Buffer file replaced with lockfile/backupfile location

Question

I have been suffering this issue for a few months, unable to reproduce it until now.

In short, the issue is that when creating a new file, editing it, and then deciding to rename the file, the buffer's file is replaced by the lockfile location and not the original file location.

I do not have a minimal config which reliably reproduces the issue, but I may have a sufficient cross-section to get close.

I have the following changes to lockfiles and auto-save:

(defconst --lockfile-directory (concat user-emacs-directory "lockfiles/")
(setq auto-save-file-name-transforms `((".*" ,--lockfile-directory t)))
(setq auto-save-default t)
(setq auto-save-timeout 20)
(setq auto-save-interval 200)

Steps to reproduce:

1. Create new file and add some text

2. Save the file

3. Rename the file with write-file

4. Make some changes to the file

5. The buffer is now in the lockfile.

Since there's a pause, I suspect a timer is at play, but listing my timers, I don't see anything that stands out except maybe the auto-revert-buffers timer.

List of timers:

               6.0s         5.0s auto-revert-buffers
               8.8s            - undo-auto--boundary-timer
   *           0.0s            t jit-lock-deferred-fontify
   *           0.1s            t show-paren-function
   *           0.2s            t ggtags-highlight-tag-at-point
   *           0.5s      :repeat blink-cursor-start
   *           0.5s            t #<subr F616e6f6e796d6f75732d6c616d626461_anonymous_lambda_14>
   *           0.5s            t #<subr F616e6f6e796d6f75732d6c616d626461_anonymous_lambda_12>
   *           0.5s            t posframe-hidehandler-daemon-function
   *           1.0s       repeat ess--idle-timer-function
   *           1.0s            t semantic-idle-scheduler-function
   *           1.2s            t reftex-view-crossref-when-idle
   *          30.0s            t js-gc
   *        1m 0.0s            t semantic-idle-scheduler-work-function
   *        1m 0.0s            t bibtex-parse-buffers-stealthily

dotfiles mostly up to date, but the backups configuration has been disabled, but I haven't pushed the changes.

Because it probably matters, here is the version information:

GNU Emacs 28.0.91 (build 1, x86_64-pc-linux-gnu, GTK+ Version 3.24.30, cairo version 1.16.0)
GNU Guix SD @ 207a1625e937f916506d00ba1407c5baf27791b6

For further clarity, this is the native elisp compiled Emacs, though that doesn't matter, I've seen this issue before I switched to native branch.

Related Questions

• How to autosave a buffer to the original file on disk?

  I see from here, that perhaps the issue is that I change the location of the auto-save file and, therefore, when explicitly saving the file, emacs doesn't know how to find the original file being visited?

  Related, I see the down sides of toggling auto-save-visited-file-name, so if possible, avoiding that "solution" is desirable.

Minimal Config (emacs -Q)

I have paired down the configuration to the following to reproduce the issue with the above steps.

(setq auto-save-visited-file-name t)
(setq auto-save-default t)
(setq auto-save-timeout 20)
(setq auto-save-interval 20)

The timeout and interval value are likely not contributing, but usefully small for reproducing the issue.

I forgot I am in fact setting auto-save-visited-file-name which turns out to be the issue. If I disable this, the issue goes away.

Now, it remains to be said if changing the buffer to the lockfile location is the correct behavior for auto-save-visited-file-name.

Answer 1

It helps to read the fine variable descriptions.

Namely, the following about auto-save-visited-file-name:

This variable is obsolete since Emacs 26.1; use `auto-save-visited-mode' instead. You can customize this variable. Probably introduced at or before Emacs version 26.1.

Changing the above minimal example to the following, the issue goes away:

(auto-save-visited-mode)
(setq auto-save-default t)
(setq auto-save-timeout 20)
(setq auto-save-interval 20)

So it may in fact be a "bug". However, I don't expect it to be fixed given that the "proper" way to set this behavior is through the minor mode auto-save-visited-mode.