}} evaluates to a non-@code{nil} value.  For example:

@lisp
(use-package abbrev
  :requires foo)
@end lisp

This is the same as:

@lisp
(use-package abbrev
  :if (featurep 'foo))
@end lisp

As a convenience, a list of such packages may be specified:

@lisp
(use-package abbrev
  :requires (foo bar baz))
@end lisp

For more complex logic, such as that supported by @code{:after},
simply use @code{:if} and the appropriate Lisp expression.

@node Load path
@section Setting a custom @code{load-path}

@findex :load-path
If a package resides in some directory that is not in your
@code{load-path}, use the @code{:load-path} keyword to add it.  It
takes a symbol, a function, a string or a list of strings.  If the
path is relative, it is expanded within @code{user-emacs-directory}.

For example:

@lisp
(use-package ess-site
  :load-path "site-lisp/ess/lisp/"
  :commands R)
@end lisp

Note that when using a symbol or a function to provide a dynamically
generated list of paths, you must inform the byte-compiler of this
definition so that the value is available at byte-compilation time.
This is done by using the special form @code{eval-and-compile} (as
opposed to @code{eval-when-compile}).  Further, this value is fixed at
whatever was determined during compilation, to avoid looking up the
same information again on each startup.  For example:

@lisp
(eval-and-compile
  (defun ess-site-load-path ()
    (shell-command "find ~ -path ess/lisp")))

(use-package ess-site
  :load-path (lambda () (list (ess-site-load-path)))
  :commands R)
@end lisp

@node Manual autoloads
@section Setting up autoloads manually

@findex :commands
@findex :autoload
To autoload an interactive command, use the @code{:commands} keyword.
When you use the @code{:commands} keyword, it creates autoloads for
those commands (which defers loading of the module until they are
used).  The @code{:commands} keyword takes either a symbol or a list
of symbols.

The @code{:autoload} keyword works like @code{:commands}, but is used
to autoload non-interactive functions.  Here is an example:

@lisp
(use-package org-crypt
  :autoload org-crypt-use-before-save-magic)
@end lisp

@c ----------------------------------------------------------------------------
@node Configuring Packages
@chapter Configuring Packages

This chapter describes the various keywords provided by
@code{use-package} that helps you configure packages.

@menu
* Lisp Configuration::          Using Lisp to configure packages.
* Key bindings::                Making your own keybindings.
* Hooks::                       Adding functions to hooks.
* Modes and interpreters::      Enabling modes automatically.
* Magic handlers::              Using regexps to enable modes.
* User options::                Setting user options.
* Faces::                       Customizing faces.
* Hiding minor modes::          Tidying up the mode line.
@end menu

@node Lisp Configuration
@section Using Lisp code for configuring packages

The most general way to add customizations are the @code{:preface},
@code{:init}, and @code{:config} keywords.  They all accept one or
more Emacs Lisp forms, up to the next keyword, that are evaluated in
order.  This lets you add arbitrary Lisp code to your
@code{use-package} declarations.

The only difference between these keywords is when they are evaluated.

@menu
* Preface keyword::             Evaluate code before anything else.
* Init keyword::                Evaluate code before loading package.
* Config keyword::              Evaluate code after loading package.
* Best practices::              When to use @code{:config}, @code{:init}, and @code{:preface}.
@end menu

@node Preface keyword
@subsection @code{:preface} is evaluated first

@findex :preface
The @code{:preface} section is evaluated before anything else, except
@code{:disabled} and @code{:ensure}.  It can be used to establish
function and variable definitions that will:

@enumerate
@item
Make the byte-compiler happy.  It will not complain about functions
whose definitions are unknown because you have them within a guard
block.

@item
Define code that can be used in an @code{:if} test.
@end enumerate

Note that whatever is specified within @code{:preface} is evaluated
both at load time and at byte-compilation time, in order to ensure
that definitions are seen by both the Lisp evaluator and the
byte-compiler.  Therefore, you should avoid having any side-effects in
your preface, and restrict it to symbol declarations and definitions.

@node Init keyword
@subsection @code{:init} is evaluated before loading package

@findex :init
The @code{:init} section is evaluated just before the package is
loaded.  Note that the @code{:init} form is run unconditionally --
even if the @code{foo} package happens to not exist on your system.
You must therefore remember to restrict @code{:init} code to only what
would succeed either way.  @code{:init} also always happens before
package load, whether @code{:config} has been deferred or not.

@node Config keyword
@subsection @code{:config} is evaluated after loading package

@findex :config
The @code{:config} section is evaluated after the package has been
loaded.  If the package is loaded immediately, this happens
immediately after that, but if loading is done lazily (@pxref{Loading
Packages}), this is deferred until after the package has been loaded.

In general, you should keep @code{:init} forms as simple and quick as
possible, and put as much as you can get away with into the
@code{:config} section.  That way, deferred loading can help your
Emacs start as quickly as possible.

@node Best practices
@subheading When to use @code{:preface}, @code{:config} and @code{:init}?

