The private module won't be byte-compiled implicitly anymore, as it can
cause unexpected problems with stale code and config portability.
doom compile -> only compiles ~/.emacs.d
doom compile :core -> ~/.emacs.d/core
doom compile :plugins -> ~/.emacs.d/.local/packages/elpa
doom compile :private -> ~/.doom.d
I don't want to litter config/default/+evil-bindings.el with conditions
for every keybind whose module may or may not be enabled. It impacts its
readability and is relatively expensive (due to the internals of map!
and general).
So instead, we no-op keybinds for commands that don't exist (and aren't
autoloaded), so you don't see missing function errors when using these
keys.
Prompts for an open project and kills all its buffers. This is useful
for closing `.emacs.d` in an effort to get some *billable* work done.
Added `doom-fixup-windows` and `doom-kill-buffer(s)-fixup-windows`
helper functions for leaving affected windows on a real buffer or the
fallback buffer.
Fixed `doom/kill-this-buffer-in-all-windows` to properly "fixup"
windows.
This command would no-op if used in major modes with "fake" comments.
i.e. modes where "comments" are merely highlighted as comments, but have
no comment entry in its syntax table. e.g. text-mode + git-commit-mode.
- Limits process-environment during scraping
- Add `doom-env-executable` and `doom-env-switches` variables
- Announce what commands were run to produce your env var within env var
file header
Bloated projectile caches can cause freezing and delays. This is
especially so if you accidentally index $HOME or /, which will bloat
them into the hundreds of MBs.
+ Adds purging of blacklisted, over-sized and non-projects from
projectile's cache when Emacs is killed.
+ Projectile's project file cache will now expire after a week.
+ Corrects file paths in core/doctor's filesize checks.
IMPORTANT: This is a breaking update for Mac users, as your shell
environment will no longer be inherited correctly (with the removal of
exec-path-from-shell). The quick fix is: 'bin/doom env refresh'. Also,
the set-env! autodef now does nothing (and is deprecated), be sure to
remove calls to it in your config.
Smaller changes:
+ This update also adds --no-* switches to doom quickstart
+ Includes general improvements to the documentation of several bin/doom
commands.
+ Moves doom/reload* commands to core/autoload/config.el
+ doom/reload-project has been removed (it didn't actually do anything)
The breaking change:
This update adds an "envvar file" to Doom Emacs. This file is generated
by `doom env refresh`, populated with variables scraped from your shell
environment (from both non-interactive and interactive sessions). This
file is then (inexpensively) loaded at startup, if it exists.
+ The file is manually generated with `doom env refresh`.
+ It can be regenerated automatically whenever `doom refresh` is run by
running `doom env enable` (`doom env clear` will reverse this and
delete the env file).
+ `doom quickstart` will ask if you want to auto-generate this envvar
file. You won't need it if you're confident Emacs will always be
started from the correct environment, however.
+ Your env file can be reloaded from a running Emacs session with `M-x
doom/reload-env`. Note: this won't work if the Emacs session you're
running it in doesn't have a correct SHELL set. i.e. don't use this to
create your first env file!
The idea isn't mine -- it's borrowed from Spacemacs -- and was
introduced to me in #1053 by @yurimx. I was impressed with it. Prior to
this, I was unhappy with exec-path-from-shell (no hate to the dev, I
understand its necessity), and 'doom patch-macos' wasn't ideal for mac
users (needed to be reapplied every time you update Emacs). What's more,
many users (even Linux users) had to install exec-path-from-shell
anyway.
This solution suffers from none of their shortcomings. More reliable
than patch-macos, more performant and complete than
exec-path-from-shell, and easily handled by bin/doom.
helm-projectile-find-file misbehaves as a workspace project-switch
handler (likely because it runs asynchronously and misses the lexical
value of `default-directory`), so we avoid it and use
projectile-find-file directly (which still uses helm, just not the
helm-projectile package).
Refactors :after-call to not fmakunbound the transient hook function, as
it may get executed multiple times before getting a chance to fully
remove itself, causing void-function errors.
+ Bump Emacs version check to 25.3
+ Fix doctor reporting missing packages that are user-disabled
+ Add Doom core checks for over-sized cache files (a possible cause of
freezes/hangs)
+ Emit a backtrace from module doctor script errors
+ Fix doom doctor not respecting DEBUG envvar
This was erroring when run inside a popup (eg. `M-x ielm` `C-x k`) due
to a missing check for `window-live-p`. We don't need to do anything if
the window has already gone away.
