This is magit.info, produced by makeinfo version 6.5 from magit.texi.
Copyright (C) 2015-2018 Jonas Bernoulli <jonas@bernoul.li>
You can redistribute this document and/or modify it under the terms
of the GNU General Public License as published by the Free Software
Foundation, either version 3 of the License, or (at your option)
any later version.
This document is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
General Public License for more details.
INFO-DIR-SECTION Emacs
START-INFO-DIR-ENTRY
* Magit: (magit). Using Git from Emacs with Magit.
END-INFO-DIR-ENTRY
File: magit.info, Node: Microsoft Windows Performance, Next: MacOS Performance, Up: Performance
Microsoft Windows Performance
.............................
In order to update the status buffer, ‘git’ has to be run a few dozen
times. That is problematic on Microsoft Windows, because that operating
system is exceptionally slow at starting processes. Sadly this is an
issue that can only be fixed by Microsoft itself, and they don’t appear
to be particularly interested in doing so.
Beside the subprocess issue, there are also other Windows-specific
performance issues. Some of these have workarounds. The maintainers of
"Git for Windows" try to improve performance on Windows. Always use the
latest release in order to benefit from the latest performance tweaks.
Magit too tries to work around some Windows-specific issues.
According to some sources, setting the following Git variables can
also help.
git config --global core.preloadindex true # default since v2.1
git config --global core.fscache true # default since v2.8
git config --global gc.auto 256
You should also check whether an anti-virus program is affecting
performance.
File: magit.info, Node: MacOS Performance, Prev: Microsoft Windows Performance, Up: Performance
MacOS Performance
.................
On macOS Emacs currently creates child processes using ‘fork’. It
appears that this also copies GUI resources. The result is that forking
takes about 30 times as long on Darwin than on Linux. And because Magit
starts many ‘git’ processes even when doing simple things, that makes
quite a difference.
On the ‘master’ branch Emacs now uses ‘vfork’ when possible, like
this was already done on Linux, and now child creation only takes about
twice as long on Darwin. See (1) for more information.
Nobody knows when the changes on the ‘master’ branch will be released
as ‘26.1’, but it is still a long way off. You might want to get your
hands on this change before then. The easiest way to get a patched
Emacs is to install the ‘emacs-plus’ formula (2) using ‘homebrew’. The
change has been backported, so you get it not only when using ‘--HEAD’,
but also when using ‘--devel’ or when installing the latest release (by
not using a version argument).
Alternatively you can apply the backport (3) manually.
---------- Footnotes ----------
(1)
<https://lists.gnu.org/archive/html/bug-gnu-emacs/2017-04/msg00201.html>
(2) <https://github.com/d12frosted/homebrew-emacs-plus>
(3)
<https://gist.githubusercontent.com/aaronjensen/f45894ddf431ecbff78b1bcf533d3e6b/raw/6a5cd7f57341aba673234348d8b0d2e776f86719/Emacs-25-OS-X-use-vfork.patch>
File: magit.info, Node: Plumbing, Next: FAQ, Prev: Customizing, Up: Top
10 Plumbing
***********
The following sections describe how to use several of Magit’s core
abstractions to extend Magit itself or implement a separate extension.
A few of the low-level features used by Magit have been factored out
into separate libraries/packages, so that they can be used by other
packages, without having to depend on Magit. These libraries are
described in separate manuals, see *note (with-editor)Top:: and *note
(magit-popup)Top::.
If you are trying to find an unused key that you can bind to a
command provided by your own Magit extension, then checkout
<https://github.com/magit/magit/wiki/Plugin-Dispatch-Key-Registry>.
* Menu:
* Calling Git::
* Section Plumbing::
* Refreshing Buffers::
* Conventions::
File: magit.info, Node: Calling Git, Next: Section Plumbing, Up: Plumbing
10.1 Calling Git
================
Magit provides many specialized functions for calling Git. All of these
functions are defined in either ‘magit-git.el’ or ‘magit-process.el’ and
have one of the prefixes ‘magit-run-’, ‘magit-call-’, ‘magit-start-’, or
‘magit-git-’ (which is also used for other things).
All of these functions accept an indefinite number of arguments,
which are strings that specify command line arguments for Git (or in
some cases an arbitrary executable). These arguments are flattened
before being passed on to the executable; so instead of strings they can
also be lists of strings and arguments that are ‘nil’ are silently
dropped. Some of these functions also require a single mandatory
argument before these command line arguments.
Roughly speaking, these functions run Git either to get some value or
for side-effects. The functions that return a value are useful to
collect the information necessary to populate a Magit buffer, while the
others are used to implement Magit commands.
The functions in the value-only group always run synchronously, and
they never trigger a refresh. The function in the side-effect group can
be further divided into subgroups depending on whether they run Git
synchronously or asynchronously, and depending on whether they trigger a
refresh when the executable has finished.
* Menu:
* Getting a Value from Git::
* Calling Git for Effect::
File: magit.info, Node: Getting a Value from Git, Next: Calling Git for Effect, Up: Calling Git
10.1.1 Getting a Value from Git
-------------------------------
These functions run Git in order to get a value, an exit status, or
output. Of course you could also use them to run Git commands that have
side-effects, but that should be avoided.
-- Function: magit-git-exit-code &rest args
Executes git with ARGS and returns its exit code.
-- Function: magit-git-success &rest args
Executes git with ARGS and returns ‘t’ if the exit code is ‘0’,
‘nil’ otherwise.
-- Function: magit-git-failure &rest args
Executes git with ARGS and returns ‘t’ if the exit code is ‘1’,
‘nil’ otherwise.
-- Function: magit-git-true &rest args
Executes git with ARGS and returns ‘t’ if the first line printed by
git is the string "true", ‘nil’ otherwise.
-- Function: magit-git-false &rest args
Executes git with ARGS and returns ‘t’ if the first line printed by
git is the string "false", ‘nil’ otherwise.
-- Function: magit-git-insert &rest args
Executes git with ARGS and inserts its output at point.