Where possible, it is better to avoid @code{:preface}, @code{:config}
and @code{:init}.  Instead, prefer autoloading keywords such as
@code{:bind}, @code{:hook}, and @code{:mode}, as they will take care
of setting up autoloads for you without any need for boilerplate code.
For example, consider the following declaration:

@lisp
(use-package foo
  :init
  (add-hook 'some-hook 'foo-mode))
@end lisp

This has two problems.  First, it will unconditionally load the
package @samp{foo} on startup, which will make things slower.  You can
fix this by adding @code{:defer t}:

@lisp
(use-package foo
  :defer t
  :init
  (add-hook 'some-hook 'foo-mode))
@end lisp

This is better, as @samp{foo} is now only loaded when it is actually
needed (that is, when the hook @samp{some-hook} is run).

The second problem is that there is a lot of boilerplate that you have
to write.  In this case, it might not be so bad, but avoiding that was
what use-package was made to avoid.  The better option in this case is
therefore to use @code{:hook} (@xref{Hooks}), which also implies
@w{@code{:defer t}}.  The above is thereby reduced down to:

@lisp
(use-package foo
  :hook some-hook)
@end lisp

use-package will set up autoloading for you, and your Emacs startup
time will not suffer one bit.

@node Key bindings
@section Key bindings

@cindex :bind
@cindex binding keys
@cindex key bindings
One common thing to do when loading a package is to bind a key to
commands within that module.  Without use-package, this would be done
using a combination of @code{keymap-local-set},
@code{keymap-global-set} and various autoloads.  With use-package, you
can simplify this using the @code{:bind} keyword.

@menu
* Global keybindings::          Bindings you can use anywhere.
* Binding in keymaps::          Bindings for particular modes.
* Binding to a keymap::         Binding a key to a keymap.
* Binding to repeat-maps::      Binding repeating keys.
* Displaying keybindings::      Displaying personal key bindings.
@end menu

@node Global keybindings
@subsection Global keybindings

@findex :bind
To bind keys globally, the @code{:bind} keyword takes either a single
cons or a list of conses.  Every cons has the form @code{(@var{key}
. @var{command}}, where @var{key} is a string indicating the key to
bind, and @var{command} is the name of a command (a symbol).  The
syntax for the keys is similar to the syntax used by the @code{kbd}
function (@pxref{Init Rebinding,,, emacs, GNU Emacs Manual} for more
information).

@subheading Using @code{:bind} with a single cons

Here is an example of using a single cons:

@lisp
(use-package ace-jump-mode
  :bind ("C-." . ace-jump-mode))
@end lisp

This does two things: first, it creates an autoload for the
@code{ace-jump-mode} command and defers loading of the
@code{ace-jump-mode} package until you actually use it.  Second, it
binds the key @code{C-.} to that command globally.

@subheading Using @code{:bind} with a list of conses

Here is an example of using @code{:bind} with a list of conses:

@lisp
(use-package hi-lock
  :bind (("M-o l" . highlight-lines-matching-regexp)
         ("M-o r" . highlight-regexp)
         ("M-o w" . highlight-phrase)))
@end lisp

@subheading Using special keys

Inside key strings, special keys like @kbd{TAB} or @kbd{F1}--@kbd{F12}
have to be written inside angle brackets, e.g.  @code{"C-<up>"}.
Standalone special keys (and some combinations) can be written in
square brackets, e.g.@ @code{[tab]} instead of @code{"<tab>"}.

Examples:

@lisp
(use-package helm
  :bind (("M-x" . helm-M-x)
         ("M-<f5>" . helm-find-files)
         ([f10] . helm-buffers-list)
         ([S-f10] . helm-recentf)))
@end lisp

@subheading Remapping commands

Remapping commands with @code{:bind} and @code{bind-key} works as
expected, because when the binding is a vector, it is passed straight
to @code{define-key}.  @xref{Remapping Commands,,, elisp, GNU Emacs
Lisp Reference Manual}) for more information about command remapping.
For example, the following declaration will rebind
@code{fill-paragraph} (bound to @kbd{M-q} by default) to
@code{unfill-toggle}:

@lisp
(use-package unfill
  :bind ([remap fill-paragraph] . unfill-toggle))
@end lisp

@subheading What @code{:bind} does behind the scenes

To understand what @code{:bind} does behind the scenes, it might be
useful to consider an example:

@lisp
(use-package ace-jump-mode
  :bind ("C-." . ace-jump-mode))
@end lisp

This could be expressed in a much more verbose way with the
@code{:commands} and @code{:init} keywords.

@lisp
(use-package ace-jump-mode
  :commands ace-jump-mode
  :init
  (bind-key "C-." 'ace-jump-mode))
@end lisp

Without using even the @code{:commands} keyword, we could also write
the above like so:

@lisp
(use-package ace-jump-mode
  :defer t
  :init
  (autoload 'ace-jump-mode "ace-jump-mode" nil t)
  (bind-key "C-." 'ace-jump-mode))
@end lisp

Although these three forms are all equivalent, the first form is
usually the best, as it will save some typing.

