Package: emacs;
Reported by: Julian Gilbey <julian-gnu <at> d-and-j.net>
Date: Wed, 25 Nov 2020 13:10:01 UTC
Severity: normal
Tags: moreinfo, wontfix
Found in version 27.1
Done: Lars Ingebrigtsen <larsi <at> gnus.org>
Bug is archived. No further changes may be made.
View this message in rfc822 format
From: Stefan Monnier <monnier <at> iro.umontreal.ca> To: Julian Gilbey <julian-gnu <at> d-and-j.net> Cc: 44864 <at> debbugs.gnu.org Subject: bug#44864: 27.1; advice.el: ad-with-originals deprecated, but no advice on replacement Date: Thu, 26 Nov 2020 12:54:51 -0500
> That sounds eminently sensible. Scouring the rest of Dave's file, I
> found the following comment at the beginning of the function you've
> just quoted from:
Hmm... seems relevant indeed.
> (defun multi-install-mode (mode &optional chunk-fn base)
> "Add MODE to the multiple major modes supported by the current buffer.
> CHUNK-FN, if non-nil, is a function to select the mode of a chunk,
> added to the list `multi-chunk-fns'. BASE non-nil means that this
> is the base mode."
> (unless (memq mode multi-indirect-buffers-alist) ; be idempotent
> ;; This is part of a grim hack for lossage in AUCTeX, which
> ;; bogusly advises `hack-one-local-variable'. This loses, due to
> ;; the way advice works, when we run `multi-hack-local-variables'
> ;; below -- there ought to be a way round this, probably with CL's
> ;; flet. Any subsequent use of it then fails because advice has
> ;; captured the now-unbound variable `late-hack'... Thus ensure
> ;; we've loaded the mode in advance to get any autoloads sorted
> ;; out. Do it generally in case other modes have similar
> ;; problems. [The AUCTeX stuff is in support of an undocumented
> ;; feature which is unnecessary and, anyway, wouldn't need advice
> ;; to implement. Unfortunately the maintainer seems not to
> ;; understand the local variables mechanism and wouldn't remove
> ;; this. To invoke minor modes, you should just use `mode:' in
> ;; `local variables'.]
I'm not sure I understand the details of the problem described, but
I think I see the general idea. Indeed when multi-mode.el does:
(let ((late-hack (symbol-function 'hack-one-local-variable)))
(fset 'hack-one-local-variable
(lambda (var val)
(unless (eq var 'mode)
(funcall late-hack var val))))
(unwind-protect
(hack-local-variables)
(fset 'hack-one-local-variable late-hack))))
if the code run within `hack-local-variables` ends up looking at the
definition of `hack-one-local-variable` and saving it somewhere for
later use (e.g. if `defadvice` is used at this moment, maybe because
AUCTeX was autoloaded during the course of running
`hack-local-variables`), then we could have a problem. To some extent
the final `fset` above should hide the problem, but it probably just
ends up putting advice in an inconsistent state or worse (depending on
the version of `advice`).
I think with the new version of `advice.el` (introduced when
`nadvice.el` was introduced, to make them interact tolerably) the
problem should be minor: the final `fset` should not put `advice.el` in
an inconsistent state but should simply undo the `defadvice` that
happened during `hack-local-variables` (which is still undesirable,
arguably, but should be mostly harmless as long as you don't rely on
the feature implemented by that advice).
> So I'm guessing that's what he's referring to.
Indeed.
You might like to try the patch below (guaranteed 100% untested).
Do you think it would be worthwhile to add it to GNU ELPA?
Stefan
diff --git a/multi-mode.el b/multi-mode.el
index cece376..06291cf 100644
--- a/multi-mode.el
+++ b/multi-mode.el
@@ -1,6 +1,6 @@
-;;; multi-mode.el --- support for multiple major modes
+;;; multi-mode.el --- support for multiple major modes -*- lexical-binding: t; -*-
-;; Copyright (C) 2003, 2004, 2007, 2009 Free Software Foundation, Inc.
+;; Copyright (C) 2003-2020 Free Software Foundation, Inc.
;; Author: Dave Love <fx <at> gnu.org>
;; Keywords: languages, extensions, files
@@ -164,27 +164,26 @@ Buffer local.")
"Original value of `imenu-create-index-function' for the buffer's mode.")
(make-variable-buffer-local 'multi-late-index-function)
+(define-obsolete-variable-alias 'multi-alist 'multi-mode-alist nil)
(defvar multi-mode-alist nil
"Alist of elements (MODE . FUNCTION) specifying a buffer's multiple modes.
MODE is a major mode and FUNCTION is a function used as an element of
`multi-chunk-fns' or nil. Use nil if MODE is detected by another element
of the alist.")
-(if (fboundp 'define-obsolete-variable-alias)
- (define-obsolete-variable-alias 'multi-alist 'multi-mode-alist)
- (make-obsolete-variable 'multi-alist 'multi-mode-alist))
-
;; See the commentary below.
(defun multi-hack-local-variables ()
"Like `hack-local-variables', but ignore `mode' items."