The real vs unreal check was backwards; we should only try
`previous-buffer` if the current buffer is *not* real.
This will later be used for doom/describe-packages to list all locations
where a package is being configured (along with def-package! and after!
blocks).
Patch the apropos button types so they call helpful instead of the
built-in describe functions. Also add some bindings to apropos-mode-map
so it behaves like other help modes.
Add `doom/describe-symbol` function, which shows documentation for
callable and variable symbols. If a symbol is both a variable and a
callable, it dispatches to apropos. This gives a better workflow than
`helpful-symbol`, which annoyingly prompts the user.
Remap `describe-symbol` to `doom/describe-symbol`, and update
`+emacs-lisp-lookup-documentation` to call it also.
Previously this would error due to undefined `doom-debug-on-error`
variable. Now behaves like a minor-mode function to toggle both
`doom-debug-mode` and `debug-on-error`.
Can now launch four different sessions:
1. Vanilla Emacs
2. Vanilla Doom (only Doom core)
3. Doom core + modules - private config
4. A full Doom session (load everything)
- Packages are initialized once, when package.el is first loaded, and
must be updated manually via doom/reload-packages.
- Package->module association is now stored in the package's PLIST under
:modules. This is an internal property and cannot be explicitly set
through `package!'
- Add doom-package-list function
- Rename doom-get-packages to doom-find-packages
- Updated doom-find-packages' docstring
- Added the :core filter to doom-find-packages
- Simplified doom-initialize-packages
- doom/reload calls doom/reload-packages if necessary.
- Fix redundant properties in doom-packages
- Remove tracking of after!, def-package! and def-package-hook! blocks.
Replaced with doom-package-list being able to see all packages, even
in disabled modules.
- Add :built-in property to package! for dummy packages. This is
important so that doom/describe-package can see built-in packages.
+ Don't use set-frame-font. Causes issues for daemon users and is
slower.
+ Revice init function docstrings
+ Load fonts a little earlier than themes
May resolve an issue where fonts in daemon Emacs are too small on
startup.
Addresses #1223
Also adds doom-after-init-modules-hook for consistency, which is an
alias to after-init-hook until I decide if it's necessary for it to be
independent.
Using general to bind leader keys was responsible for 40-50% of Doom's
startup time. This change reduces that significantly, but not entirely.
It may be better that the config/default module not use map!. It is a
convenient macro, but general is a huge bottleneck.
SPC h now maps to help-map to reduce redundancy. Some adjustments and
additions were made to this map so all users (evil and vanilla) can
benefit from Doom's extra help commands.
+ Add doom-switch-frame-hook
+ Replace doom-{enter,exit}-{buffer,window}-hook with
doom-switch-{buffer,window}-hook
+ New switch-buffer hooks run on buffer-list-update-hook rather than
in select-window advice.
+ Blank our buffer-list-update-hook in some places to reduce how many
times it gets triggered.
By removing the cl-flet call, we reduce the size of backtraces produced
during bin/doom commands by a whopping 80%. Noice.
Also renames doom-ansi-apply -> doom-color-apply
- doom-post-init-hook was renamed doom-init-modules-hook
- doom-init-hook was renamed doom-before-init-modules-hook
- doom-init-modules-hook now runs before the user's config.el is run
- Moved doom-init-ui-hook to run later (on window-setup-hook rather than
emacs-startup-hook).
Yield a modest improvement in startup times.
- Code reduction and refactor across the board (cull unneeded minor
advise, hooks and hacks or update them)
- Revise outdated comments and docstrings
- Reorganize core autoload libraries
- Remove large file check (Emacs already has a built-in one, which we
augment to be even more performant when it does kick in)
- helpful.el can now be disabled completely through package!
They've been removed from feature/workspaces and moved into
core/autoload/sessions, which falls back to desktop.el if persp-mode
isn't present. This also offers a substantial speed up to
restart+restoring and restoring sessions in general.
Also fixes#1210, where the newly spawned frame after doom/restart
wasn't focused.