@node Binding in keymaps
@subsection Key bindings in local keymaps

@findex :map, inside :bind
Slightly different from binding a key to a keymap, is binding a key
@emph{within} a local keymap that only exists after the package is
loaded.  @code{use-package} supports this with a @code{:map} modifier,
taking the local keymap to bind to:

@lisp
(use-package helm
  :bind (:map helm-command-map
         ("C-c h" . helm-execute-persistent-action)))
@end lisp

The effect of this statement is to wait until @code{helm} has loaded,
and then to bind the key @code{C-c h} to
@code{helm-execute-persistent-action} within Helm's local keymap,
@code{helm-command-map}.

Multiple uses of @code{:map} may be specified. Any binding occurring
before the first use of @code{:map} are applied to the global keymap:

@lisp
(use-package term
  :bind (("C-c t" . term)
         :map term-mode-map
         ("M-p" . term-send-up)
         ("M-n" . term-send-down)
         :map term-raw-map
         ("M-o" . other-window)
         ("M-p" . term-send-up)
         ("M-n" . term-send-down)))
@end lisp

@node Binding to a keymap
@subsection Binding to keymaps

@findex :bind-keymap, inside :bind
Normally @code{:bind} expects that commands are functions that will be
autoloaded from the given package. However, this does not work if one of
those commands is actually a keymap, since keymaps are not functions,
and cannot be autoloaded using the built-in @code{autoload} function.

To handle this case, @code{use-package} offers a special, limited
variant of @code{:bind} called @code{:bind-keymap}. The only difference
is that the ``commands'' bound to by @code{:bind-keymap} must be keymaps
defined in the package, rather than command functions. This is handled
behind the scenes by generating custom code that loads the package
containing the keymap, and then re-executes your keypress after the
first load, to reinterpret that keypress as a prefix key.

For example:

@lisp
(use-package foo
  :bind-keymap ("C-c p" . foo-command-map))
@end lisp

@node Binding to repeat-maps
@subsection Binding to repeat-maps

@findex :repeat-map, inside :bind
@cindex repeat-mode and use-package, using
A special case of binding within a local keymap is when that keymap is
used by @code{repeat-mode} @pxref{Repeating,,, emacs, GNU Emacs
Manual}.  These keymaps are usually defined specifically for
this.  Using the @code{:repeat-map} keyword, and passing it a name for
the map it defines, will bind all following keys inside that map, and
(by default) set the @code{repeat-map} property of each bound command
to that map.

The following example creates a keymap called
@code{git-gutter+-repeat-map}, makes four bindings in it as above,
then sets the @code{repeat-map} property of each bound command
(@code{git-gutter+-next-hunk} @code{git-gutter+-previous-hunk},
@code{git-gutter+-stage-hunks} and @code{git-gutter+-revert-hunk}) to
that keymap.

@lisp
(use-package git-gutter+
  :bind
  (:repeat-map git-gutter+-repeat-map
   ("n" . git-gutter+-next-hunk)
   ("p" . git-gutter+-previous-hunk)
   ("s" . git-gutter+-stage-hunks)
   ("r" . git-gutter+-revert-hunk)))
@end lisp

@findex :exit, inside :repeat-map and :bind
Specifying @code{:exit} inside the scope of @code{:repeat-map} will
prevent the @code{repeat-map} property being set, so that the command
can be used from within the repeat map, but after it using it the repeat
map will no longer be available.  This is useful for commands often used
at the end of a series of repeated commands:

@lisp
(use-package git-gutter+
  :bind
  (:repeat-map my/git-gutter+-repeat-map
   ("n" . git-gutter+-next-hunk)
   ("p" . git-gutter+-previous-hunk)
   ("s" . git-gutter+-stage-hunks)
   ("r" . git-gutter+-revert-hunk)
   :exit
   ("c" . magit-commit-create)
   ("C" . magit-commit)
   ("b" . magit-blame)))
@end lisp

@findex :continue, inside :repeat-map and :bind
Specifying @code{:continue} @emph{forces} setting the
@code{repeat-map} property (just like @emph{not} specifying
@code{:exit}), so the above snippet is equivalent to:

@lisp
(use-package git-gutter+
  :bind
  (:repeat-map my/git-gutter+-repeat-map
   :exit
   ("c" . magit-commit-create)
   ("C" . magit-commit)
   ("b" . magit-blame)
   :continue
   ("n" . git-gutter+-next-hunk)
   ("p" . git-gutter+-previous-hunk)
   ("s" . git-gutter+-stage-hunks)
   ("r" . git-gutter+-revert-hunk)))
@end lisp

@node Displaying keybindings
@subsection Displaying personal keybinding

@findex describe-personal-keybindings
The @code{:bind} keyword uses the @code{bind-keys} macro from the
@samp{bind-key.el} library to set up keybindings.  It keeps track of
all keybindings you make, so that you can display them separately from
the default keybindings.

Use @w{@code{M-x describe-personal-keybindings}} to see all
keybindings you've set using either the @code{:bind} keyword or the
@code{bind-keys} macro.

@node Hooks
@section Hooks

@cindex hooks
@findex :hook
The @code{:hook} keyword allows adding functions onto hooks.  It takes
one argument of the form @var{hooks}, specifying one or more functions
to add to one or more hooks.  For the purposes of @code{:hook}, the
name of hook variables should always exclude the @samp{-hook} suffix.
It is appended automatically for you, to save some typing.