-- Function: magit-git-string &rest args
Executes git with ARGS and returns the first line of its output.
If there is no output or if it begins with a newline character,
then this returns ‘nil’.
-- Function: magit-git-lines &rest args
Executes git with ARGS and returns its output as a list of lines.
Empty lines anywhere in the output are omitted.
-- Function: magit-git-items &rest args
Executes git with ARGS and returns its null-separated output as a
list. Empty items anywhere in the output are omitted.
If the value of option ‘magit-git-debug’ is non-nil and git exits
with a non-zero exit status, then warn about that in the echo area
and add a section containing git’s standard error in the current
repository’s process buffer.
If an error occurs when using one of the above functions, then that
is usually due to a bug, i.e. using an argument which is not actually
supported. Such errors are usually not reported, but when they occur we
need to be able to debug them.
-- User Option: magit-git-debug
Whether to report errors that occur when using ‘magit-git-insert’,
‘magit-git-string’, ‘magit-git-lines’, or ‘magit-git-items’. This
does not actually raise an error. Instead a message is shown in
the echo area, and git’s standard error is insert into a new
section in the current repository’s process buffer.
-- Function: magit-git-str &rest args
This is a variant of ‘magit-git-string’ that ignores the option
‘magit-git-debug’. It is mainly intended to be used while handling
errors in functions that do respect that option. Using such a
function while handing an error could cause yet another error and
therefore lead to an infinite recursion. You probably won’t ever
need to use this function.
File: magit.info, Node: Calling Git for Effect, Prev: Getting a Value from Git, Up: Calling Git
10.1.2 Calling Git for Effect
-----------------------------
These functions are used to run git to produce some effect. Most Magit
commands that actually run git do so by using such a function.
Because we do not need to consume git’s output when using these
functions, their output is instead logged into a per-repository buffer,
which can be shown using ‘$’ from a Magit buffer or ‘M-x magit-process’
elsewhere.
These functions can have an effect in two distinct ways. Firstly,
running git may change something, i.e. create or push a new commit.
Secondly, that change may require that Magit buffers are refreshed to
reflect the changed state of the repository. But refreshing isn’t
always desirable, so only some of these functions do perform such a
refresh after git has returned.
Sometimes it is useful to run git asynchronously. For example, when
the user has just initiated a push, then there is no reason to make her
wait until that has completed. In other cases it makes sense to wait
for git to complete before letting the user do something else. For
example after staging a change it is useful to wait until after the
refresh because that also automatically moves to the next change.
-- Function: magit-call-git &rest args
Calls git synchronously with ARGS.
-- Function: magit-call-process program &rest args
Calls PROGRAM synchronously with ARGS.
-- Function: magit-run-git &rest args
Calls git synchronously with ARGS and then refreshes.
-- Function: magit-run-git-with-input input &rest args
Calls git synchronously with ARGS and sends it INPUT on standard
input.
INPUT should be a buffer or the name of an existing buffer. The
content of that buffer is used as the process’ standard input.
After the process returns a refresh is performed.
As a special case, INPUT may also be nil. In that case the content
of the current buffer is used as standard input and *no* refresh is
performed.
This function actually runs git asynchronously. But then it waits
for the process to return, so the function itself is synchronous.
-- Function: magit-run-git-with-logfile file &rest args
Calls git synchronously with ARGS. The process’ output is saved in
FILE. This is rarely useful and so this function might be removed
in the future.
This function actually runs git asynchronously. But then it waits
for the process to return, so the function itself is synchronous.
-- Function: magit-git &rest args
Calls git synchronously with ARGS for side-effects only. This
function does not refresh the buffer.
-- Function: magit-git-wash washer &rest args
Execute Git with ARGS, inserting washed output at point. Actually
first insert the raw output at point. If there is no output call
‘magit-cancel-section’. Otherwise temporarily narrow the buffer to
the inserted text, move to its beginning, and then call function
WASHER with ARGS as its sole argument.
And now for the asynchronous variants.
-- Function: magit-run-git-async &rest args
Start Git, prepare for refresh, and return the process object.
ARGS is flattened and then used as arguments to Git.
Display the command line arguments in the echo area.
After Git returns some buffers are refreshed: the buffer that was
current when this function was called (if it is a Magit buffer and
still alive), as well as the respective Magit status buffer.
Unmodified buffers visiting files that are tracked in the current
repository are reverted if ‘magit-revert-buffers’ is non-nil.
-- Function: magit-run-git-with-editor &rest args
Export GIT_EDITOR and start Git. Also prepare for refresh and
return the process object. ARGS is flattened and then used as
arguments to Git.
Display the command line arguments in the echo area.
After Git returns some buffers are refreshed: the buffer that was
current when this function was called (if it is a Magit buffer and
still alive), as well as the respective Magit status buffer.
-- Function: magit-start-git &rest args
Start Git, prepare for refresh, and return the process object.
If INPUT is non-nil, it has to be a buffer or the name of an
existing buffer. The buffer content becomes the processes standard
input.
Option ‘magit-git-executable’ specifies the Git executable and
option ‘magit-git-global-arguments’ specifies constant arguments.
The remaining arguments ARGS specify arguments to Git. They are
flattened before use.
After Git returns, some buffers are refreshed: the buffer that was
current when this function was called (if it is a Magit buffer and
still alive), as well as the respective Magit status buffer.
Unmodified buffers visiting files that are tracked in the current
repository are reverted if ‘magit-revert-buffers’ is non-nil.
-- Function: magit-start-process &rest args
Start PROGRAM, prepare for refresh, and return the process object.
If optional argument INPUT is non-nil, it has to be a buffer or the
name of an existing buffer. The buffer content becomes the
processes standard input.
The process is started using ‘start-file-process’ and then setup to
use the sentinel ‘magit-process-sentinel’ and the filter
‘magit-process-filter’. Information required by these functions is
stored in the process object. When this function returns the
process has not started to run yet so it is possible to override
the sentinel and filter.
