Error occurs on Emacs 30.0.92 or newer, because this variable (what it's
aliased to, at least) was moved to comp-run.el, upstream.
Ref: emacsmirror/emacs@e6a955d242
The lexical bindings for the load suffix variables in early-init.el does
this work already, so affecting the global state of these variables is
redundant and overkill.
It seems either Emacs' warning library and/or debugger relies on
custom.el functionality at load time to properly function. If
`custom-dont-initialize` is non-nil when a warning or error occurs,
Emacs fails to fully load the `warning` or `backtrace` libraries,
causing this error to obscure the true warning/error with:
Error in delayed-warnings-hook (display-delayed-warnings): (void-variable
warning-minimum-log-level)
`doom-module-load-path` will be removed in the next big refactor commit,
so as a stop gap, I simply move it to doom.el to resolve the reference
error.
This also removes the obsolete alias `doom-modules-dirs`, since it's
been deprecated long enough (and `doom-module-load-path` itself won't be
around much longer anyway).
Ref: #8147
Amend: 8cafbe4408
Seems these two macros were marked obsolete very recently on Emacs 31.1
in favor of (if|when)-let*. Since the change seems premature (judging
from the mailing list discussion), and because I disagree with the
change (and will redefine (if|when)-let if they actually go through with
removing them), I simply silence the warnings altogether. They're not
helpful for the end user and end up only spamming them with unactionable
warnings.
I'll wait until upstream figures it out before I make any decisions.
Ref: https://lists.gnu.org/archive/html/emacs-devel/2024-10/msg00637.html
BREAKING CHANGE: This restructures Doom's core in an effort to slim it
down and partially mirror architectural changes coming in v3. This is
part 2 of 3 commits (part 1 being 1590434), done to facilitate a change
in part 3 that will introduce a new `doom!` syntax for pulling
third-party module libraries from remote sources (similar to `package!`
statements). I am backporting this from V3 so I can move our modules out
into separate repos sooner than later, so development on modules can
continue separately without interfering with v3's roll out.
Though this is labeled a breaking change, it shouldn't affect most users
except those few tinkering directly with Doom's internals.
Ref: 15904349cf
BREAKING CHANGE: This backports some architectural choices from v3.0.
This changes Doom's module API, renaming some functions and removing
others, in order to facilitate some new features, prepare to move Doom's
modules into separate repos, and make way for two, much larger breaking
commits coming in the next few days.
This commit won't break anything for users unless they're tinkering with
Doom's internals/using its `doom-module-*` API directly. I am avoiding
broader backwards incompatibilities until the 3.0 release.
What's new:
- Negated flags. (modulep! :editor evil -everywhere) will return non-nil
if :editor evil is active without its +everywhere flag.
- `modulep!` now takes multiple flags to simplify AND checks. E.g.
(and (modulep! +foo)
(modulep! +bar)
(not (modulep! +baz)))
Can now be expressed with:
(modulep! +foo +bar -baz)
- Adds pcase matchers for `doom-module-context` and `doom-module`
structs, making the following destructuring binds possible:
(pcase-dolist ((doom-module group name flags features)
(hash-table-values doom-modules))
...)
This will be used more in v3.0.
- Adds file cookie support to module init.el and config.el files.
Here's a summary of breaking changes made in this commit:
- `doom-module-context` was changed from a vector to a struct (record).
- `doom-modules` is now a table of `doom-module` structs, rather than
free-form plists.
- The following macros have been renamed:
- `doom-context-with` -> `with-doom-context`
- `doom-module-context-with` -> `with-doom-module`
- The followings functions have been replaced/removed:
- `doom-module-context`+`doom-module-context-get` -> `doom-module`
- `doom-module-set` -> `doom-module--put`
- `doom-module-p` -> `doom-module-active-p`
- `doom-module-context-key` (is now a getter with the same name)
- `doom-module-put` (removed)
- `doom-module--context-field` (removed)
- The signatures for these functions have changed:
- `doom-module-get CATEGORY &optional MODULE PROP` ->
`doom-module-get (GROUP . MODULE) &optional PROP`
- `doom-module-locate-path CATEGORY &optional MODULE FILE` ->
`doom-module-locate-path (GROUP . MODULE) &optional FILE`
- `doom-module-expand-path CATEGORY MODULE &optional FILE` ->
`doom-module-expand-path (GROUP . MODULE) &optional FILE`
- Adds the following functions
- `doom-module-exists-p`
- `doom-module-key`
- `doom-module->context`
- `doom-module<-context`
- Removes the following variables
- `doom-module--empty-context`
This commit results in a little redundancy, which I will address in
parts 2/3 and/or v3.0.
Eventually, Emacs will process any files passed to it via the command
line, and will do so *really* early in the startup process. These might
contain special file paths like TRAMP paths (e.g.
/sudo://etc/ssh/ssh_config), so restore `file-name-handler-alist' just
for this small portion of startup.
There are still a few edge cases where inhibit-redisplay is never
reset (usually involving errors at startup), leaving Emacs as a blank
screen until the user performs an action that forces it to redraw (e.g.
pressing M-x). This tries to address more of those.
The motivations for delaying the site-lisp files are different between
interactive and non-interactive sessions. I've revised these comments to
reflect that.
These checks and warnings were added on some recent commit of Emacs.
They're annoying to deal with and the end-user typically can't do much
about them (e.g. old packages), so I suppress them across the board.
That said, there are a few we cannot catch in non-interactive
sessions (where they aren't delayed), and those early warnings don't
respect delayed-warnings-list. This ought to be considered a bug
upstream.
I revisit all our startup optimizations to see how they fair in Emacs
29.x and 30.x. Most of them still hold up. I've revised and updated most
of the accompanying comments to better explain them, given what I know
now compared to when I first wrote them.
Before this, startup optimizations were disabled in debug mode, but more
often than not, this just made it difficult to reproduce some errors at
startup.
One various OSes, Emacs ships with site-lisp files that load
OS/architecture-specific config (like native-comp config), or load-lines
for Emacs packages installed via your OS package manager (like mu4e).
Output from these are rarely suppressed, for some reason, which causes
noise in *Messages* at startup, which triggers a redraw, which can be
very expensive during startup, depending on your window system.
BREAKING CHANGE: This deprecates the IS-(MAC|WINDOWS|LINUX|BSD) family
of global constants in favor of a native `featurep` check:
IS-MAC -> (featurep :system 'macos)
IS-WINDOWS -> (featurep :system 'windows)
IS-LINUX -> (featurep :system 'linux)
IS-BSD -> (featurep :system 'bsd)
The constants will stick around until the v3 release so folks can still
use it -- and there are still some modules that use it, but I'll phase
those uses out gradually.
Fix: #7479
Ensures that lexical contexts are never taken into account, in the case
where Doom's core is loaded in an isolated environment (e.g. the
sandbox). Also improves my startup time by 10%? I'll take it.
Silences some byte-compiler warnings about:
- 'nreverse on constant list' on add-hook! calls.
- inhibit-changing-match-data deprecation warning.
- unescaped quotes in docstring in some doom-*-dir variables.
- Variable non-essential should be quoted (though it isn't referring to
a variable).
- CONTEXT -> CONTEXTS to match argument name.
