* bug#17351: Trunk emacs splats .emacs.desktop at startup @ 2014-04-26 18:04 Alan Mackenzie 2014-04-27 8:22 ` Juri Linkov 0 siblings, 1 reply; 19+ messages in thread From: Alan Mackenzie @ 2014-04-26 18:04 UTC (permalink / raw) To: 17351 Hello, Emacs. This is horrible. I started a GUI emacs session, and there were errors loading the desktop. I loaded the file .emacs.desktop into my emacs session to see what the problem was. And lo and behold, that file had already been overwritten by Emacs. :-( This is crazy. What's the point of overwriting .emacs.desktop at startup? It's either going to be overwriting itself with a functionally identical copy, or (as in the current case) its going to be losing information needed for debugging. This is surely a bug. My trunk is at revision number 116992. -- Alan Mackenzie. ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-04-26 18:04 bug#17351: Trunk emacs splats .emacs.desktop at startup Alan Mackenzie @ 2014-04-27 8:22 ` Juri Linkov 2014-04-27 9:15 ` Alan Mackenzie 0 siblings, 1 reply; 19+ messages in thread From: Juri Linkov @ 2014-04-27 8:22 UTC (permalink / raw) To: Alan Mackenzie; +Cc: 17351 > This is horrible. I started a GUI emacs session, and there were errors > loading the desktop. I loaded the file .emacs.desktop into my emacs > session to see what the problem was. And lo and behold, that file had > already been overwritten by Emacs. :-( I lost my desktop too. :-( After starting a GUI session, loading the desktop failed to read \.\.\. characters. I see the only way how \.\.\. got into the desktop: some mode temporarily changed the buffer-local value of `print-length' that abbreviated a value while saving the desktop. I added precautions against this problem on the emacs-24 branch. But the remaining problem is that it overwrote the desktop with the broken file while leaving no backups of the previous desktop. > This is crazy. What's the point of overwriting .emacs.desktop at > startup? It's either going to be overwriting itself with a functionally > identical copy, or (as in the current case) its going to be losing > information needed for debugging. If after startup you paused for 30 seconds then auto-saving overwrote your desktop indeed. It seems we need to keep a backup of the desktop from the previous session when saving the desktop in a new session for the case when the desktop gets broken by the errors in it. ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-04-27 8:22 ` Juri Linkov @ 2014-04-27 9:15 ` Alan Mackenzie 2014-04-27 9:52 ` Juri Linkov 0 siblings, 1 reply; 19+ messages in thread From: Alan Mackenzie @ 2014-04-27 9:15 UTC (permalink / raw) To: Juri Linkov; +Cc: 17351 Morning, Juri. On Sun, Apr 27, 2014 at 11:22:39AM +0300, Juri Linkov wrote: > > This is horrible. I started a GUI emacs session, and there were errors > > loading the desktop. I loaded the file .emacs.desktop into my emacs > > session to see what the problem was. And lo and behold, that file had > > already been overwritten by Emacs. :-( > I lost my desktop too. :-( After starting a GUI session, loading the > desktop failed to read \.\.\. characters. I see the only way how > \.\.\. got into the desktop: some mode temporarily changed the buffer-local > value of `print-length' that abbreviated a value while saving the desktop. > I added precautions against this problem on the emacs-24 branch. > But the remaining problem is that it overwrote the desktop with the > broken file while leaving no backups of the previous desktop. > > This is crazy. What's the point of overwriting .emacs.desktop at > > startup? It's either going to be overwriting itself with a functionally > > identical copy, or (as in the current case) its going to be losing > > information needed for debugging. > If after startup you paused for 30 seconds then auto-saving overwrote > your desktop indeed. It seems we need to keep a backup of the desktop > from the previous session when saving the desktop in a new session > for the case when the desktop gets broken by the errors in it. Yes, I think so. This needs to be mentioned in the manual, too. 30 seconds seems a very short time (by default) to trigger this saving of the desktop. Especially when there's been no change since the previous save. -- Alan Mackenzie (Nuremberg, Germany). ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-04-27 9:15 ` Alan Mackenzie @ 2014-04-27 9:52 ` Juri Linkov 2014-04-27 14:31 ` Stefan Monnier 0 siblings, 1 reply; 19+ messages in thread From: Juri Linkov @ 2014-04-27 9:52 UTC (permalink / raw) To: Alan Mackenzie; +Cc: 17351 >> If after startup you paused for 30 seconds then auto-saving overwrote >> your desktop indeed. It seems we need to keep a backup of the desktop >> from the previous session when saving the desktop in a new session >> for the case when the desktop gets broken by the errors in it. > > Yes, I think so. This needs to be mentioned in the manual, too. 30 > seconds seems a very short time (by default) to trigger this saving of > the desktop. 30 seconds of idle time before desktop auto-saving is the default value inherited from `auto-save-timeout' used for file auto-saving. But they don't need to be the same. In ~/.emacs I customized it to 60 seconds, and maybe this default value would be better for everyone. > Especially when there's been no change since the previous save. Actually auto-saving already checks for changes and doesn't write the identical desktop. In practice this happens rarely, only when the session left inactive, because any minuscule change in window configuration (such as displaying a buffer with error messages) causes the desktop file to be overwritten with new contents. ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-04-27 9:52 ` Juri Linkov @ 2014-04-27 14:31 ` Stefan Monnier 2014-04-27 22:13 ` Stefan Monnier 0 siblings, 1 reply; 19+ messages in thread From: Stefan Monnier @ 2014-04-27 14:31 UTC (permalink / raw) To: Juri Linkov; +Cc: Alan Mackenzie, 17351 >> Yes, I think so. This needs to be mentioned in the manual, too. 30 >> seconds seems a very short time (by default) to trigger this saving of >> the desktop. No, it's not long at all. It only looks that way when the event is threatening, so the problem is the threat, not the delay. Stefan ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-04-27 14:31 ` Stefan Monnier @ 2014-04-27 22:13 ` Stefan Monnier 2014-06-22 7:05 ` Glenn Morris 0 siblings, 1 reply; 19+ messages in thread From: Stefan Monnier @ 2014-04-27 22:13 UTC (permalink / raw) To: Juri Linkov; +Cc: Alan Mackenzie, 17351 > No, it's not long at all. It only looks that way when the event is ^^^^ short Sorry, Stfean ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-04-27 22:13 ` Stefan Monnier @ 2014-06-22 7:05 ` Glenn Morris 2014-06-22 23:01 ` Juri Linkov 0 siblings, 1 reply; 19+ messages in thread From: Glenn Morris @ 2014-06-22 7:05 UTC (permalink / raw) To: Stefan Monnier; +Cc: Alan Mackenzie, 17351 As a stopgap, how about disabling the autosave (with a warning, eg "error loading desktop; automatic save disabled until next manual save") if there were errors loading the desktop at startup? More generally, why does the autosaving actually save directly into the real desktop file? Why doesn't it use the normal autosave mechanism, ie write to a #desktop# file? Naively, it seems like this would be both simpler to implement and give better results. ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-06-22 7:05 ` Glenn Morris @ 2014-06-22 23:01 ` Juri Linkov 2014-06-23 0:05 ` Glenn Morris 0 siblings, 1 reply; 19+ messages in thread From: Juri Linkov @ 2014-06-22 23:01 UTC (permalink / raw) To: Glenn Morris; +Cc: Alan Mackenzie, 17351 > As a stopgap, how about disabling the autosave (with a warning, eg > "error loading desktop; automatic save disabled until next manual save") > if there were errors loading the desktop at startup? It's not clear where to catch errors and handle them, because they can occur everywhere and it's better to leave the default error processing that enters the debugger. But without such warning message, it's possible to disable autosaving when loading desktop with a simple patch below. > More generally, why does the autosaving actually save directly into the > real desktop file? Why doesn't it use the normal autosave mechanism, ie > write to a #desktop# file? Naively, it seems like this would be both > simpler to implement and give better results. This would keep the desktop file open all the time, and users might not like to see it in the buffer list. If this is not desirable then a backup copy needs to be saved once per session after reading the desktop. === modified file 'lisp/desktop.el' --- lisp/desktop.el 2014-06-06 23:38:40 +0000 +++ lisp/desktop.el 2014-06-22 22:52:31 +0000 @@ -174,11 +174,8 @@ (define-minor-mode desktop-save-mode :global