Introduces the following commands:
- doom/restart
- doom/restart-and-restore
- doom/quickload-session
- doom/quicksave-session
- doom/load-session
- doom/save-session
- +workspace/restore-last-session (alias for doom/quickload-session)
And removes
- +workspace/load-session
- +workspace/save-session
- +workspace/load-last-session (renamed to +workspace/restore-last-session)
- +workspace/restart-emacs-then-restore (replaced by doom/restart-and-restore)
- :ss (ex command)
- :sl (ex command)
The prompt in the minibuffer is read-only. You are able to move the
cursor into it before this fix.
This also more effectively silences echo-area output when deleting text
in the minibuffer. No more "Text is read-only" blocking what you're
typing.
;;;###autodef FORM
FORM was used as a predicate for inclusion as an autodef. Now it is used
as the replacement sexp in case the module is disabled.
Oh, you don't know what autdefs are? Well let me explain (thanks for
asking, by the way). An autdef'ed function, macro, or function alias is
always available to be called, anywhere in Doom, even if its containing
module is disabled. For instance:
;;;###autodef
(defun say-hello! (name) ; the trailing ! denotes an autodef
(message "Hello %s" name))
This makes it safe to call `do-something` without a check whether it
exists (or if its module is enabled). When the module is enabled, an
autoload entry is added to the Doom autoloads file:
(autoload 'do-something "path/to/some/modules/autoloads")
And it is autoloaded as normal when it is first used. However, if the
module is disabled, then this is inserted instead:
(defmacro do-something (&rest _))
This no-ops; it does nothing and doesn't evaluate its arguments. If FORM
above was provided, that is used instead of a noop macro.
It's a little smarter than simple substitution, but that's the gist of
it.
These weren't reliable, often times buggy or overzealous about killing
buffers and processes. Best to do it manually or come up with a better
solution.
- SPC f . -> counsel-file-jump or find-file
- SPC f > -> doom/browse-in-other-projects
- SPC f / -> projectile-find-file
- SPC f ? -> doom/find-file-in-other-project
- Moved doom/sudo-find-file to SPC f S
This change was made to accommodate the new
doom/browse-in-other-projects and doom/find-file-in-other-project
commands, which make it easy to jump to files in other known projects.
The purpose of this is to highlight indentation characters that betray
your indent-tabs-mode setting. i.e. If you're using tab indentation,
highlight space indentation. If you're using spaces, highlight tab
characters.
On MacOS, command used to be 'meta, which Emacs (and many packages) use
for many keybinds. I don't want to pollute the command key, so it is now
meta (as is the Emacs default).
The MacOS keybind fixes have been moved back to super.
When indexing it ignores hidden files, which is especially annoying for
dotfiles where everything starts with a dot. If you just have -H it'll
index .git too though, so we exclude that
Using meta is a little more flexible. Since there is no dedicated meta
key on our keyboards (anymore), it can be remapped to another modifier
by changing mac-command-modifier, x-meta-keysym, etc. without
sacrificing super or alt keybinds.
This changes how leader keys are bound, to fix an issue where the wrong
which-key label was assigned to the wrong keys, and cases where the
leader key was being shadowed by other minor mode mappings.
Unfortunately, this new method adds 10-20% to startup times. I'll
address this in a future patch. For now, correctness is more important.
Also fixes dashboard keybind detection.
The global leader keybind was conflicting with a global M-SPC keybind in
helm-map. This keybind should only be set in non-evil sessions, so we
unset it if evil is found.
Due to issues with preset prefixes in general definers and nested
:prefix's supplied from a map! call not cooperating, many localleader
keybinds were broken and causing errors. For :leader/:localleader keys,
we now use :infix for sub-prefixes.
However, with this change, the :alt-prefix property has been removed, as
there is no simple way to support this without some major state
gymnastics in map!.
Fixes#1059
This resolves issues with :leader/:localleader keys not working when
evil states are specified. Evil states are now ignored. Also, some of
map!'s internals have been optimized to yield a ~10% improvement in
macro expansion time.
The former approach was the cause for a huge increase in startup
time (adding ~0.4s) when :leader and :localleader were used. This is
because general-define-key was called for every key-def pair.
This new approach batches these calls, which has decreased the
performance impact by at least 80%.
Before general.el was introduced to Doom, commands were bound to keys
whether or not their containing modules were enabled. This was out of
laziness and readability. I intend to change this, as such it is no
longer necessary to hide unavailable keybinds from which-key.