For example, consider the following @code{use-package} declaration
that sets up autoloads for @code{company-mode} from the @samp{company}
package, and adds @samp{company-mode} to @code{prog-mode-hook}:

@lisp
(use-package company
  :commands company-mode
  :init
  (add-hook 'prog-mode-hook #'company-mode))
@end lisp

Using @code{:hook}, this can be simplified to:

@lisp
(use-package company
  :hook (prog-mode . company-mode))
@end lisp

Here, @code{:hook} will automatically set up autoloads for the
@code{company-mode} command, so there is no need to use
@code{:commands}.

The @code{:hook} keyword will also assume that the name of the
function you want to add is the same as the package name with
@samp{-mode} appended to it.  Taking this into account, you can
simplify the above to the equivalent:

@lisp
(use-package company
  :hook prog-mode)
@end lisp

@cindex multiple hooks
You can also provide a list of hooks.  When multiple hooks should be
applied, the following examples are all equivalent:

@lisp
(use-package company
  :hook (prog-mode text-mode))

(use-package company
  :hook ((prog-mode text-mode) . company-mode))

(use-package company
  :hook ((prog-mode . company-mode)
         (text-mode . company-mode)))

(use-package company
  :commands company-mode
  :init
  (add-hook 'prog-mode-hook #'company-mode)
  (add-hook 'text-mode-hook #'company-mode))
@end lisp

One common mistake when using @code{:hook} is to forget to omit the
@samp{-hook} suffix, which, as already explained, is appended
automatically.  Therefore, the following will not work, as it attempts
to add a function to non-existent @code{prog-mode-hook-hook}:

@lisp
;; DOES NOT WORK
(use-package ace-jump-mode
  :hook (prog-mode-hook . ace-jump-mode))
@end lisp

@vindex use-package-hook-name-suffix
If you do not like this behavior, you can customize the user option
@code{use-package-hook-name-suffix} to @code{nil}.  The value of this
variable is @samp{"-hook"} by default.

The use of @code{:hook}, as with @code{:bind}, @code{:mode},
@code{:interpreter}, etc., causes the functions being hooked to
implicitly be read as @code{:commands}.  This means that they will
establish interactive @code{autoload} definitions for that module, if
not already defined as functions), and so @code{:defer t} is also
implied by @code{:hook}.

@node Modes and interpreters
@section Modes and interpreters

@findex :mode
@findex :interpreter
Similar to @code{:bind}, you can use @code{:mode} and
@code{:interpreter} to establish a deferred binding within the
@code{auto-mode-alist} and @code{interpreter-mode-alist} variables.
The specifier to either keyword can be a cons cell, a list of cons
cells, or a string or regexp.

The following example reproduces the default @code{ruby-mode}
configuration, exactly as it is in Emacs out-of-the-box.  That mode is
enabled automatically when a file whose name matches the regexp
@code{"\\.rb\\'"} (a file with the @samp{.rb} extension), or when the
first line of the file (known as the ``shebang'') matches the string
@code{"ruby"}:

@lisp
(use-package ruby-mode
  :mode "\\.rb\\'"
  :interpreter "ruby")
@end lisp

The default @code{python-mode} configuration can be reproduced using
the below declaration.  Note that the package that should be loaded
differs from the mode name in this case, so we must use a cons:

@lisp
;; The package is "python" but the mode is "python-mode":
(use-package python
  :mode ("\\.py\\'" . python-mode)
  :interpreter ("python" . python-mode))
@end lisp

Both the @code{:mode} and @code{:interpreter} keywords also accept a
list of regexps:

@lisp
(use-package foo
  ;; Equivalent to "\\(ba[rz]\\)\\'":
  :mode ("\\.bar\\'" "\\.baz\\'")
  ;; Equivalent to "\\(foo[ab]\\)":
  :interpreter ("fooa" "foob"))
@end lisp

@node Magic handlers
@section Magic handlers

@findex :magic
@findex :magic-fallback
Similar to @code{:mode} and @code{:interpreter}, you can also use
@code{:magic} and @code{:magic-fallback} to cause certain function to
be run if the beginning of a file matches a given regular expression.
The difference between @code{:magic} and @code{:magic-fallback}, is
that the latter has a lower priority than @code{:mode}.

Here is an example:

@lisp
(use-package pdf-tools
  :magic ("%PDF" . pdf-view-mode)
  :config
  (pdf-tools-install :no-query))
@end lisp

This registers an autoloaded command for @code{pdf-view-mode}, defers
loading of @code{pdf-tools}, and runs @code{pdf-view-mode} if the
beginning of a buffer matches the string @code{"%PDF"}.

@node User options
@section User options

@findex :custom
In Emacs, you normally set customizable variables (user options) using
the @code{M-x customize} interface (@pxref{Easy Customization,,,
emacs, GNU Emacs Manual}).  We recommended this method for most users.
However, it is also possible to set them in your @code{use-package}
declarations by using the @code{:custom} keyword.

