Bug#928662: A quick and dirty observations for GNOME mutter about this

[hoxp18 at noramail.jp]

To: 928662-maintonly at bugs.debian.org
Subject: A quick and dirty observations for GNOME mutter about this

Dear maintainers,

Summary
=======

Debian buster GNOME/Wayland apps produces a lot of
last_user_time GNOME mutter warnings in syslog.
Debian stretch, GNOME/X.org does not.

As I mentioned at my last post,
this issue is a combination of log flood and user privacy.

1. Too many last_user_time related GNOME mutter warnings,
   almost same content, occurring in many apps
   (at least evince, Firefox, gedit).

2. They contain each windows' title,
   which may includes which file a user editing and/or
   where a user browsing and so on, in syslog.

Scenarios
---------

1) A box is just posting logs to its own /var/log/syslog only.
   #2 is not so much important.

2) Logs are managed by central syslog server.
   both #1 and #2 are problem.

3) /var/log is in SSD/NVMe/eMMC devices.
   #1 is a bit problem.

   As you know, SSD/NVMe has write I/O limitation,
   even they are significantly improved their lifetime.

   * QLC models are not so good for a lot of tiny writes.
 
   * Most of amd64 PCs and so many arm64 devices
     only has SSD/NVMe/eMMC.

   * This bug (?) posts tiny repeated log entries,
     may cause small and many write operations,
     which I think harmful unless those logs are important.

4) logcheck and related log management trouble.
   #1 could make those reports filled by too many entries.

   Administrators have to do something to exclude them,
   or could miss important logs.

My observation for GNOME mutter
===============================

I think those logs are generated by GNOME mutter,
so I did a small survey about this issue,
but I have no idea about the cause
(at first I thought it could be a time rag somewhere
around AppArmor and/or ibus, but now I have doubt around
each GUI app codes and/or GUI middlewares).

GNOME mutter source review
---------------------------

src/core/display.c: meta_display_sanity_check_timestamps
  This calls
  meta_warning ("%s appears to be one of the offending windows "
  ...
  %s is window->desc and may be a string of below codes,
  causing last_user_time related log flood and the privacy issue.

src/core/window.c: meta_window_update_desc
  This sets (pointer) address only.

src/core/stack-tracker.c: meta_stack_op_dump
  #ifdef WITH_VERBOSE_MODE
  This sets get_window_desc result, maybe contains window title.

BTW I have no idea how important meta_display_sanity_check_timestamps
warnings are.

Do GNOME developers/maintainers needs those warnings in syslog,
repeatedly?

My Proposals
============

I did not build mutter nor test it, apology
(I just glanced/greped mutter src in salsa).

A) disable WITH_VERBOSE_MODE for buster and stretch

   cons: GNOME/Wayland related app bugs might be missed.
         last_user_time related log flood (#1) be left intact.

   pros: GNOME/Wayland related bugs can still be tracked in bullseye.
         User privacy issue (#2) might be solved.

B) remove one of meta_warning in meta_display_sanity_check_timestamps

   only a part of "%s appears to be one of the offending windows "

   pros: user window title might not be shown in syslog.
         It may resolve a privacy issue (#2)

   cons: makes syslog entry hard to trace which window caused
         the last_user_time issues.

C) leave them

   pros: Perhaps most of users/maintainers/developers don't care.
         Time and resource saving (except syslog).

   cons: syslog contains users window titles, a privacy issue (#2).
         syslog has too many last_user_time warnings
         which (I think) is not so much important (#1).

D) add a warning counter

   in meta_warning in meta_display_sanity_check_timestamps,
   not to repeat almost same warnings in a short period of time.

   pros: log flood (#1) might be resolved.

   cons: window title issue (#2) be left, a potential privacy issue.
         GNOME mutter logging code would be complicated.

   I myself think this is a bad idea.

Thanks for reading.

Regards