+ Now uses an overriding keymap for leader keys, so that it is always
available, even outside of normal/visual states. In insert/emacs
states, or in sessions where evil is absent, an alternative prefix is
used for leader/localleader keys. See these variables:
+ doom-leader-prefix
+ doom-leader-alt-prefix
+ doom-localleader-prefix
+ doom-localleader-alt-prefix
+ Keybinds now support alternative prefixes through the new :alt-prefix
property. This is useful for non-evil users and non-normal evil
states. By default, this is M-SPC (leader) and M-SPC m (localleader).
+ Removed +evil-commands flag from config/default (moved to
feature/evil/+commands.el).
+ config/default/+bindings.el has been split into
config/default/+{evil,emacs}-bindings.el, which one is loaded depends
on whether evil is present or not. The latter is blank, but will soon
be populated with a keybinding scheme for non-evil users (perhaps
inspired by #641).
+ The define-key! macro has been replaced; it is now an alias for
general-def.
+ Added unmap! as an alias for general-unbind.
+ The following modifier key conventions are now enforced for
consistency, across all OSes:
alt/option = meta
windows/command = super
It used to be
alt/option = alt
windows/command = meta
Many of the default keybinds have been updated to reflect this switch,
but it is likely to affect personal meta/super keybinds!
The map! macro has also been rewritten to use general-define-key. Here
is what has been changed:
+ map! no longer works with characters, e.g. (map! ?x #'do-something) is
no longer supported. Keys must be kbd-able strings like "C-c x" or
vectors like [?C-c ?x].
+ The :map and :map* properties are now the same thing. If specified
keymaps aren't defined when binding keys, it is automatically
deferred.
+ The way you bind local keybinds has changed:
;; Don't do this
(map! :l "a" #'func-a
:l "b" #'func-b)
;; Do this
(map! :map 'local "a" #'func-a
"b" #'func-b)
+ map! now supports the following new blocks:
+ (:if COND THEN-FORM ELSE-FORM...)
+ (:alt-prefix PREFIX KEYS...) -- this prefix will be used for
non-normal evil states. Equivalent to :non-normal-prefix in general.
+ The way you declare a which-key label for a prefix key has changed:
;; before
(map! :desc "label" :prefix "a" ...)
;; now
(map! :prefix ("a" . "label") ...)
+ It used to be that map! supported binding a key to a key sequence,
like so:
(map! "a" [?x]) ; pressing a is like pressing x
This functionality was removed *temporarily* while I figure out the
implementation.
Addresses: #448, #814, #860
Mentioned in: #940
Emacs occasionally hangs when polling for the emacs server (with
server-running-p). Since this is used for such a trivial feature (to
decide whether or not to display the "you need to restart" message), I
removed it. Now it always shows that message (if the autoload files have
changed).
undo-tree-load-history was formerly advised with doom*shut-up, which
uses the quiet! macro to suppress output. quiet! accomplishes this by
temporarily redefining message to a no-op function. However, if a fatal
error occurs while this binding is active, in some cases, message will
remain redefined, perpetually silencing all output to the minibuffer.
This tries to mitigate that, at least where undo-tree is concerned.
Also sharp-quotes an unquoted function.
Fixes an issue where reading TAGS files could cause "%s is a large file,
open literally to avoid performance issues?" prompts every time you open
a project file, if the tags file was larger than `doom-large-file-size'
Uses a less destructive method (the same that Spacemacs uses) than the
one introduced in 13cee68, by introducing MODE-local-vars-hook hooks,
which run after local vars have been initialized.
The old method was to call `hack-local-variables` *before* mode hooks
run, however, this causes variables set by modes to have higher
precedence than local vars, which is unacceptable.
Also moved intero-mode & dante-mode to haskell-mode-local-vars-hook
The default behavior is to read file+directory-local variables after the
major mode and its hooks have run. I think this is backwards. What if we
want to use these local variables to customize the things running in
hooks?
This does mean hack-local-variables will run at least twice when the
mode changes, but this is an acceptable compromise.
A minor optimization. assq is significantly faster than assoc (not that
it matters for this incredibly insignificant use-case, but yay for
premature optimization!)
+ Add doom-serif-font variable
+ Update docstrings of all doom-*font variables to mention that they all
support font-specs, font objects, XFT strings and XLFD strings.
+ Set doom-font if the user hasn't, ensuring that other functions know
what the current, default font is set to (fixes doom-big-font-mode not
switching back to normal when disabled).
