Check yas--original-auto-fill-function before using it

[chep]

Sometimes when using tramp, I've got an error (Symbol's function definition is
void: nil) because the variable yas--original-auto-fill-function is not
initialized.

A simple check of the value solved my problem.

[npostavs]

Can you explain how to reproduce the problem? I think the real bug is that yas--original-auto-fill-function is not being set.

[chep]

I can't. I can't reproduce it when starting with emacs -Q and my configuration is not simple.
I'll try to find a way to reproduce it.

[npostavs]

Even if you can't reproduce with emacs -Q, if you can set debug-on-error and get a backtrace that could be helpful.

[chep]

I just reproduced it but I don't know how.
It happens when I do a query-replace (M-%) and then press the spacebar (in the minibuffer).
It worked 1 hour ago, it is now broken...

Here is the backtrace:

Debugger entered--Lisp error: (void-function nil)
  nil()
  funcall(nil)
  (let ((yas--inhibit-overlay-hooks t)) (funcall yas--original-auto-fill-function))
  (let* ((orig-point (point)) (end (progn (forward-paragraph) (point))) (beg (progn (backward-paragraph) (point))) (snippets (yas-active-snippets beg end)) (remarkers nil) (reoverlays nil)) (let ((--dolist-tail-- snippets) snippet) (while --dolist-tail-- (setq snippet (car --dolist-tail--)) (let ((--dolist-tail-- (yas--collect-snippet-markers snippet)) m) (while --dolist-tail-- (setq m (car --dolist-tail--)) (if (and (<= beg m) (<= m end)) (progn (setq remarkers (cons (yas--snapshot-marker-location m beg end) remarkers)))) (setq --dolist-tail-- (cdr --dolist-tail--)))) (setq reoverlays (cons (yas--snapshot-overlay-location (progn (or (and (memq (type-of snippet) cl-struct-yas--snippet-tags) t) (signal 'wrong-type-argument (list 'yas--snippet snippet))) (aref snippet 5)) beg end) reoverlays)) (setq --dolist-tail-- (cdr --dolist-tail--)))) (goto-char orig-point) (let ((yas--inhibit-overlay-hooks t)) (funcall yas--original-auto-fill-function)) (save-excursion (setq end (progn (forward-paragraph) (point))) (setq beg (progn (backward-paragraph) (point)))) (save-excursion (save-restriction (narrow-to-region beg end) (mapc (function yas--restore-marker-location) remarkers) (mapc (function yas--restore-overlay-location) reoverlays)) (mapc (function (lambda (snippet) (let ((envvar (progn (or (and (memq (type-of snippet) cl-struct-yas--snippet-tags) t) (signal 'wrong-type-argument (list 'yas--snippet snippet))) (aref snippet 1)))) (progn (let* ((syms (mapcar (function car) envvar)) (vals (mapcar (function (lambda (v-f) (eval (car (cdr v-f))))) envvar)) (body (function (lambda nil (yas--update-mirrors snippet)))) (binds nil)) (while syms (setq binds (cons (list (car-safe (prog1 syms (setq syms (cdr syms)))) (list 'quote (car-safe (prog1 vals (setq vals (cdr vals)))))) binds))) (eval (list 'let binds (list 'funcall (list 'quote body))))))))) snippets)))
  yas--auto-fill()
  internal-auto-fill()
  self-insert-command(1)
  funcall-interactively(self-insert-command 1)
  call-interactively(self-insert-command nil nil)
  command-execute(self-insert-command)
  read-from-minibuffer(#("Query replace (default (0), →  = 0;): " 27 30 (separator t face minibuffer-prompt display " → ")) nil nil nil nil "waitpid" t)
  query-replace-read-from("Query replace" nil)
  query-replace-read-args("Query replace" nil)
  byte-code("\301\302\010\203\024\0\010\303=\203\020\0\304\202\025\0\305\202\025\0\306\307 \203\036\0\310\202\037\0\306Q\311\"\211@\001A@\312\0038\307 \2051\0\313 \307 \2058\0\314 \315\006\0068\307 \205C\0\316 \257\007\207" [current-prefix-arg query-replace-read-args "Query replace" - " backward" " word" "" use-region-p " in region" nil 2 region-beginning region-end 3 region-noncontiguous-p] 8)
  call-interactively(query-replace nil nil)
  command-execute(query-replace)

[npostavs]

It happens when I do a query-replace (M-%) and then press the spacebar (in the minibuffer).
[...]

yas--auto-fill()
internal-auto-fill()
self-insert-command(1)
funcall-interactively(self-insert-command 1)
call-interactively(self-insert-command nil nil)
command-execute(self-insert-command)
read-from-minibuffer(#("Query replace (default (0), → = 0;): " 27 30 (separator t face minibuffer-prompt display " → ")) nil nil nil nil "waitpid" t)
query-replace-read-from("Query replace" nil)

Hmm, I don't think space is supposed to perform self-insert-command in a query replace buffer. Can you check the current buffer in those stack frames (move point to a line in the backtrace, then e (current-buffer)RET).