After the process returns, ‘magit-process-sentinel’ refreshes the
buffer that was current when ‘magit-start-process’ was called (if
it is a Magit buffer and still alive), as well as the respective
Magit status buffer. Unmodified buffers visiting files that are
tracked in the current repository are reverted if
‘magit-revert-buffers’ is non-nil.
-- Variable: magit-this-process
The child process which is about to start. This can be used to
change the filter and sentinel.
-- Variable: magit-process-raise-error
When this is non-nil, then ‘magit-process-sentinel’ raises an error
if git exits with a non-zero exit status. For debugging purposes.
File: magit.info, Node: Section Plumbing, Next: Refreshing Buffers, Prev: Calling Git, Up: Plumbing
10.2 Section Plumbing
=====================
* Menu:
* Creating Sections::
* Section Selection::
* Matching Sections::
File: magit.info, Node: Creating Sections, Next: Section Selection, Up: Section Plumbing
10.2.1 Creating Sections
------------------------
-- Macro: magit-insert-section &rest args
Insert a section at point.
TYPE is the section type, a symbol. Many commands that act on the
current section behave differently depending on that type. Also if
a variable ‘magit-TYPE-section-map’ exists, then use that as the
text-property ‘keymap’ of all text belonging to the section (but
this may be overwritten in subsections). TYPE can also have the
form ‘(eval FORM)’ in which case FORM is evaluated at runtime.
Optional VALUE is the value of the section, usually a string that
is required when acting on the section.
When optional HIDE is non-nil collapse the section body by default,
i.e. when first creating the section, but not when refreshing the
buffer. Otherwise, expand it by default. This can be overwritten
using ‘magit-section-set-visibility-hook’. When a section is
recreated during a refresh, then the visibility of predecessor is
inherited and HIDE is ignored (but the hook is still honored).
BODY is any number of forms that actually insert the section’s
heading and body. Optional NAME, if specified, has to be a symbol,
which is then bound to the struct of the section being inserted.
Before BODY is evaluated the ‘start’ of the section object is set
to the value of ‘point’ and after BODY was evaluated its ‘end’ is
set to the new value of ‘point’; BODY is responsible for moving
‘point’ forward.
If it turns out inside BODY that the section is empty, then
‘magit-cancel-section’ can be used to abort and remove all traces
of the partially inserted section. This can happen when creating a
section by washing Git’s output and Git didn’t actually output
anything this time around.
-- Function: magit-insert-heading &rest args
Insert the heading for the section currently being inserted.
This function should only be used inside ‘magit-insert-section’.
When called without any arguments, then just set the ‘content’ slot
of the object representing the section being inserted to a marker
at ‘point’. The section should only contain a single line when
this function is used like this.
When called with arguments ARGS, which have to be strings, then
insert those strings at point. The section should not contain any
text before this happens and afterwards it should again only
contain a single line. If the ‘face’ property is set anywhere
inside any of these strings, then insert all of them unchanged.
Otherwise use the ‘magit-section-heading’ face for all inserted
text.
The ‘content’ property of the section struct is the end of the
heading (which lasts from ‘start’ to ‘content’) and the beginning
of the body (which lasts from ‘content’ to ‘end’). If the value of
‘content’ is nil, then the section has no heading and its body
cannot be collapsed. If a section does have a heading then its
height must be exactly one line, including a trailing newline
character. This isn’t enforced; you are responsible for getting it
right. The only exception is that this function does insert a
newline character if necessary.
-- Function: magit-cancel-section
Cancel the section currently being inserted. This exits the
innermost call to ‘magit-insert-section’ and removes all traces of
what has already happened inside that call.
-- Function: magit-define-section-jumper sym title &optional value
Define an interactive function to go to section SYM. TITLE is the
displayed title of the section.
File: magit.info, Node: Section Selection, Next: Matching Sections, Prev: Creating Sections, Up: Section Plumbing
10.2.2 Section Selection
------------------------
-- Function: magit-current-section
Return the section at point.
-- Function: magit-region-sections
Return a list of the selected sections.
When the region is active and constitutes a valid section
selection, then return a list of all selected sections. This is
the case when the region begins in the heading of a section and
ends in the heading of a sibling of that first section. When the
selection is not valid then return nil. Most commands that can act
on the selected sections, then instead just act on the current
section, the one point is in.
When the region looks like it would in any other buffer then the
selection is invalid. When the selection is valid then the region
uses the ‘magit-section-highlight’. This does not apply to diffs
where things get a bit more complicated, but even here if the
region looks like it usually does, then that’s not a valid
selection as far as this function is concerned.
-- Function: magit-region-values &rest types
Return a list of the values of the selected sections.
Also see ‘magit-region-sections’ whose doc-string explains when a
region is a valid section selection. If the region is not active
or is not a valid section selection, then return nil. If optional
TYPES is non-nil then the selection not only has to be valid; the
types of all selected sections additionally have to match one of
TYPES, or nil is returned.
File: magit.info, Node: Matching Sections, Prev: Section Selection, Up: Section Plumbing
10.2.3 Matching Sections
------------------------
‘M-x magit-describe-section-briefly’ (‘magit-describe-section-briefly’)
Show information about the section at point. This command is
intended for debugging purposes.
-- Function: magit-section-ident
Return an unique identifier for SECTION. The return value has the
form ‘((TYPE . VALUE)...)’.
-- Function: magit-get-section
Return the section identified by IDENT. IDENT has to be a list as
returned by ‘magit-section-ident’.
-- Function: magit-section-match condition &optional section
Return ‘t’ if SECTION matches CONDITION. SECTION defaults to the
section at point. If SECTION is not specified and there also is no
section at point, then return ‘nil’.
CONDITION can take the following forms:
• ‘(CONDITION...)’
matches if any of the CONDITIONs matches.
• ‘[TYPE...]’
matches if the first TYPE matches the type of the section, the
second matches that of its parent, and so on.
• ‘[* TYPE...]’
matches sections that match [TYPE...] and also recursively all
their child sections.