Now that projectile-project-root has changed not throw errors (depending
on projectile-require-project-root), it is safe to change this, in case
you use other projectile commands that do require a project.
projectile-project-root no longer returns `default-directory` if not in
a project (it returns nil). As such, doom-project-* functions (and their
uses) have been refactored.
+ doom-project-p & doom-project-root are aliases for
projectile-project-p & projectile-project-root.
+ doom-project-{p,root,name,expand} now has a DIR argument (for
consistency, since projectile-project-name and
projectile-project-expand do not).
+ The nocache parameter is no longer necessary, as projectile's caching
behavior is now more sane.
+ Removed some projectile advice/hacks that are no longer necessary.
+ Updated unit tests
It's possible for the debugger to be invoked from inside code wrapped in
a (quiet! ...) call. The debugger pauses Emacs in a broken state where
the functions locally rebound by quiet! (e.g. message, load-file,
write-region, etc) are never returned to their original definitions.
This attempts to reduce that probabilityby changing how quiet! silences
code. Rather than silencing them completely, they will be logged
to *Messages* but not displayed in the echo area.
Also, quiet! is now used less, where it isn't strictly needed (or where
inhibit-message is sufficient).
Done to make bin/doom produce better debugger output (and more readily).
A lot of bin/doom errors aren't recurring, so it's better to produce the
full error report ASAP.
This is experimental. I'd eventually like users to be able to management
Doom from inside Emacs, if they desire. This may be the only way to have
a decent user experience on Windows, for that matter.
Also adds a popup rule for it.
This should fix an issue where a package A, which uses macros from a
package B, is installed before package B, causing void-function errors.
The currently known and affected packages are neotree, parinfer, and
evil-collection.
Even when an `.elc` is preset `load-theme` would attempt to load the `.el` first,
to give the opportunity to inspect the theme.
However we are loading themes automatically from (M)ELPA, and loading a
theme package is no more dangerous than loading a regular package.
So override the search order for themes and atttempt to load the `.elc`
first.
This improves startup time by ~25ms with the default theme.
Improved implementation by @hlissner.
Signed-off-by: Edwin Török <edwin@etorok.net>
custom-set-faces is somewhat overkill for this. It overwrites any user
customizations for the default & variable-pitch faces, and saves them to
custom-file, which may cause those settings to persist even when we
don't want them to.
This new solution is a little faster, is more respective of
customizations, and adds XFT font string support.
XFT font strings are a little more powerful. For example, the :antialias
font-spec property was unreliable, but now you can use:
(setq doom-font "Fira Mono:pixelsize=12:antialias=off")
And replace it with buffer-local mode-name setters. This is more
explicit and less magical, which is easier for users to discover and
change, if they'd like.
Hooks run with this should already be run in a gc-deferred state (at
startup), so deferring garbage collection _again_ may just cause
unnecessary consing.
This is for loading large packages (like org and magit) quietly in the
background during idle time. It is already set up to incrementally load
org and magit.
This is still experimental, however. the idle timers may need to be
tweaked.
Only add doom-private-dir to load-path during autoload generation,
otherwise a $DOOMDIR/autoload.el will shadow the built-in autoload.el
Emacs package.
Note: with `doom-private-dir' in `load-path', Doom autoloads files will
be unable to declare autoloads for the built-in autoload.el Emacs
package, should $DOOMDIR/autoload.el exist. Not sure why they'd want to
though, so it's an acceptable compromise for simpler autoload
declarations.
The function isn't really expecting a function, but the 'eq, 'eql or
'equal symbols. User supplied test/hash functions must be defined with
define-hash-table-test.
It is easier to spot real problems if the code is warning-free.