@lisp
(use-package comint
  :defer t
  :custom
  (comint-buffer-maximum-size 20000 "Increase comint buffer size.")
  (comint-prompt-read-only t "Make the prompt read only."))
@end lisp

This is better than using @code{setq} in a @code{:config} block, as
customizable variables might have some code associated with it that
Emacs will execute when you assign values to them.  In Emacs 29, there
is also the new @code{setopt} macro that does this for you.

Note that the values customized using this keyword are @emph{not}
saved in the standard Emacs @code{custom-file}.  You should therefore
set each user option using either the @code{:custom} keyword @emph{or}
@w{@code{M-x customize-option}}, which will save customized values in
the Emacs @code{custom-file}.  Do not use both for the same variable,
as this risk having conflicting values in your use-package declaration
and your @code{custom-file}.  This can lead to problems that are both
tricky and tedious to debug.

@node Faces
@section Faces

@findex :custom-face
The @code{:custom-face} keyword allows customization of package custom
faces.

@lisp
(use-package eruby-mode
  :custom-face
  (eruby-standard-face ((t (:slant italic)))))

(use-package example
  :custom-face
  (example-1-face ((t (:foreground "LightPink"))))
  (example-2-face ((t (:foreground "LightGreen"))) face-defspec-spec))

(use-package zenburn-theme
  :preface
  (setq my/zenburn-colors-alist
        '((fg . "#DCDCCC") (bg . "#1C1C1C") (cyan . "#93E0E3")))
  :custom-face
  (region ((t (:background ,(alist-get my/zenburn-colors-alist 'cyan)))))
  :config
  (load-theme 'zenburn t))
@end lisp

@node Hiding minor modes
@section Hiding minor modes with diminish and delight

@code{use-package} supports the diminish and delight packages, both of
which make it possible remove or change minor mode strings in your
mode-line.  Which one to use is up to you, but you should normally
only use one or the other -- never both.@footnote{When in doubt, you
might as well use diminish.}  To use either of them, you must first
install the corresponding package from @acronym{GNU ELPA}.

@menu
* Diminish::                    Hiding minor modes with Diminish.
* Delight::                     Hiding minor modes with Delight.
@end menu

@node Diminish
@subsection Diminish

@findex :diminish
When diminish@footnote{The diminish package is installable from
@acronym{GNU ELPA}.} is installed, you can use the @code{:diminish}
keyword.  First, add the following declaration to the beginning of
your init file.  The optional @w{@code{:ensure t}} makes sure the
package is installed if it isn't already (@pxref{Installing
packages}).

@lisp
(use-package diminish :ensure t)
@end lisp

The @code{:diminish} keyword takes either a minor mode symbol, a cons
of the symbol and its replacement string, or just a replacement
string, in which case the minor mode symbol is guessed to be the
package name with @samp{-mode} appended at the end:

@lisp
(use-package abbrev
  :diminish abbrev-mode
  :config
  (if (file-exists-p abbrev-file-name)
      (quietly-read-abbrev-file)))
@end lisp

@node Delight
@subsection Delight

@findex :delight
When delight@footnote{The @samp{delight} package is installable from
GNU ELPA.} is installed, you can use the @code{:delight} keyword.
First, add the following declaration to the beginning of your init
file.  The optional @w{@code{:ensure t}} makes sure the package is
installed if it isn't already (@pxref{Installing packages}).

@lisp
(use-package delight :ensure t)
@end lisp

The @code{:delight} keyword takes a minor mode symbol, a replacement
string, or quoted mode line data (in which case the minor mode symbol
is guessed to be the package name with @samp{-mode} appended at the
end), both of these, or several lists of both.  @xref{Mode Line
Data,,, elisp, GNU Emacs Lisp Reference Manual}.  If no arguments are
provided, the default mode name is hidden completely.

@lisp
;; Don't show anything for rainbow-mode.
(use-package rainbow-mode
  :delight)

;; Don't show anything for auto-revert-mode, which doesn't match
;; its package name.
(use-package autorevert
  :delight auto-revert-mode)

;; Remove the mode name for projectile-mode, but show the project name.
(use-package projectile
  :delight '(:eval (concat " " (projectile-project-name))))

;; Completely hide visual-line-mode and change auto-fill-mode to " AF".
(use-package emacs
  :delight
  (auto-fill-function " AF")
  (visual-line-mode))
@end lisp

@c ----------------------------------------------------------------------------
@node Installing packages
@chapter Installing packages automatically

The standard Emacs package manager is documented in the Emacs manual
(@pxref{Package Installation,,, emacs, GNU Emacs Manual}).  The
@code{use-package} macro provides the @code{:ensure} and @code{:pin}
keywords, that interface with that package manager to automatically
install packages.  This is particularly useful if you use your init
file on more than one system.

@menu
* Install package::
* Pinning packages::
* Other package managers::
@end menu

@node Install package
@section Installing package

@findex :ensure
The @code{:ensure} keyword makes use-package ask the Emacs package
manager to install a package if it is not already present on your
system.

For example:

@lisp
(use-package magit
  :ensure t)
@end lisp

If you need to install a different package from the one named by
@code{use-package}, you can use a symbol:

@lisp
(use-package tex
  :ensure auctex)
@end lisp

You can customize the user option @code{use-package-always-ensure} to
non-@code{nil} if you want this behavior to be global for all
packages.