• ‘TYPE’
matches sections of TYPE regardless of the types of the parent
sections.
Each TYPE is a symbol. Note that it is not necessary to specify
all TYPEs up to the root section as printed by
‘magit-describe-type’, unless of course you want to be that
precise.
-- Function: magit-section-when condition &rest body
If the section at point matches CONDITION evaluate BODY.
If the section matches, then evaluate BODY forms sequentially with
‘it’ bound to the section and return the value of the last form.
If there are no BODY forms, then return the value of the section.
If the section does not match or if there is no section at point
then return nil.
See ‘magit-section-match’ for the forms CONDITION can take.
-- Function: magit-section-case &rest clauses
Choose among clauses on the type of the section at point.
Each clause looks like (CONDITION BODY...). The type of the
section is compared against each CONDITION; the BODY forms of the
first match are evaluated sequentially and the value of the last
form is returned. Inside BODY the symbol ‘it’ is bound to the
section at point. If no clause succeeds or if there is no section
at point return nil.
See ‘magit-section-match’ for the forms CONDITION can take.
Additionally a CONDITION of t is allowed in the final clause and
matches if no other CONDITION match, even if there is no section at
point.
-- Variable: magit-root-section
The root section in the current buffer. All other sections are
descendants of this section. The value of this variable is set by
‘magit-insert-section’ and you should never modify it.
For diff related sections a few additional tools exist.
-- Function: magit-diff-type &optional section
Return the diff type of SECTION.
The returned type is one of the symbols ‘staged’, ‘unstaged’,
‘committed’, or ‘undefined’. This type serves a similar purpose as
the general type common to all sections (which is stored in the
‘type’ slot of the corresponding ‘magit-section’ struct) but takes
additional information into account. When the SECTION isn’t
related to diffs and the buffer containing it also isn’t a
diff-only buffer, then return nil.
Currently the type can also be one of ‘tracked’ and ‘untracked’,
but these values are not handled explicitly in every place they
should be. A possible fix could be to just return nil here.
The section has to be a ‘diff’ or ‘hunk’ section, or a section
whose children are of type ‘diff’. If optional SECTION is nil,
return the diff type for the current section. In buffers whose
major mode is ‘magit-diff-mode’ SECTION is ignored and the type is
determined using other means. In ‘magit-revision-mode’ buffers the
type is always ‘committed’.
-- Function: magit-diff-scope &optional section strict
Return the diff scope of SECTION or the selected section(s).
A diff’s "scope" describes what part of a diff is selected, it is a
symbol, one of ‘region’, ‘hunk’, ‘hunks’, ‘file’, ‘files’, or
‘list’. Do not confuse this with the diff "type", as returned by
‘magit-diff-type’.
If optional SECTION is non-nil, then return the scope of that,
ignoring the sections selected by the region. Otherwise return the
scope of the current section, or if the region is active and
selects a valid group of diff related sections, the type of these
sections, i.e. ‘hunks’ or ‘files’. If SECTION (or if the current
section that is nil) is a ‘hunk’ section and the region starts and
ends inside the body of a that section, then the type is ‘region’.
If optional STRICT is non-nil then return nil if the diff type of
the section at point is ‘untracked’ or the section at point is not
actually a ‘diff’ but a ‘diffstat’ section.
File: magit.info, Node: Refreshing Buffers, Next: Conventions, Prev: Section Plumbing, Up: Plumbing
10.3 Refreshing Buffers
=======================
All commands that create a new Magit buffer or change what is being
displayed in an existing buffer do so by calling ‘magit-mode-setup’.
Among other things, that function sets the buffer local values of
‘default-directory’ (to the top-level of the repository),
‘magit-refresh-function’, and ‘magit-refresh-args’.
Buffers are refreshed by calling the function that is the local value
of ‘magit-refresh-function’ (a function named ‘magit-*-refresh-buffer’,
where ‘*’ may be something like ‘diff’) with the value of
‘magit-refresh-args’ as arguments.
-- Macro: magit-mode-setup buffer switch-func mode refresh-func
&optional refresh-args
This function displays and selects BUFFER, turns on MODE, and
refreshes a first time.
This function displays and optionally selects BUFFER by calling
‘magit-mode-display-buffer’ with BUFFER, MODE and SWITCH-FUNC as
arguments. Then it sets the local value of
‘magit-refresh-function’ to REFRESH-FUNC and that of
‘magit-refresh-args’ to REFRESH-ARGS. Finally it creates the
buffer content by calling REFRESH-FUNC with REFRESH-ARGS as
arguments.
All arguments are evaluated before switching to BUFFER.
-- Function: magit-mode-display-buffer buffer mode &optional
switch-function
This function display BUFFER in some window and select it. BUFFER
may be a buffer or a string, the name of a buffer. The buffer is
returned.
Unless BUFFER is already displayed in the selected frame, store the
previous window configuration as a buffer local value, so that it
can later be restored by ‘magit-mode-bury-buffer’.
The buffer is displayed and selected using SWITCH-FUNCTION. If
that is ‘nil’ then ‘pop-to-buffer’ is used if the current buffer’s
major mode derives from ‘magit-mode’. Otherwise ‘switch-to-buffer’
is used.
-- Variable: magit-refresh-function
The value of this buffer-local variable is the function used to
refresh the current buffer. It is called with ‘magit-refresh-args’
as arguments.
-- Variable: magit-refresh-args
The list of arguments used by ‘magit-refresh-function’ to refresh
the current buffer. ‘magit-refresh-function’ is called with these
arguments.
The value is usually set using ‘magit-mode-setup’, but in some
cases it’s also useful to provide commands which can change the
value. For example, the ‘magit-diff-refresh-popup’ can be used to
change any of the arguments used to display the diff, without
having to specify again which differences should be shown.
‘magit-diff-more-context’, ‘magit-diff-less-context’, and
‘magit-diff-default-context’ change just the ‘-U<N>’ argument. In
both case this is done by changing the value of this variable and
then calling this ‘magit-refresh-function’.
