I backported `find-sibling-file` from Emacs 29+ in 198fe82, but it
passes a third argument (REGEXP) to `file-expand-wildcards` which wasn't
introduced until 29, so users on 28 or earlier would see a
wrong-number-of-arguments error (see #7795).
Rather than backport the entire `file-expand-wildcards`
function (risking other edge cases), I've advised its usage in
`find-sibling-file-search` to behave as it would in 29+ with a non-nil
REGEXP argument.
Amend: 198fe82b6dFix: #7795
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.
Doom makes multiple attempts to re-clone repos if they failed the first
time, however, if a user provides a :local-repo and that location isn't
a git repo, Doom assumes this it is the result of a failed clone, and so
deletes it to "try" again (which will fail). This can result in lost
work.
This prevents this from happening to packages/repos outside of
$DOOMLOCALDIR (all packages under $DOOMLOCALDIR must be git repos).
Fix: #7785
Amend: 3643c4dadd
Amend: 1fa8d3a4b9
If the buffer is an indirect clone, it may not have a buffer-file-name.
Also changes the command to throw an error if the current file has no
file path *and* is not a dired buffer.
directory-empty-p was introduced in 28.1, but Doom still supports 27.1,
so 27.x users would see this error when running any bin/doom command
that clones packages.
Fix: #7779
Amend: 1fa8d3a4b9
It seems the temporary branch that 'doom upgrade' creates is sometimes
left over after a 'doom upgrade', preventing the next 'git fetch' from
completing.
Ref: #7771
Caused by a regression originally introduced in b6b755d, but incorrectly
fixed in 89c56a3. Don't code while sleep deprived, kids.
Amend: 89c56a3393
Amend: b6b755dea4
12a765c introduced suppression of package-rebuilding prompts on 'doom
upgrade', but inadvertently reversed the check. This fixes that (and
refactors the if/or tree into a `cond`).
Amend: 12a765c509
Prior to this, -B would suppress the prompt for package rebuilding if
your Emacs version or hostname changed. Now, it fully inhibits
rebuilding in either case.
Fix: #7760
Ref: cff091982e
'doom purge' is now deprecated.
Also changes 'doom sync's -p option to --gc. Since GCing causes the loss
of historical data, I'd rather it be a long option to make it a little
harder to do accidentally.
This changes 'doom sync' to be smarter about responding to changed
package recipes/pins, changes in Emacs version, or instances where the user
has copied a config to a new system.
In all these cases, the user would formerly have to know about a
specific combination of 'doom sync -u' and 'doom build' to ensure Doom
is in a good state. With this change, 'doom sync' handles all these
cases.
Also, 'doom build' is now deprecated (and 'doom sync' now has a
--rebuild option to mimic its old behavior).
Also also, sometimes, a package may silently fail when cloned (which
used to result in an empty repo). Now, if this is detected, cloning will
be re-attempted up to 3 times before aborting with much more visible
error.
Note: these are stopgap solutions, until v3 is finished.
Something often reported are file-missing errors when a package that
should be present isn't. This can easily happen if, say, during a 'doom
sync' or 'doom upgrade' a package fails to clone correctly and the user
misses the errors, then tries to carry on as normal. What's worse is
that Straight leaves behind an empty directory, which it treats as a
sign that the package has been cloned correctly, so it doesn't raise any
fuss over them.
With this change, 'doom sync' (and 'doom upgrade') will now try again,
if the clone process fails the first time (up to 3 times) before
aborting the whole process altogether, which should be loud enough for
users not to miss. Note that these failures at 99.99% because of
network (or upstream downtime) issues.
For now, this does leave Doom in an incomplete state (until you try
again when the connection issue is resolved), but a rollback step will
be added in v3 to prevent this, as well as better error messages (as
well as @doomelpa mirror for packages on less reliable hosts, like
codeberg, savannah, etc).
BREAKING CHANGE: This removes the 'doom compile' and 'doom clean'
commands, and offers no immediate replacement for them (and no plan to
include one). In the future, byte-compilation of Doom's internals will
be baked into 'doom sync', but until then, Doom is not optimized to take
advantage of byte-compilation, and forcing it provides no benefit.
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.
Due to a suppressed void-variable error, font initialization is
short-circuited across daemon frames when the fonts are reloaded or
changed (which is also triggered by changing themes), resulting in
malformed (often tiny) fonts in GUI frames.
Fix: #7730
In some contexts, like org-agenda-switch-to visiting a TODO in an agenda
file, the visiting command will move the cursor after opening the file.
If save-place moves the cursor yet again, the cursor will end up in an
unpredictable place.
Some site files will forcibly undo `inhibit-message` or set
`force-load-messages`. This ensures site lisp files don't make
unnecessary noise at startup.
Tools like these will be moved to a benchmark module later (or perhaps
to `:lang emacs-lisp`). For the time being, it only takes up extra space
that few users use.
I will slowly phase out projectile in favor of project.el, starting with
projectile-find-other-file, which -- as of Emacs 29 -- has a native
alternative: `find-sibling-file`.
Ref: doomemacs/community#1