@lisp
(require 'use-package-ensure)
(setq use-package-always-ensure t)
@end lisp

@noindent
You can override the above setting for a single package by adding
@w{@code{:ensure nil}} to its declaration.

@node Pinning packages
@section Pinning packages using @code{:pin}

@findex :pin
use-package can pin a package to a specific archive using the
@code{:pin} keyword.@footnote{The @code{:pin} keyword has no effect on
Emacs versions older than 24.4.}  This allows you to mix and match
packages from different archives.  The primary use-case for this is
preferring to install packages from @acronym{GNU ELPA} or
@acronym{NonGNU ELPA} (indicated by @code{gnu} and @code{nongnu},
respectively), while installing specific packages from third-party
archives.

For example:

@lisp
(use-package company
  :ensure t
  :pin gnu)   ; GNU ELPA
@end lisp

@vindex use-package-always-pin
Unfortunately, the third-party archive @acronym{MELPA} uses a
versioning scheme based on dates, which means that packages from that
archive are always preferred.  If you are using that archive, we
strongly encourage you to customize @code{use-package-always-pin} to
@code{nongnu}.  This guarantees that you are using a version of that
package that has been specifically marked for release by its
developer, and not a development snapshot.

@c FIXME: This needs clarifying.  AFAIK, :ensure does not update packages.
If you want to manually keep a package updated and ignore upstream
updates, you can pin it to @samp{manual}.  This will work as long as
you have not customized a repository to use that name in the
@code{package-archives} variable.

Example:

@lisp
(use-package org
  :ensure t
  ;; ignore org-mode from upstream and use a manually installed version
  :pin manual)
@end lisp

@code{use-package} signals an error if you try to pin a package to an
archive that is not configured using @code{package-archives} (except
from the special @samp{manual} archive).

@node Other package managers
@section Non-standard package managers

By default, use-package assumes that you are using the built-in
@code{package.el} package manager.  We expect that most users will
find that it is more than capable enough, even for advanced use cases.

@vindex use-package-ensure-function
However, some users might prefer to use a third-party package manager
for a specific circumstance or use case.  By setting the user option
@code{use-package-ensure-function} to the name of a function, you can
direct @code{:ensure} to use a different package manager for
installing packages.

For more details, please see the documentation of the package manager
you are using.  If you run into any bugs, it is often best to report
them directly to the developers of that package manager.

@c ----------------------------------------------------------------------------
@node Byte-compiling
@chapter Byte-compiling your init file

Some users might want to byte-compile their init file to make Emacs
startup even faster.  This is not recommended in most cases, as the
speed-up is often too small to be worth it, and can lead to confusion
if the byte-compiled files are out-of-date.  If you still want to do
it, read on.

@code{use-package} always loads every library that it can while a file
is being byte-compiled.  This helps silence spurious warnings about
unknown variables and functions.

@findex :defines
@findex :functions
However, there are times when this is just not enough.  For those
times, use the @code{:defines} and @code{:functions} keywords to
introduce dummy variable and function declarations solely for the sake
of silencing byte-compiler warnings.  For example:

@lisp
(use-package texinfo
  :defines texinfo-section-list
  :commands texinfo-mode
  :init
  (add-to-list 'auto-mode-alist '("\\.texi$" . texinfo-mode)))
@end lisp

If you need to silence a missing function warning, you can use
@code{:functions}:

@lisp
(use-package ruby-mode
  :mode "\\.rb\\'"
  :interpreter "ruby"
  :functions inf-ruby-keys
  :config
  (defun my-ruby-mode-hook ()
    (require 'inf-ruby)
    (inf-ruby-keys))

  (add-hook 'ruby-mode-hook 'my-ruby-mode-hook))
@end lisp

@findex :no-require
@cindex prevent a package from loading at compile-time
Normally, @code{use-package} will load each package at compile time
before compiling the configuration, to ensure that any necessary
symbols are in scope to satisfy the byte-compiler.  At times this can
cause problems, since a package may have special loading requirements,
and all that you want to use @code{use-package} for is to add a
configuration to the @code{eval-after-load} hook.  In such cases, use
the @code{:no-require} keyword:

@lisp
(use-package foo
  :no-require t
  :config
  (message "Evaluate this immediately after loading `foo'"))
@end lisp

@c ----------------------------------------------------------------------------
@node Troubleshooting
@chapter Troubleshooting

@cindex troubleshooting
@cindex debugging
If an error occurs while initializing or configuring a package, this
will not stop your Emacs from loading.  Instead, @code{use-package}
captures the error and reports it in a special @code{*Warnings*} popup
buffer, so that you can debug the situation in an otherwise functional
Emacs.

