I have defined a .dir-locals.el file with the following content:
((python-mode . ((cr/virtualenv-name . "saas"))))
In my .emacs I have the following function to retrieve this value and provide a virtualenv path:
(defun cr/virtualenv ()
(cond (cr/virtualenv-name (format "%s/%s" virtualenv-base cr/virtualenv-name))
((getenv "EMACS_VIRTUAL_ENV") (getenv "EMACS_VIRTUAL_ENV"))
(t "~/.emacs.d/python")))
Finally, in my python-mode-hook list, I have this hook function:
(add-hook 'python-mode-hook 'cr/python-mode-shell-setup)
(defun cr/python-mode-shell-setup ()
(message "virtualenv-name is %s" cr/virtualenv-name)
(let ((python-base (cr/virtualenv)))
(cond ((and (fboundp 'ipython-shell-hook) (file-executable-p (concat python-base "/bin/ipython")))
(setq python-python-command (concat python-base "/bin/ipython"))
(setq py-python-command (concat python-base "/bin/ipython"))
(setq py-python-command-args '( "-colors" "NoColor")))
(t
(setq python-python-command (concat python-base "/bin/python"))
(setq py-python-command (concat python-base "/bin/python"))
(setq py-python-command-args nil)))))
When I open a new python file, the message logged by cr/python-mode-shell-setup
indicates that cr/virtualenv-name
is nil
. However, when I C-h v the name, I get "saas" instead.
Obviously there's a load order issue here; is there a way to have my mode hook statements respond to directory-local variables?
This happens because
normal-mode
calls(set-auto-mode)
and(hack-local-variables)
in that order.However
hack-local-variables-hook
is run after the local variables have been processed, which enables some solutions:The first is to make Emacs run a new "local variables hook" for each major mode:
(Your original function can be used unmodified, with that approach.)
A second option is to utilise the optional
LOCAL
argument toadd-hook
that makes the specified function buffer-local. With this approach you could write your hook as follows:i.e.
python-mode-hook
runs first and registers the anonymous function withhack-local-variables-hook
for the current buffer only; and that function is then called after the local variables have been processed.Lindydancer's comment prompts a third approach. It's not nearly as clean as the other two, but proved interesting regardless. I didn't like the idea of causing
(hack-local-variables)
to be called twice, but I see that if you set thelocal-enable-local-variables
buffer-locally, it prevents(hack-local-variables)
from doing anything, so you could do this:Obviously that modifies the normal sequence of execution a little, so side effects may be possible. I was worried that if the same major mode is set by a local variable comment in the file, this might cause infinite recursion, but that doesn't actually appear to be a problem.
Local variable header comments (e.g.
-*- mode: foo -*-
) are handled by(set-auto-mode)
, so those are fine; but amode: foo
Local Variables:
comment seems like it would be an issue as it is handled by(hack-local-variables)
, and so if the mode is set that way I thought it would cause recursion.In practice I was able to trigger the problem by using a simple function as a 'mode' which did nothing more than try to run its hooks; however testing with a 'proper' mode did not exhibit the problem, so it's probably safe in reality. I didn't look into this further (as the other two solutions are much cleaner than this), but I would guess the delayed mode hooks mechanism probably explains it?