- (let ((late-hack (symbol-function 'hack-one-local-variable)))
- (fset 'hack-one-local-variable
- (lambda (var val)
- (unless (eq var 'mode)
- (funcall late-hack var val))))
+ (let ((f (lambda (orig-fun var val)
+ (unless (eq var 'mode)
+ (funcall orig-fun var val)))))
(unwind-protect
- (hack-local-variables)
- (fset 'hack-one-local-variable late-hack))))
+ (progn
+ (advice-add 'hack-one-local-variable :around f)
+ (hack-local-variables))
+ (advice-remove 'hack-one-local-variable f))))
+
+(defvar multi-mode)
(defun multi-install-mode (mode &optional chunk-fn base)
"Add MODE to the multiple major modes supported by the current buffer.
@@ -235,21 +234,7 @@ is the base mode."
(funcall mode))
;; Now we can make it local:
(set (make-local-variable 'multi-mode) t)
- ;; Use file's local variables section to set variables in
- ;; this buffer. (Don't just copy local variables from the
- ;; base buffer because it may have set things locally that
- ;; we don't want in the other modes.) We need to prevent
- ;; `mode' being processed and re-setting the major mode.
- ;; It all goes badly wrong if `hack-one-local-variable' is
- ;; advised. The appropriate mechanism to get round this
- ;; appears to be `ad-with-originals', but we don't want to
- ;; pull in the advice package unnecessarily. `flet'-like
- ;; mechanisms lose with advice because `fset' acts on the
- ;; advice anyway.
- (if (featurep 'advice)
- (ad-with-originals (hack-one-local-variable)
- (multi-hack-local-variables))
- (multi-hack-local-variables))
+ (multi-hack-local-variables)
;; Indentation should first narrow to the chunk. Modes
;; should normally just bind `indent-line-function' to
;; handle indentation.
@@ -290,9 +275,9 @@ is the base mode."
;; Kill the base buffer along with the indirect one; careful not
;; to infloop.
(add-hook 'kill-buffer-hook
- '(lambda ()
- (setq kill-buffer-hook nil)
- (kill-buffer (buffer-base-buffer)))
+ (lambda ()
+ (setq kill-buffer-hook nil)
+ (kill-buffer (buffer-base-buffer)))
t t)
;; This should probably be at the front of the hook list, so
;; that other hook functions get run in the (perhaps)
@@ -306,7 +291,7 @@ is the base mode."
(setq buffer-file-coding-system coding)
;; For benefit of things like VC
(setq buffer-file-name file)
- (vc-find-file-hook))
+ (vc-refresh-state))
;; Propagate updated values of the relevant buffer-local
;; variables to the indirect buffers.
(dolist (x alist)
@@ -373,8 +358,7 @@ Fontifies chunk-by-chunk within the region from START for up to
Works piece-wise in all the chunks with the same major mode.
Assigned to `imenu-create-index-function'."
(let ((selected-mode major-mode)
- imenu-alist ; accumulator
- last mode)
+ imenu-alist) ; accumulator
(multi-map-over-chunks
(point-min) (point-max)
(lambda ()
@@ -454,8 +438,7 @@ Destructively modifies `multi-mode-list' to avoid consing in
Return a list (MODE START END), the value returned by the function in the
list for which START is closest to POS (and before it); i.e. the innermost
mode is selected. POS defaults to point."
- (let ((fns multi-chunk-fns)
- (start (point-min))
+ (let ((start (point-min))
(mode (with-current-buffer (multi-base-buffer)
major-mode))
(end (point-max))
@@ -500,28 +483,6 @@ mode is selected. POS defaults to point."
(fundamental-mode)
(error "`multi-mode-alist' not defined for multi-mode")))
-;; In 21.3, Flyspell breaks things, apparently by getting an error in
-;; post-command-hook and thus clobbering it. In development code it
-;; doesn't do that, but does check indirect buffers it shouldn't. I'm
-;; not sure exactly how this happens, but checking flyspell-mode in
-;; the hook functions cures this. For the moment, we'll hack this up.
-;; (Let's not bring advice into it...)
-(eval-after-load "flyspell"
- '(progn
- (defalias 'flyspell-post-command-hook
- `(lambda ()
- ,(concat (documentation 'flyspell-post-command-hook)
- "\n\n[Wrapped by multi-mode.]")
- (if flyspell-mode
- (funcall ,(symbol-function 'flyspell-post-command-hook)))))
-
- (defalias 'flyspell-pre-command-hook
- `(lambda ()
- (concat (documentation 'flyspell-pre-command-hook)
- "\n\n[Wrapped by multi-mode.]")
- (if 'flyspell-mode
- (funcall ,(symbol-function 'flyspell-pre-command-hook)))))))
-
;; This is useful in composite modes to determine whether a putative
;; major mode is safe to invoke.
(defun multi-mode-major-mode-p (value)
GNU bug tracking system
Copyright (C) 1999 Darren O. Benham,
1997,2003 nCipher Corporation Ltd,
1994-97 Ian Jackson.