If you are having trouble when starting Emacs, you can pass Emacs the
@samp{--debug-init} command line flag.  @xref{Initial Options,,,
emacs, GNU Emacs Manual}.  To get even more information when using
that flag, add the following to your init file (these options are
documented below):

@lisp
(when init-file-debug
  (setq use-package-verbose t
        use-package-expand-minimally nil
        use-package-compute-statistics t
        debug-on-error t))
@end lisp

@cindex reporting bugs
@cindex expanding macro, for troubleshooting
Since @code{use-package} is a macro, the first step when you need to
dig deeper is usually to see what Emacs Lisp code your declaration
expands to.  You can either use the command @w{@kbd{M-x
pp-macroexpand-last-sexp}}, or wrap the use-package declaration in
@code{macroexpand} and evaluate it.  It is a good idea to include
their output in any bugs you file for use-package.

@menu
* Troubleshooting Options::
* Gathering Statistics::
* Disabling a package::
@end menu

@node Troubleshooting Options
@section Options that help when troubleshooting

@vindex use-package-expand-minimally
By default, use-package will attempts to catch and report errors that
occur during expansion of use-package declarations in your init file.
Customize the user option @code{use-package-expand-minimally} to a
non-@code{nil} value to disable this checking.

@findex :catch
This behavior may be overridden locally using the @code{:catch}
keyword.  If @code{t} or @code{nil}, it enables or disables catching
errors at load time.  It can also be a function taking two arguments:
the keyword being processed at the time the error was encountered, and
the error object (as generated by @code{condition-case}).  For
example:

@lisp
(use-package example
  ;; Note that errors are never trapped in the preface, since doing so would
  ;; hide definitions from the byte-compiler.
  :preface (message "I'm here at byte-compile and load time")
  :init (message "I'm always here at startup")
  :config
  (message "I'm always here after the package is loaded")
  (error "oops")
  ;; Don't try to (require 'example), this is just an example!
  :no-require t
  :catch (lambda (keyword err)
           (message (error-message-string err))))
@end lisp

Evaluating the above form will print these messages:

@verbatim
I’m here at byte-compile and load time
I’m always here at startup
Configuring package example...
I’m always here after the package is loaded
oops
@end verbatim

@node Gathering Statistics
@section Gathering Statistics

@vindex use-package-verbose
When a package is loaded, and if you have @code{use-package-verbose}
set to @code{t}, or if the package takes longer than 0.1 seconds to
load, you will see a message to indicate this loading activity in the
@code{*Messages*} buffer.  The same will happen for configuration, or
@code{:config} blocks, that take longer than 0.1 seconds to execute.

@vindex use-package-compute-statistics
If you'd like to see a summary how many packages you've loaded, what
stage of initialization they've reached, and how much aggregate time
they've spent (roughly), you can customize the user option
@code{use-package-compute-statistics} to a non-@code{nil} value.  Then
reload your packages, normally by restarting Emacs, to make sure that
use-package can gather statistics for all your packages.

@cindex use-package-report
Run the command @kbd{M-x use-package-report} to see the results.  The
buffer displayed is a tabulated list.  To sort rows based on a
particular column, move point to it and type @kbd{S}, or click the
column name at the top of the buffer on graphical displays.

@cindex use-package-reset-statistics
To reset all statistics that use-package has gathered for the current
Emacs invocation, run the command @kbd{M-x use-package-reset-statistics}.

Note that, if you are setting @code{use-package-compute-statistics}
directly in your init file, and not with @code{customize}, you must do
this after loading @code{use-package}, but before any
@code{use-package} forms.

@node Disabling a package
@section Disabling a package

@cindex disable package
@findex :disabled
The @code{:disabled} keyword inhibits loading a package, and all it's
customizations.  It is equivalent to commenting out or deleting the
definition.

You could use this, for example, to temporarily disable a package that
you're having difficulties with, or to avoid loading a package that
you're not currently using.

This example disables the @samp{foo} package:

@lisp
(use-package foo
  :disabled)
@end lisp

When byte-compiling your init file, use-package omits disabled
declarations from the output entirely, in order to make Emacs startup
faster.

@c ----------------------------------------------------------------------------
@node Keyword extensions
@appendix Keyword extensions

use-package is based on an extensible framework that makes it easy for
package authors to add new keywords, or modify the behavior of
existing keywords.

Some keyword extensions are included with @code{use-package}, and can
be optionally enabled.

@menu
* use-package-ensure-system-package::
* Creating an extension::
@end menu

@node use-package-ensure-system-package
@section :use-package-ensure-system-package

@findex :ensure-system-package
The @code{:ensure-system-package} keyword allows you to ensure certain
executables are available on your system alongside your package
declarations.@footnote{On macOS, you will want to make sure
@code{exec-path} is cognisant of all binary package names that you
would like to ensure are installed.  The
@uref{https://github.com/purcell/exec-path-from-shell,@samp{exec-path-from-shell}}
package is often a good way to do this.}

To use this extension, add this immediately after loading
@code{use-package}:

@lisp
(use-package use-package-ensure-system-package)
@end lisp

Now you can use the @code{:ensure-system-package} keyword.
Here's an example usage:

@lisp
(use-package foo
  :ensure-system-package foo)
@end lisp

This will expect a global binary package to exist called @code{foo}.
If it does not, it will use your system package manager to attempt an
install of a binary by the same name asynchronously.  This requires
the GNU ELPA package
@uref{https://gitlab.com/jabranham/system-packages,@samp{system-packages}},
so for this to work you must install that first.