File: magit.info, Node: Conventions, Prev: Refreshing Buffers, Up: Plumbing
10.4 Conventions
================
Also see *note Completion and Confirmation::.
* Menu:
* Theming Faces::
File: magit.info, Node: Theming Faces, Up: Conventions
10.4.1 Theming Faces
--------------------
The default theme uses blue for local branches, green for remote
branches, and goldenrod (brownish yellow) for tags. When creating a new
theme, you should probably follow that example. If your theme already
uses other colors, then stick to that.
In older releases these reference faces used to have a background
color and a box around them. The basic default faces no longer do so,
to make Magit buffers much less noisy, and you should follow that
example at least with regards to boxes. (Boxes were used in the past to
work around a conflict between the highlighting overlay and text
property backgrounds. That’s no longer necessary because highlighting
no longer causes other background colors to disappear.) Alternatively
you can keep the background color and/or box, but then have to take
special care to adjust ‘magit-branch-current’ accordingly. By default
it looks mostly like ‘magit-branch-local’, but with a box (by default
the former is the only face that uses a box, exactly so that it sticks
out). If the former also uses a box, then you have to make sure that it
differs in some other way from the latter.
The most difficult faces to theme are those related to diffs,
headings, highlighting, and the region. There are faces that fall into
all four groups - expect to spend some time getting this right.
The ‘region’ face in the default theme, in both the light and dark
variants, as well as in many other themes, distributed with Emacs or by
third-parties, is very ugly. It is common to use a background color
that really sticks out, which is ugly but if that were the only problem
then it would be acceptable. Unfortunately many themes also set the
foreground color, which ensures that all text within the region is
readable. Without doing that there might be cases where some foreground
color is too close to the region background color to still be readable.
But it also means that text within the region loses all syntax
highlighting.
I consider the work that went into getting the ‘region’ face right to
be a good indicator for the general quality of a theme. My
recommendation for the ‘region’ face is this: use a background color
slightly different from the background color of the ‘default’ face, and
do not set the foreground color at all. So for a light theme you might
use a light (possibly tinted) gray as the background color of ‘default’
and a somewhat darker gray for the background of ‘region’. That should
usually be enough to not collide with the foreground color of any other
face. But if some other faces also set a light gray as background
color, then you should also make sure it doesn’t collide with those (in
some cases it might be acceptable though).
Magit only uses the ‘region’ face when the region is "invalid" by its
own definition. In a Magit buffer the region is used to either select
multiple sibling sections, so that commands which support it act on all
of these sections instead of just the current section, or to select
lines within a single hunk section. In all other cases, the section is
considered invalid and Magit won’t act on it. But such invalid sections
happen, either because the user has not moved point enough yet to make
it valid or because she wants to use a non-magit command to act on the
region, e.g. ‘kill-region’.
So using the regular ‘region’ face for invalid sections is a feature.
It tells the user that Magit won’t be able to act on it. It’s
acceptable if that face looks a bit odd and even (but less so) if it
collides with the background colors of section headings and other things
that have a background color.
Magit highlights the current section. If a section has subsections,
then all of them are highlighted. This is done using faces that have
"highlight" in their names. For most sections,
‘magit-section-highlight’ is used for both the body and the heading.
Like the ‘region’ face, it should only set the background color to
something similar to that of ‘default’. The highlight background color
must be different from both the ‘region’ background color and the
‘default’ background color.
For diff related sections Magit uses various faces to highlight
different parts of the selected section(s). Note that hunk headings,
unlike all other section headings, by default have a background color,
because it is useful to have very visible separators between hunks.
That face ‘magit-diff-hunk-heading’, should be different from both
‘magit-diff-hunk-heading-highlight’ and ‘magit-section-highlight’, as
well as from ‘magit-diff-context’ and ‘magit-diff-context-highlight’.
By default we do that by changing the foreground color. Changing the
background color would lead to complications, and there are already
enough we cannot get around. (Also note that it is generally a good
idea for section headings to always be bold, but only for sections that
have subsections).
When there is a valid region selecting diff-related sibling sections,
i.e. multiple files or hunks, then the bodies of all these sections use
the respective highlight faces, but additionally the headings instead
use one of the faces ‘magit-diff-file-heading-selection’ or
‘magit-diff-hunk-heading-selection’. These faces have to be different
from the regular highlight variants to provide explicit visual
indication that the region is active.
When theming diff related faces, start by setting the option
‘magit-diff-refine-hunk’ to ‘all’. You might personally prefer to only
refine the current hunk or not use hunk refinement at all, but some of
the users of your theme want all hunks to be refined, so you have to
cater to that.
(Also turn on ‘magit-diff-highlight-indentation’,
‘magit-diff-highlight-trailing’, and ‘magit-diff-paint-whitespace’; and
insert some whitespace errors into the code you use for testing.)
For e.g. "added lines" you have to adjust three faces:
‘magit-diff-added’, ‘magit-diff-added-highlight’, and
‘smerge-refined-added’. Make sure that the latter works well with both
of the former, as well as ‘smerge-other’ and ‘diff-added’. Then do the
same for the removed lines, context lines, lines added by us, and lines
added by them. Also make sure the respective added, removed, and
context faces use approximately the same saturation for both the
highlighted and unhighlighted variants. Also make sure the file and
diff headings work nicely with context lines (e.g. make them look
different). Line faces should set both the foreground and the
background color. For example, for added lines use two different
greens.
It’s best if the foreground color of both the highlighted and the
unhighlighted variants are the same, so you will need to have to find a
color that works well on the highlight and unhighlighted background, the
refine background, and the highlight context background. When there is
an hunk internal region, then the added- and removed-lines background
color is used only within that region. Outside the region the
highlighted context background color is used. This makes it easier to
see what is being staged. With an hunk internal region the hunk heading
is shown using ‘magit-diff-hunk-heading-selection’, and so are the thin
lines that are added around the lines that fall within the region. The
background color of that has to be distinct enough from the various
other involved background colors.
