After some profiling, it turns out map-put and map-delete are 5-7x
slower (more on Emacs 25) than delq, setf/alist-get and add-to-list for
small lists (under 250 items), which is exactly how I've been using
them.
The only caveat is alist-get's signature is different on Emacs 25, thus
a polyfill is necessary in core-lib.
The way Doom was using eval-after-load ensured its form were never
byte-compiled or even checked by the byte-compiler, because they were
treated as quoted forms (data), and thus eval'ed.
Friends don't let friends use eval.
Another refactor, again to improve the locality of doom errors and make
the data that accompanies them more useful in determining the origin and
source of issues. Also, bin/doom is now a little more informative about
how to debug errors.
Caused because of symbol-file advice assuming SYMBOL would always be a
symbol.
This would prevent backtraces from happening, making debugging
impossibly obtuse.
This may indirectly fix#701
Stubs are defined as ordinary macros (with defmacro), but if an autodef
is defined with cl-defun, defmacro can't handle the clisp-style argument
lists, causing wrong-number-of-arguments errors.
The error handlers were a little too effective. They obscured a large
chunk of the stacktrace after errors, even in debug mode. This fixes
that and ensures backtraces in debug mode are more helpful.
This fix prevented the byte-compiler from trying to load packages that
were disabled or failed their :when/:unless/:if predicates. This commit
moves it into doom-byte-compile, so the :no-require predicate doesn't
have to run on every package in an interactive session, eating MY
PRECIOUS cpu cycles.
I do love my cpu cycles, yessiree.
This will hopefully reveal more information as to the cause and origin
of errors at startup. It should also make doom-debug-mode more likely to
produce a backtrace in non-interactive sessions.
This naming convention was meant to be for batch commands, but it grew
to include "commands that were helpful with managing Doom", but many of
these commands shouldn't be interactive in the first place!
A regression caused the former to error out (if the buffer name was less
than 2 characters long) and the latter to return t too eagerly (because
of an incorrect condition chain).
I can feel it; the power coursing through my veins. Every millisecond
saved at startup makes me more powerful. This world will be mine. Mine I
tell you!