One way of making sure it is installed is with @code{use-package}
together with @code{:ensure}.

@lisp
(use-package system-packages
  :ensure t)
@end lisp

For example, on a @code{Debian GNU/Linux} system, this would call
@samp{apt-get install foo}.

If the package is named differently than the binary, you can use a
cons in the form of @code{(binary . package-name)}.  For example:

@lisp
(use-package foo
  :ensure-system-package
  (foocmd . foo))
@end lisp

On a @code{Debian GNU/Linux} system, this would call @code{apt install
foo} if Emacs could not locate the executable
@code{foocmd}.@footnote{For manual testing, you could use the
@code{executable-find} function, which is what @samp{system-packages}
uses internally.}

@code{:ensure-system-package} can also take a cons where its
@code{cdr} is a string that will get called by
@code{(async-shell-command)} to install if it isn't found.  This does
not depend upon any external package.

@lisp
(use-package tern
  :ensure-system-package (tern . "npm i -g tern"))
@end lisp

To install several packages, you can pass in a list of conses:

@lisp
(use-package ruby-mode
  :ensure-system-package
  ((rubocop     . "gem install rubocop")
   (ruby-lint   . "gem install ruby-lint")
   (ripper-tags . "gem install ripper-tags")
   (pry         . "gem install pry")))
@end lisp

Finally, in case the package dependency does not provide a global
executable, you can ensure packages exist by checking the presence of a
file path by providing a string like so:

@lisp
(use-package dash-at-point
  :if (eq system-type 'darwin)
  :ensure-system-package
  ("/Applications/Dash.app" . "brew cask install dash"))
@end lisp

@code{:ensure-system-package} will use @code{system-packages-install}
to install system packages, except where a custom command has been
specified, in which case it will be executed verbatim by
@code{async-shell-command}.

The user options @code{system-packages-package-manager} and
@code{system-packages-use-sudo} are honored, but not for custom
commands.  Custom commands should include the call to sudo in the
command if needed.

@node Creating an extension
@section How to create an extension keyword

This section describes how to create a new keyword.

@enumerate
@item
Add the keyword.

The first step is to add your keyword at the right place in
@code{use-package-keywords}.  This list determines the order in which
things will happen in the expanded code.  You should never change this
order, but it gives you a framework within which to decide when your
keyword should fire.

@item
Create a normalizer.

The job of the normalizer is take a list of arguments (possibly
@code{nil}), and turn it into the single argument (which could still
be a list) that should appear in the final property list used by
@code{use-package}.

Define a normalizer for your keyword by defining a function named
after the keyword, for example:

@lisp
(defun use-package-normalize/:pin (name-symbol keyword args)
  (use-package-only-one (symbol-name keyword) args
    (lambda (label arg)
      (cond
       ((stringp arg) arg)
       ((symbolp arg) (symbol-name arg))
       (t
        (use-package-error
         ":pin wants an archive name (a string)"))))))
@end lisp

@item
Create a handler.

Once you have a normalizer, you must create a handler for the keyword.

Handlers can affect the handling of keywords in two ways.  First, it
can modify the @code{state} plist before recursively processing the
remaining keywords, to influence keywords that pay attention to the
state (one example is the state keyword @code{:deferred}, not to be
confused with the @code{use-package} keyword @code{:defer}).  Then,
once the remaining keywords have been handled and their resulting
forms returned, the handler may manipulate, extend, or just ignore
those forms.

The task of each handler is to return a @emph{list of forms}
representing code to be inserted.  It does not need to be a
@code{progn} list, as this is handled automatically in other places.
Thus it is common to see the idiom of using @code{use-package-concat}
to add new functionality before or after a code body, so that only the
minimum code necessary is emitted as the result of a
@code{use-package} expansion.

This is an example handler:

@lisp
(defun use-package-handler/:pin (name-symbol keyword archive-name rest state)
  (let ((body (use-package-process-keywords name-symbol rest state)))
    ;; This happens at macro expansion time, not when the expanded code is
    ;; compiled or evaluated.
    (if (null archive-name)
        body
      (use-package-pin-package name-symbol archive-name)
      (use-package-concat
       body
       `((push '(,name-symbol . ,archive-name)
               package-pinned-packages))))))
@end lisp

@item
Test it.

After the keyword has been inserted into @code{use-package-keywords},
and a normalizer and a handler defined, you can now test it by seeing
how usages of the keyword will expand.  For this, use @code{M-x
pp-macroexpand-last-sexp} with the cursor set immediately after the
@code{(use-package ...)} expression.
@end enumerate

@c ----------------------------------------------------------------------------
@node History
@appendix History and acknowledgments

use-package was written by John Wiegley.  Its development started in
2012, and it got merged into Emacs in 2022, in preparation of the
release of Emacs 29.1.

Dozens of people have contributed to use-package over the years with
bug reports, documentation and code.  They are too many to list here,
but we thank them all for their contributions.

This Texinfo manual was written by Stefan Kangas, as a significant
rewrite of the old use-package manual and @file{README}.

@node GNU Free Documentation License
@appendix GNU Free Documentation License
@include doclicense.texi

@node Index
@unnumbered Index
@printindex cp

@bye