This reverts commit 29c78ce841 and all other fixes
that have been made afterwards.
The motivation is that use-package is seen by many as a replacement for
`require`. Is use-package always defer the loading of packages then is breaks
this use case, this does not respect POLA so even if it was making Spacemacs
loading faster (up to 3s faster on some startup on my machine) we just cannot
use it, it would be irresponsible. Spacemacs should be easy to use, loading
performance will come with time but it is not a priority.
Enabling a company backend for a specific mode was a tedious tasks with code
scattered at different locations, one for local variable definitions, one for
company hook function definitions and another where the backends were pushed to
the local variables (which was problematic, since we ended up pushing the same
backends over and over again with `SPC f e R`, pushes have been replaced by
add-to-list calls in the new macro).
All these steps are now put together at one place with the new macro
spacemacs|add-company-backends, check its docstring for more info on its
arguments.
This macro also allows to define arbitrary buffer local variables to tune
company for specific modes (similar to layer variables via a keyword :variables)
The code related to company backends management has been moved to the
auto-completion layer in the funcs.el file. A nice side effect of this move is
that it enforces correct encapsulation of company backends related code. We can
now easily detect if there is some configuration leakage when the
auto-completion layer is not used. But we loose macro expansion at file loading
time (not sue it is a big concern though).
The function spacemacs|enable-auto-complete was never used so it has been
deleted which led to the deletion of the now empty file core-auto-completion.el.
The example in LAYERS.org regarding auto-completion is now out of date and has
been deleted. An example to setup auto-completion is provided in the README.org
file of the auto-completion layer.
Squashing the 3 commits of #5688:
Commit 1
--------
Add prefix doc in idris layer
Also remove `idris-ensure-process-and-repl-buffer` as this is not an
interactive command
Add `SPC m l` keybinding for extracting lemma
Bind `SPC m c` to `idris-case-dwim`
Commit 2
--------
Fix registering idris repl
Commit 3
--------
Add basic auto-completion for idris mode
Motivation
While disabling Evil in holy-mode makes its implementation shorter and
sounds elegant on the paper, in practice it puts a big burden on the
configuration parts which need to know if Evil is enable or not. This is
a bad separation of concerns and the bunch of fixes that we were forced
to do in the past weeks shows this issue. Those fixes were about
removing the knowledge of the activation of Evil by implementing new
dispatching functions to be used by layers, this is cumbersome and makes
Spacemacs layer configuration more subtle which is not good. There was
additional bad consequences of the removal of Evil state like the
impossibility to use Evil lisp state or iedit states, or we would have
been forced to implement a temporary activation of Evil which is
awkward.
Instead I reintroduce Evil as the central piece of Spacemacs design thus
Evil is now re-enabled in holy-mode. It provides the abstraction we need
to isolate editing styles and be able to grow the Spacemacs
configuration coverage sanely. Layers don't need to check whether the
holy mode is active or not and they don't need to know if Evil is
available (it is always available). We also don't need to write
additional dispatching functions, this is the job of Evil, and I think
it provides everything for this. Ideally configuration layer should be
implemented with only Evil in mind and the holy-mode (and hybrid-mode)
should magically make it work for Emacs style users, for instance we can
freely use `evil-insert-state` anywhere in the code without any guard.
Evil is now even more part of Spacemacs, we can really say that
Spacemacs is Emacs+Evil which is now an indivisible pair. Spacemacs
needed this stable API to continue on the right track.
While these changes should be rather transparent to the user, I'm sorry
for this experimental period, I failed to see all the implications of
such a change, I was just excited about the possibility to make Evil
optional. The reality is that Spacemacs has to embrace it and keep its
strong position on being Emacs+Evil at the core.
Implementation
- insert, motion and normal states are forced to emacs state using an
advice on `evil-insert-state`, `evil-motion-state` and
`evil-normal-state` respectively. These functions can be used freely in
the layer configuration.
- A new general hook `spacemacs-editing-style-hook` allow to hook any
code that need to be configured based on the editing style. Functions
hooked to this hook takes the current style as parameter, this
basically generalize the hook used to setup hjkl navigation bindings.
- ESC has been removed from the emacs state map.
- Revert unneeded changes
- Revert "evil: enter insert-state only from normal-state"
commit bdd702dfbe.
- Revert "avoid being evil in deft with emacs editing style"
commit f3a16f49ed.
Additional changes
All editing style packages have been moved to a layer called
`spacemacs-editing-styles`
Notes
I did not have time to attack hybrid mode, I should be able to do it
later.
Idris development results in a lot of transient popup buffers, which
the layer currently doesn't handle particularly well. This commit uses
popwin so that these special buffers appear in a predictable location
and opens them in motion state so that they can be closed immediately by
pressing ~q~.
These changes are largely based on the Clojure layer, which has worked
very well in my experience.
The idris-mode package has been updated to no longer use the
idris-packages variable, which allows the Idris layer to safely be
renamed to the standard for other languages.