(defvar query-replace-map
  (let ((map (make-sparse-keymap)))
    (define-key map " " 'act)
  [...]

[npostavs]

@chep Are you using Emacs 26? Based on the properly escaped NUL bytes in your backtrace I guess you are. Could you try the following to hopefully catch where yas--original-auto-fill-function becomes nil? EDIT: this code is too eager, it will catch false positives. Don't use it.

(when nil ; don't use this
 (add-variable-watcher
 'yas--original-auto-fill-function
 (lambda (sym newval op where)
   (when (and (null newval) (eq auto-fill-function 'yas--auto-fill))
     (debug nil "yas--original-auto-fill-function unexpectedly nil!"))))
)

[chep]

I'm using emacs 27 (master branch). The watcher is now in my .emacs file...

[npostavs]

I've now (conditionally) added the watcher to yasnippet, so you can remove it from your .emacs after updating #920.

[stardiviner]

I got following error today. Noticed yasnippet is updated, and I upgrade it. I have not changed yasnippet config recently.

Debugger entered: ("`yas--original-auto-fill-function' unexpectedly nil! Please report this backtrace (hit `c' to continue)")
  yas--watch-auto-fill(yas--original-auto-fill-function nil makunbound #<buffer *scratch*>)
  kill-all-local-variables()
  prog-mode()
  emacs-lisp-mode()
  lisp-interaction-mode()
  command-line()
  normal-top-level()

[drot]

Managed to trigger it with Magit:

GNU Emacs 26.0.91 (build 1, i686-pc-linux-gnu, GTK+ Version 3.22.28) of 2018-03-15

Debugger entered: ("`yas--original-auto-fill-function' unexpectedly nil! Please report this backtrace (hit `c' to continue)")
  yas--watch-auto-fill(yas--original-auto-fill-function nil makunbound #<buffer COMMIT_EDITMSG>)
  kill-all-local-variables()
  normal-mode(t)
  git-commit-setup()
  git-commit-setup-check-buffer()
  run-hooks(find-file-hook)
  after-find-file(nil t)
  find-file-noselect-1(#<buffer COMMIT_EDITMSG> "~/dotfiles/.git/COMMIT_EDITMSG" nil nil "~/dotfiles/.git/COMMIT_EDITMSG" (2633930 2054))
  find-file-noselect("/home/drot/dotfiles/.git/COMMIT_EDITMSG")
  #f(compiled-function (files proc &optional nowait) "Find FILES and return a list of buffers created.\nFILES is an alist whose elements are (FILENAME . FILEPOS)\nwhere FILEPOS can be nil or a pair (LINENUMBER . COLUMNNUMBER).\nPROC is the client that requested this operation.\nNOWAIT non-nil means this client is not waiting for the results,\nso don't mark these buffers specially, just visit them normally." #<bytecode 0x200000000d498300>)((("/home/drot/dotfiles/.git/COMMIT_EDITMSG")) #<process server <1>> nil)
  apply(#f(compiled-function (files proc &optional nowait) "Find FILES and return a list of buffers created.\nFILES is an alist whose elements are (FILENAME . FILEPOS)\nwhere FILEPOS can be nil or a pair (LINENUMBER . COLUMNNUMBER).\nPROC is the client that requested this operation.\nNOWAIT non-nil means this client is not waiting for the results,\nso don't mark these buffers specially, just visit them normally." #<bytecode 0x200000000d498300>) ((("/home/drot/dotfiles/.git/COMMIT_EDITMSG")) #<process server <1>> nil))
  server-visit-files((("/home/drot/dotfiles/.git/COMMIT_EDITMSG")) #<process server <1>> nil)
  server-execute(#<process server <1>> (("/home/drot/dotfiles/.git/COMMIT_EDITMSG")) nil nil nil nil nil)
  #f(compiled-function () #<bytecode 0x200000000d4d0280>)()
  server-execute-continuation(#<process server <1>>)
  server-process-filter(#<process server <1>> "-dir /home/drot/dotfiles/ -current-frame -tty /dev/pts/1 dumb -file /home/drot/dotfiles/.git/COMMIT_EDITMSG \n")

[npostavs]

Ah, I should have tested out my debug code more thoroughly, the kill-all-local-variables case should be skipped. Pushed a new update for that (#921).

[npostavs]

As mentioned in #919 (comment), nobody seems to be hitting this anymore, so I'm going to close.