Nobody said this would be easy. If your theme restricts itself to a
certain set of colors, then you should make an exception here.
Otherwise it would be impossible to make the diffs look good in each and
every variation. Actually you might want to just stick to the default
definitions for these faces. You have been warned. Also please note
that if you do not get this right, this will in some cases look to users
like bugs in Magit - so please do it right or not at all.
File: magit.info, Node: FAQ, Next: Debugging Tools, Prev: Plumbing, Up: Top
Appendix A FAQ
**************
The next two nodes lists frequently asked questions. For a list of
frequently *and recently* asked questions, i.e. questions that haven’t
made it into the manual yet, see
<https://github.com/magit/magit/wiki/FAQ>.
Please also use the *note Debugging Tools::.
* Menu:
* FAQ - How to ...?::
* FAQ - Issues and Errors::
File: magit.info, Node: FAQ - How to ...?, Next: FAQ - Issues and Errors, Up: FAQ
A.1 FAQ - How to ...?
=====================
* Menu:
* How to show git's output?::
* How to install the gitman info manual?::
* How to show diffs for gpg-encrypted files?::
* How does branching and pushing work?::
* Can Magit be used as ediff-version-control-package?::
File: magit.info, Node: How to show git's output?, Next: How to install the gitman info manual?, Up: FAQ - How to ...?
A.1.1 How to show git’s output?
-------------------------------
To show the output of recently run git commands, press ‘$’ (or, if that
isn’t available, ‘M-x magit-process-buffer’). This will show a buffer
containing a section per git invocation; as always press ‘TAB’ to expand
or collapse them.
By default, git’s output is only inserted into the process buffer if
it is run for side-effects. When the output is consumed in some way,
also inserting it into the process buffer would be too expensive. For
debugging purposes, it’s possible to do so anyway by setting
‘magit-git-debug’ to ‘t’.
File: magit.info, Node: How to install the gitman info manual?, Next: How to show diffs for gpg-encrypted files?, Prev: How to show git's output?, Up: FAQ - How to ...?
A.1.2 How to install the gitman info manual?
--------------------------------------------
Git’s manpages can be exported as an info manual called ‘gitman’.
Magit’s own info manual links to nodes in that manual instead of the
actual manpages because Info doesn’t support linking to manpages.
Unfortunately some distributions do not install the ‘gitman’ manual
by default and you will have to install a separate documentation package
to get it.
Magit patches Info adding the ability to visit links to the ‘gitman’
Info manual by instead viewing the respective manpage. If you prefer
that approach, then set the value of ‘magit-view-git-manual-method’ to
one of the supported packages ‘man’ or ‘woman’, e.g.:
(setq magit-view-git-manual-method 'man)
File: magit.info, Node: How to show diffs for gpg-encrypted files?, Next: How does branching and pushing work?, Prev: How to install the gitman info manual?, Up: FAQ - How to ...?
A.1.3 How to show diffs for gpg-encrypted files?
------------------------------------------------
Git supports showing diffs for encrypted files, but has to be told to do
so. Since Magit just uses Git to get the diffs, configuring Git also
affects the diffs displayed inside Magit.
git config --global diff.gpg.textconv "gpg --no-tty --decrypt"
echo "*.gpg filter=gpg diff=gpg" > .gitattributes
File: magit.info, Node: How does branching and pushing work?, Next: Can Magit be used as ediff-version-control-package?, Prev: How to show diffs for gpg-encrypted files?, Up: FAQ - How to ...?
A.1.4 How does branching and pushing work?
------------------------------------------
Please see *note Branching:: and
<http://emacsair.me/2016/01/17/magit-2.4>
File: magit.info, Node: Can Magit be used as ediff-version-control-package?, Prev: How does branching and pushing work?, Up: FAQ - How to ...?
A.1.5 Can Magit be used as ‘ediff-version-control-package’?
-----------------------------------------------------------
No, it cannot. For that to work the functions ‘ediff-magit-internal’
and ‘ediff-magit-merge-internal’ would have to be implemented, and they
are not. These two functions are only used by the three commands
‘ediff-revision’, ‘ediff-merge-revisions-with-ancestor’, and
‘ediff-merge-revisions’.
These commands only delegate the task of populating buffers with
certain revisions to the "internal" functions. The equally important
task of determining which revisions are to be compared/merged is not
delegated. Instead this is done without any support whatsoever from the
version control package/system - meaning that the user has to enter the
revisions explicitly. Instead of implementing ‘ediff-magit-internal’ we
provide ‘magit-ediff-compare’, which handles both tasks like it is 2005.
The other commands ‘ediff-merge-revisions’ and
‘ediff-merge-revisions-with-ancestor’ are normally not what you want
when using a modern version control system like Git. Instead of letting
the user resolve only those conflicts which Git could not resolve on its
own, they throw away all work done by Git and then expect the user to
manually merge all conflicts, including those that had already been
resolved. That made sense back in the days when version control systems
couldn’t merge (or so I have been told), but not anymore. Once in a
blue moon you might actually want to see all conflicts, in which case
you *can* use these commands, which then use ‘ediff-vc-merge-internal’.
So we don’t actually have to implement ‘ediff-magit-merge-internal’.
Instead we provide the more useful command ‘magit-ediff-resolve’ which
only shows yet-to-be resolved conflicts.
File: magit.info, Node: FAQ - Issues and Errors, Prev: FAQ - How to ...?, Up: FAQ
A.2 FAQ - Issues and Errors
===========================
* Menu:
* Magit is slow::
* I changed several thousand files at once and now Magit is unusable::
* I am having problems committing::
* I am using MS Windows and cannot push with Magit::
* I am using OS X and SOMETHING works in shell, but not in Magit: I am using OS X and SOMETHING works in shell but not in Magit.