t :group 'desktop (if desktop-save-mode - (when (and (integerp desktop-auto-save-timeout) - (> desktop-auto-save-timeout 0)) - (add-hook 'window-configuration-change-hook 'desktop-auto-save-set-timer)) - (remove-hook 'window-configuration-change-hook 'desktop-auto-save-set-timer) - (desktop-auto-save-cancel-timer))) + (desktop-auto-save-enable) + (desktop-auto-save-disable))) (defun desktop-save-mode-off () "Disable `desktop-save-mode'. Provided for use in hooks." @@ -219,9 +216,8 @@ (defcustom desktop-auto-save-timeout aut (set-default symbol value) (ignore-errors (if (and (integerp value) (> value 0)) - (add-hook 'window-configuration-change-hook 'desktop-auto-save-set-timer) - (remove-hook 'window-configuration-change-hook 'desktop-auto-save-set-timer) - (desktop-auto-save-cancel-timer)))) + (desktop-auto-save-enable value) + (desktop-auto-save-disable)))) :group 'desktop :version "24.4") @@ -1132,6 +1128,7 @@ (defun desktop-read (&optional dirname) (unless desktop-dirname (message "Desktop file in use; not loaded."))) (desktop-lazy-abort) + (desktop-auto-save-disable) ;; Evaluate desktop buffer and remember when it was modified. (load (desktop-full-file-name) t t t) (setq desktop-file-modtime (nth 5 (file-attributes (desktop-full-file-name)))) @@ -1184,6 +1181,7 @@ (defun desktop-read (&optional dirname) (set-window-prev-buffers window nil) (set-window-next-buffers window nil)))) (setq desktop-saved-frameset nil) + (desktop-auto-save-enable) t)) ;; No desktop file found. (desktop-clear) @@ -1230,6 +1228,15 @@ (defun desktop-save-in-desktop-dir () ;; Auto-Saving. (defvar desktop-auto-save-timer nil) +(defun desktop-auto-save-enable (&optional timeout) + (when (and (integerp (or timeout desktop-auto-save-timeout)) + (> (or timeout desktop-auto-save-timeout) 0)) + (add-hook 'window-configuration-change-hook 'desktop-auto-save-set-timer))) + +(defun desktop-auto-save-disable () + (remove-hook 'window-configuration-change-hook 'desktop-auto-save-set-timer) + (desktop-auto-save-cancel-timer)) + (defun desktop-auto-save () "Save the desktop periodically. Called by the timer created in `desktop-auto-save-set-timer'." ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-06-22 23:01 ` Juri Linkov @ 2014-06-23 0:05 ` Glenn Morris 2014-06-23 22:57 ` Juri Linkov 0 siblings, 1 reply; 19+ messages in thread From: Glenn Morris @ 2014-06-23 0:05 UTC (permalink / raw) To: Juri Linkov; +Cc: Alan Mackenzie, 17351 Juri Linkov wrote: > This would keep the desktop file open all the time, > and users might not like to see it in the buffer list. So read it into an invisible buffer, set buffer-file-name, and enable auto-save-mode. ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-06-23 0:05 ` Glenn Morris @ 2014-06-23 22:57 ` Juri Linkov 2014-06-24 18:30 ` Glenn Morris 0 siblings, 1 reply; 19+ messages in thread From: Juri Linkov @ 2014-06-23 22:57 UTC (permalink / raw) To: Glenn Morris; +Cc: Alan Mackenzie, 17351 >> This would keep the desktop file open all the time, >> and users might not like to see it in the buffer list. > > So read it into an invisible buffer, set buffer-file-name, and enable > auto-save-mode. auto-save-mode doesn't seem necessary because the contents of the desktop file buffer has to be generated at the given time interval anyway, so at the end it can be saved as well like in this patch: === modified file 'lisp/desktop.el' --- lisp/desktop.el 2014-06-06 23:38:40 +0000 +++ lisp/desktop.el 2014-06-23 22:45:20 +0000 @@ -1038,8 +1038,15 @@ (defun desktop-save (dirname &optional r (point)))) (checksum (and beg (md5 (current-buffer) beg (point-max) 'emacs-mule)))) (unless (and checksum (equal checksum desktop-file-checksum)) - (let ((coding-system-for-write 'emacs-mule)) - (write-region (point-min) (point-max) (desktop-full-file-name) nil 'nomessage)) + (let ((coding-system-for-write 'emacs-mule) + (buffer-string (buffer-string))) + (with-current-buffer (find-file-noselect (desktop-full-file-name)) + ;; Make the buffer invisible + (unless (eq (aref (buffer-name) 0) ?