Replacing `gensym` with `make-symbol` is an idea taken from here:
b44c08dd45
In defer-until!:
core-lib.el:150:19:Warning: function ‘gensym’ from cl package called at
runtime
In add-transient-hook!:
core-lib.el:216:16:Warning: function ‘gensym’ from cl package called at
runtime
In toplevel form:
autoload/message.el:35:1:Warning: Unused lexical variable ‘spec’
In toplevel form:
autoload/line-numbers.el:31:1:Warning: defcustom for
‘display-line-numbers-type’ fails to specify containing group
autoload/line-numbers.el:31:1:Warning: defcustom for
‘display-line-numbers-type’ fails to specify containing group
autoload/line-numbers.el:39:1:Warning: defcustom for
‘display-line-numbers-grow-only’ fails to specify containing group
autoload/line-numbers.el:39:1:Warning: defcustom for
‘display-line-numbers-grow-only’ fails to specify containing group
autoload/line-numbers.el:44:1:Warning: defcustom for
‘display-line-numbers-width-start’ fails to specify containing group
autoload/line-numbers.el:44:1:Warning: defcustom for
‘display-line-numbers-width-start’ fails to specify containing group
In toplevel form:
cli/autoloads.el:137:1:Warning: Unused lexical variable ‘type’
Preserve name of unused lexical var _type
Makes it obvious what is stored there.
We would need to use `'equal` for comparison, but Emacs 25 only allows `'eq`.
Using `advice-add` to override `alist-get` does not work, because `setf`
has special handling for `alist-get`.
`repl.el`: Switch to a hash table which already supports multiple comparison
functions, and changing of elements even in Emacs 25.
`eshell/autoload/settings.el`: use conditional set-or-push.
Drop `doom*alist-get`, it is unused now.
Thanks to @hlissner for the reimplementation.
+ Associates enh-ruby-mode with :lang ruby for doom/describe-module
+ Enables editorconfig to detect extension-less or file-less ruby
files (from shebang lines or file local variables).
The defaults were designed for much, much older systems. We can safely
increase this to stave off premature abortion of intentional deeply
nested loops or excessive allocations.
Color let-functions no longer take format string arguments. e.g.
(format! (red "Hello %s" "world"))
Becomes
(format! (red "Hello %s") "world")
The same goes for print!. Also, doom-ansi-apply now takes two arguments
instead of three.
Also merges doom-message-{fg,bg,fx} into doom-ansi-alist, and reduces
backtrace noise when errors originate from inside these macros.
Also fixes an issue where doom install wouldn't catch missing
dependnecies of desired packages.
If :dep is non-nil, doom-get-packages will include package dependencies
in the resulting package list.
Checking for the 'display-line-numbers symbol property is clumsy and
checking for boundp is pointless now that we have a Emacs 25 polyfill
for display-line-numbers-mode.
This prevents absolute paths appearing in many find-in/browse commands,
like +default/find-in-emacsd, +default/find-in-config, or
+default/browse-project.
+ Add doom-gc-cons-upper-limit variable
+ Don't use most-positive-fixnum, in case Emacs somehow wants to
allocate that much!
+ Don't use startup optimizations in noninteractive sessions
+ Restore gc-cons-threshold on idle timer, instead of hook (to defer the
possible GC pause and so deferred packages can take advantage of these
optimizations).
Only jit-lock-stealth-nice is different from the default, and slows down
some scrolling operations without helping much. It may be better to
conditionally enable this in larger buffer.
This was done to help the modeline build its file path in indirect
buffers, but this has been fixed by using buffer-base-buffer.
Both ui/modeline and ui/doom-modeline have been refactored to
accommodate this.
+ Remove doom/switch-theme (replaced with an advice for load-theme)
+ Reorganize core-ui
+ Fix Emacs 26+ hl-line hack being applied for Emacs 25 users
+ Rename doom|show-whitespace-maybe to doom|highlight-non-default-indentation
+ Disable tool-bar, menu-bar and vertical-scroll-bars via
default-frame-alist; this is a little faster than using the minor
modes.
ws-butler only strips trailing spaces on lines that have been modified.
+ ws-butler is disabled if editorconfig enables
trim_trailing_whitespace, which resorts to delete-trailing-whitespace
instead.
+ Updates doom|(enable|disable)-delete-trailing-whitespace hooks to use
ws-butler-mode.
Also adds doom|disable-delete-trailing-whitespace hook, which can be
used to disable this behavior on a per-buffer basis. If you want to turn
off this global behavior:
(remove-hook 'write-file-functions #'delete-trailing-whitespace)
If HOME is a repo, projectile resolves all project roots to HOME. This
fixes any commands that rely on this project resolution by explicitly
telling them what project they're in instead.
This change facilitates the users that prefer their custom-file
somewhere other than doom-local-dir (e.g. in ~/.doom.d/custom.el).
This change prevents loading both files during startup. Just change
`custom-file` and Doom will load it after your config.el is loaded (but
before doom-post-init-hook runs).