* Diffs contain control sequences::
* Expanding a file to show the diff causes it to disappear::
* Point is wrong in the COMMIT_EDITMSG buffer::
* The mode-line information isn't always up-to-date::
* A branch and tag sharing the same name breaks SOMETHING::
* My Git hooks work on the command-line but not inside Magit::
* git-commit-mode isn't used when committing from the command-line::
* Point ends up inside invisible text when jumping to a file-visiting buffer::
File: magit.info, Node: Magit is slow, Next: I changed several thousand files at once and now Magit is unusable, Up: FAQ - Issues and Errors
A.2.1 Magit is slow
-------------------
See *note Performance::.
File: magit.info, Node: I changed several thousand files at once and now Magit is unusable, Next: I am having problems committing, Prev: Magit is slow, Up: FAQ - Issues and Errors
A.2.2 I changed several thousand files at once and now Magit is unusable
------------------------------------------------------------------------
Magit is *currently* not expected to work under such conditions. It
sure would be nice if it did, and v2.5 will hopefully be a big step into
that direction. But it might take until v3.1 to accomplish fully
satisfactory performance, because that requires some heavy refactoring.
But for now we recommend you use the command line to complete this
one commit. Also see *note Performance::.
File: magit.info, Node: I am having problems committing, Next: I am using MS Windows and cannot push with Magit, Prev: I changed several thousand files at once and now Magit is unusable, Up: FAQ - Issues and Errors
A.2.3 I am having problems committing
-------------------------------------
That likely means that Magit is having problems finding an appropriate
emacsclient executable. See *note (with-editor)Configuring
With-Editor:: and *note (with-editor)Debugging::.
File: magit.info, Node: I am using MS Windows and cannot push with Magit, Next: I am using OS X and SOMETHING works in shell but not in Magit, Prev: I am having problems committing, Up: FAQ - Issues and Errors
A.2.4 I am using MS Windows and cannot push with Magit
------------------------------------------------------
It’s almost certain that Magit is only incidental to this issue. It is
much more likely that this is a configuration issue, even if you can
push on the command line.
Detailed setup instructions can be found at
<https://github.com/magit/magit/wiki/Pushing-with-Magit-from-Windows>.
File: magit.info, Node: I am using OS X and SOMETHING works in shell but not in Magit, Next: Diffs contain control sequences, Prev: I am using MS Windows and cannot push with Magit, Up: FAQ - Issues and Errors
A.2.5 I am using OS X and SOMETHING works in shell, but not in Magit
--------------------------------------------------------------------
This usually occurs because Emacs doesn’t have the same environment
variables as your shell. Try installing and configuring
<https://github.com/purcell/exec-path-from-shell>. By default it
synchronizes ‘$PATH’, which helps Magit find the same ‘git’ as the one
you are using on the shell.
If SOMETHING is "passphrase caching with gpg-agent for commit and/or
tag signing", then you’ll also need to synchronize ‘$GPG_AGENT_INFO’.
File: magit.info, Node: Diffs contain control sequences, Next: Expanding a file to show the diff causes it to disappear, Prev: I am using OS X and SOMETHING works in shell but not in Magit, Up: FAQ - Issues and Errors
A.2.6 Diffs contain control sequences
-------------------------------------
This happens when you configure Git to always color diffs and/or all of
its output. The valid values for relevant Git variables ‘color.ui’ and
‘color.diff’ are ‘false’, ‘true’ and ‘always’, and the default is
‘true’. You should leave it that way because then you get colorful
output in terminals by default but when git’s output is consumed by
something else, then no color control sequences are used.
If you actually use some other tool that requires setting ‘color.ui’
and/or ‘color.diff’ to ‘always’ (which is highly unlikely), then you can
override these settings just for Magit by using:
(setq magit-git-global-arguments
(nconc magit-git-global-arguments
'("-c" "color.ui=false"
"-c" "color.diff=false")))
File: magit.info, Node: Expanding a file to show the diff causes it to disappear, Next: Point is wrong in the COMMIT_EDITMSG buffer, Prev: Diffs contain control sequences, Up: FAQ - Issues and Errors
A.2.7 Expanding a file to show the diff causes it to disappear
--------------------------------------------------------------
This is probably caused by a change of a ‘diff.*’ Git variable. You
probably set that variable for a reason, and should therefore only undo
that setting in Magit by customizing ‘magit-git-global-arguments’.
File: magit.info, Node: Point is wrong in the COMMIT_EDITMSG buffer, Next: The mode-line information isn't always up-to-date, Prev: Expanding a file to show the diff causes it to disappear, Up: FAQ - Issues and Errors
A.2.8 Point is wrong in the ‘COMMIT_EDITMSG’ buffer
---------------------------------------------------
Neither Magit nor ‘git-commit‘ fiddle with point in the buffer used to
write commit messages, so something else must be doing it.
You have probably globally enabled a mode which does restore point in
file-visiting buffers. It might be a bit surprising, but when you write
a commit message, then you are actually editing a file.
So you have to figure out which package is doing. ‘saveplace’,
‘pointback’, and ‘session’ are likely candidates. These snippets might
help:
(setq session-name-disable-regexp "\\(?:\\`'\\.git/[A-Z_]+\\'\\)")
(with-eval-after-load 'pointback
(lambda ()
(when (or git-commit-mode git-rebase-mode)
(pointback-mode -1))))
File: magit.info, Node: The mode-line information isn't always up-to-date, Next: A branch and tag sharing the same name breaks SOMETHING, Prev: Point is wrong in the COMMIT_EDITMSG buffer, Up: FAQ - Issues and Errors
A.2.9 The mode-line information isn’t always up-to-date
-------------------------------------------------------
Magit is not responsible for the version control information that is
being displayed in the mode-line and looks something like ‘Git-master’.
The built-in "Version Control" package, also known as "VC", updates that
information, and can be told to do so more often:
(setq auto-revert-check-vc-info t)
But doing so isn’t good for performance. For more (overly
optimistic) information see *note (emacs)VC Mode Line::.