\s) + (rename-buffer (concat " " (buffer-name)))) + (erase-buffer) + (insert buffer-string) + (save-buffer))) (setq desktop-file-checksum checksum) ;; We remember when it was modified (which is presumably just now). (setq desktop-file-modtime (nth 5 (file-attributes (desktop-full-file-name))))))))))) ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-06-23 22:57 ` Juri Linkov @ 2014-06-24 18:30 ` Glenn Morris 2014-06-24 23:23 ` Juri Linkov 0 siblings, 1 reply; 19+ messages in thread From: Glenn Morris @ 2014-06-24 18:30 UTC (permalink / raw) To: Juri Linkov; +Cc: Alan Mackenzie, 17351 Juri Linkov wrote: > auto-save-mode doesn't seem necessary because the contents of the > desktop file buffer has to be generated at the given time interval > anyway, so at the end it can be saved as well like in this patch: As I see it, the point is that normal auto-save distinguishes automatic save from a manual save initiated by the user, by using separate files. Whereas desktop's version of auto-save removes that distinction. (But I'm not familiar with the details.) ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-06-24 18:30 ` Glenn Morris @ 2014-06-24 23:23 ` Juri Linkov 2014-06-26 19:31 ` Glenn Morris 2014-07-02 23:52 ` Juri Linkov 0 siblings, 2 replies; 19+ messages in thread From: Juri Linkov @ 2014-06-24 23:23 UTC (permalink / raw) To: Glenn Morris; +Cc: Alan Mackenzie, 17351 > As I see it, the point is that normal auto-save distinguishes automatic > save from a manual save initiated by the user, by using separate files. > Whereas desktop's version of auto-save removes that distinction. > (But I'm not familiar with the details.) It's possible to periodically update the desktop file buffer, but not to save it, thus allowing normal auto-save to create a #desktop# backup. But this might cause some problems, e.g. modes that call `save-some-buffers' like `compilation-mode' will ask whether to save the updated desktop file. Also it's not clear how to restore the auto-saved backup copy after a crash. OTOH, saving the desktop file buffer will create a desktop~ backup that the user can restore when something goes wrong. This is implemented in the latest patch that is intended for the trunk. Meanwhile, I installed the patch that temporarily disables the autosave in `desktop-read' into the emacs-24 branch, so people won't lose the desktop in case of errors. ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-06-24 23:23 ` Juri Linkov @ 2014-06-26 19:31 ` Glenn Morris 2014-06-26 23:44 ` Juri Linkov 2014-07-02 23:52 ` Juri Linkov 1 sibling, 1 reply; 19+ messages in thread From: Glenn Morris @ 2014-06-26 19:31 UTC (permalink / raw) To: Juri Linkov; +Cc: Alan Mackenzie, 17351 Juri Linkov wrote: > But this might cause some problems, e.g. modes that call > `save-some-buffers' like `compilation-mode' will ask whether to save > the updated desktop file. I iamgine a solution could be found. I could equally well say that using save-buffer will cause the desktop to be included in the recentf list, which some people may dislike. (Cf http://debbugs.gnu.org/17155) > Also it's not clear how to restore the auto-saved backup copy after a > crash. Using the normal "file foo has auto-save, recover it" prompt? (This topic makes me wonder if desktop could have used the standard interlock mechanism, rather than inventing its own version of that too.) ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-06-26 19:31 ` Glenn Morris @ 2014-06-26 23:44 ` Juri Linkov 2014-08-11 1:18 ` Glenn Morris 0 siblings, 1 reply; 19+ messages in thread From: Juri Linkov @ 2014-06-26 23:44 UTC (permalink / raw) To: Glenn Morris; +Cc: Alan Mackenzie, 17351 > I could equally well say that using save-buffer will cause the desktop > to be included in the recentf list, which some people may dislike. > (Cf http://debbugs.gnu.org/17155) A solution in bug#17155 was to replace `save-buffer' with `write-region' that is already currently used by desktop.el. This means continuing using low-level `write-region' would be less problematic than `save-buffer' (and normal auto-saving of a file buffer that still requires `save-buffer' at the end of the session). > (This topic makes me wonder if desktop could have used the standard > interlock mechanism, rather than inventing its own version of that too.) This would be possible if the desktop file used `save-buffer' on a normal file buffer instead of `write-region'. ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-06-26 23:44 ` Juri Linkov @ 2014-08-11 1:18 ` Glenn Morris 2014-08-12 6:52 ` Juri Linkov 0 siblings, 1 reply; 19+ messages in thread From: Glenn Morris @ 2014-08-11 1:18 UTC (permalink / raw) To: Juri Linkov; +Cc: Alan Mackenzie, 17351 I lost track of this. Did the "starting Emacs may delete your deskop file" aspect get fixed? If so, this can be downgraded to severity normal (or closed). ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-08-11 1:18 ` Glenn Morris @ 2014-08-12 6:52 ` Juri Linkov 0 siblings, 0 replies; 19+ messages in thread From: Juri Linkov @ 2014-08-12 6:52 UTC (permalink / raw) To: Glenn Morris; +Cc: Alan Mackenzie, 17351-done > I lost track of this. > Did the "starting Emacs may delete your deskop file" aspect get fixed? > If so, this can be downgraded to severity normal (or closed). The fixes already installed in the emacs-24 branch were intended to reduce the number of cases that might break the desktop file, so I'm closing this bug report. The remaining feature to create a backup copy is optional and could be implemented later for the trunk in a new separate feature request. ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-06-24 23:23 ` Juri Linkov 2014-06-26 19:31 ` Glenn Morris @ 2014-07-02 23:52 ` Juri Linkov 2014-07-03 2:42 ` Glenn Morris 1 sibling, 1 reply; 19+ messages in thread From: Juri Linkov @ 2014-07-02 23:52 UTC (permalink / raw) To: Glenn Morris; +Cc: Alan Mackenzie, 17351 > Meanwhile, I installed the patch that temporarily disables the > autosave in `desktop-read' into the emacs-24 branch, so people > won't lose the desktop in case of errors. This handles the case when there are errors in the desktop file. But another case is when the errors are in the init file, and the buffer *Warnings* is displayed. The remaining problem is that the buffer *Warnings* becomes hidden behind the restored window configuration, so the user has no indication about the errors in the init file. ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-07-02 23:52 ` Juri Linkov @ 2014-07-03 2:42 ` Glenn Morris 2014-07-03 23:35 ` Juri Linkov 0 siblings, 1 reply; 19+ messages in thread From: Glenn Morris @ 2014-07-03 2:42 UTC (permalink / raw) To: Juri Linkov; +Cc: Alan Mackenzie, 17351 Juri Linkov wrote: > But another case is when the errors are in the init file, and the buffer > *Warnings* is displayed. The remaining problem is that the buffer > *Warnings* becomes hidden behind the restored window configuration, > so the user has no indication about the errors in the init file. That seems unrelated to this report? ^ permalink raw reply [flat|nested] 19+ messages in thread
* bug#17351: Trunk emacs splats .emacs.desktop at startup 2014-07-03 2:42 ` Glenn Morris @ 2014-07-03 23:35 ` Juri Linkov 0 siblings, 0 replies; 19+ messages in thread From: Juri Linkov @ 2014-07-03 23:35 UTC (permalink / raw) To: Glenn Morris; +Cc: Alan Mackenzie, 17351 >> But another case is when the errors are in the init file, and the buffer >> *Warnings* is displayed. The remaining problem is that the buffer >> *Warnings* becomes hidden behind the restored window configuration, >> so the user has no indication about the errors in the init file. > > That seems unrelated to this report? Fortunately, it doesn't cause the desktop file corruption, so I created a separate bug#17927. It seems for this bug#17351 to be closed, what remains to do is to generate a backup copy of the desktop file. I see two ways: 1. manually write a backup file once after reading the desktop and locking it. This solution fits into the current design of desktop.el. 2. Use normal mechanisms of making a backup and locking by using save-buffer instead of write-region, but this will cause more problems (including in the recentf list, etc.) ^ permalink raw reply [flat|nested] 19+ messages in thread
end of thread, other threads:[~2014-08-12 6:52 UTC | newest] Thread overview: 19+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2014-04-26 18:04 bug#17351: Trunk emacs splats .emacs.desktop at startup Alan Mackenzie 2014-04-27 8:22 ` Juri Linkov 2014-04-27 9:15 ` Alan Mackenzie 2014-04-27 9:52 ` Juri Linkov 2014-04-27 14:31 ` Stefan Monnier 2014-04-27 22:13 ` Stefan Monnier 2014-06-22 7:05 ` Glenn Morris 2014-06-22 23:01 ` Juri Linkov 2014-06-23 0:05 ` Glenn Morris 2014-06-23 22:57 ` Juri Linkov 2014-06-24 18:30 ` Glenn Morris 2014-06-24 23:23 ` Juri Linkov 2014-06-26 19:31 ` Glenn Morris 2014-06-26 23:44 ` Juri Linkov 2014-08-11 1:18 ` Glenn Morris 2014-08-12 6:52 ` Juri Linkov 2014-07-02 23:52 ` Juri Linkov 2014-07-03 2:42 ` Glenn Morris 2014-07-03 23:35 ` Juri Linkov
Code repositories for project(s) associated with this public inbox https://git.savannah.gnu.org/cgit/emacs.git This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).