If you don’t really care about seeing that information in the
mode-line, but just don’t want to see _incorrect_ information, then
consider disabling VC when using Git:
(setq vc-handled-backends (delq 'Git vc-handled-backends))
Or to disable it completely:
(setq vc-handled-backends nil)
File: magit.info, Node: A branch and tag sharing the same name breaks SOMETHING, Next: My Git hooks work on the command-line but not inside Magit, Prev: The mode-line information isn't always up-to-date, Up: FAQ - Issues and Errors
A.2.10 A branch and tag sharing the same name breaks SOMETHING
--------------------------------------------------------------
Or more generally, ambiguous refnames break SOMETHING.
Magit assumes that refs are named non-ambiguously across the
"refs/heads/", "refs/tags/", and "refs/remotes/" namespaces (i.e., all
the names remain unique when those prefixes are stripped). We consider
ambiguous refnames unsupported and recommend that you use a
non-ambiguous naming scheme. However, if you do work with a repository
that has ambiguous refnames, please report any issues you encounter so
that we can investigate whether there is a simple fix.
File: magit.info, Node: My Git hooks work on the command-line but not inside Magit, Next: git-commit-mode isn't used when committing from the command-line, Prev: A branch and tag sharing the same name breaks SOMETHING, Up: FAQ - Issues and Errors
A.2.11 My Git hooks work on the command-line but not inside Magit
-----------------------------------------------------------------
When Magit calls ‘git’ it adds a few global arguments including
‘--literal-pathspecs’ and the ‘git’ process started by Magit then passes
that setting on to other ‘git’ process it starts itself. It does so by
setting the environment variable ‘GIT_LITERAL_PATHSPECS’, not by calling
subprocesses with the ‘--literal-pathspecs’. You can therefore override
this setting in hook scripts using ‘unset GIT_LITERAL_PATHSPECS’.
File: magit.info, Node: git-commit-mode isn't used when committing from the command-line, Next: Point ends up inside invisible text when jumping to a file-visiting buffer, Prev: My Git hooks work on the command-line but not inside Magit, Up: FAQ - Issues and Errors
A.2.12 ‘git-commit-mode’ isn’t used when committing from the command-line
-------------------------------------------------------------------------
The reason for this is that ‘git-commit.el’ has not been loaded yet
and/or that the server has not been started yet. These things have
always already been taken care of when you commit from Magit because in
order to do so, Magit has to be loaded and doing that involves loading
‘git-commit’ and starting the server.
If you want to commit from the command-line, then you have to take
care of these things yourself. Your ‘init.el’ file should contain:
(require 'git-commit)
(server-mode)
Instead of ‘(require ’git-commit)‘ you may also use:
(load "/path/to/magit-autoloads.el")
You might want to do that because loading ‘git-commit’ causes large
parts of Magit to be loaded.
There are also some variations of ‘(server-mode)’ that you might want
to try. Personally I use:
(use-package server
:config (or (server-running-p) (server-mode)))
Now you can use:
$ emacs&
$ EDITOR=emacsclient git commit
However you cannot use:
$ killall emacs
$ EDITOR="emacsclient --alternate-editor emacs" git commit
This will actually end up using ‘emacs’, not ‘emacsclient’. If you
do this, then can still edit the commit message but ‘git-commit-mode’
won’t be used and you have to exit ‘emacs’ to finish the process.
Tautology ahead. If you want to be able to use ‘emacsclient’ to
connect to a running ‘emacs’ instance, even though no ‘emacs’ instance
is running, then you cannot use ‘emacsclient’ directly.
Instead you have to create a script that does something like this:
Try to use ‘emacsclient’ (without using ‘--alternate-editor’). If
that succeeds, do nothing else. Otherwise start ‘emacs &’ (and
‘init.el’ must call ‘server-start’) and try to use ‘emacsclient’ again.
File: magit.info, Node: Point ends up inside invisible text when jumping to a file-visiting buffer, Prev: git-commit-mode isn't used when committing from the command-line, Up: FAQ - Issues and Errors
A.2.13 Point ends up inside invisible text when jumping to a file-visiting buffer
---------------------------------------------------------------------------------
This can happen when you type ‘RET’ on a hunk to visit the respective
file at the respective position. One solution to this problem is to use
‘global-reveal-mode’. It makes sure that text around point is always
visible. If that is too drastic for your taste, then you may instead
use ‘magit-diff-visit-file-hook’ to reveal the text, possibly using
‘reveal-post-command’ or for Org buffers ‘org-reveal’.
File: magit.info, Node: Debugging Tools, Next: Keystroke Index, Prev: FAQ, Up: Top
B Debugging Tools
*****************
Magit and its dependencies provide a few debugging tools, and we
appreciate it very much if you use those tools before reporting an
issue. Please include all relevant output when reporting an issue.
‘M-x magit-version’ (‘magit-version’)
This command shows the currently used versions of Magit, Git, and
Emacs in the echo area. Non-interactively this just returns the
Magit version.
‘M-x magit-emacs-Q-command’ (‘magit-emacs-Q-command’)
This command shows a debugging shell command in the echo area and
adds it to the kill ring. Paste that command into a shell and run
it.
This shell command starts ‘emacs’ with only ‘magit’ and its
dependencies loaded. Neither your configuration nor other
installed packages are loaded. This makes it easier to determine
whether some issue lays with Magit or something else.
If you run Magit from its Git repository, then you should be able
to use ‘make emacs-Q’ instead of the output of this command.
‘M-x magit-debug-git-executable’ (‘magit-debug-git-executable’)
This command displays a buffer containing information about the
available and used ‘git’ executable(s), and can be useful when
investigating ‘exec-path’ issues.
Also see *note Git Executable::.
‘M-x with-editor-debug’ (‘with-editor-debug’)
This command displays a buffer containing information about the
available and used ‘~emacsclient’ executable(s), and can be useful
when investigating why Magit (or rather ‘with-editor’) cannot find
an appropriate ‘emacsclient’ executable.
Also see *note (with-editor)Debugging::.
Please also see the *note FAQ::.
File: magit.info, Node: Keystroke Index, Next: Command Index, Prev: Debugging Tools, Up: Top
Appendix C Keystroke Index
**************************