On Fri, 2017-12-15 at 08:55 -0600, Michael Cronenworth wrote:
On 12/14/2017 09:46 PM, Adam Williamson wrote:
> Sorry, totally forgot the second part of the post:
> Meanwhile, I recalled we actually have a RHBZ and a github report for
> the abrt false-dupe side of things already. I have updated those too:
> They've made a change upstream to sort of blacklist (AIUI) _g_log_abort
> and another function in the commonly-encountered trace, and sent it to
> Rawhide, but not sent it out as an update for F26 and F27 yet; I asked
> if they can do that. I also suggested another variant path to
> _g_log_abort which may indicate another function or two to add to the
Thanks, Adam, for your due diligence and sorting out the mess.
No problem. Since I just *love* work, today I had the great idea of
looking through all open gnome-shell bugs for other ones reported by
abrt which reference _g_log_abort - and surprise surprise, I found
several others like this. I think abrt considers the distribution
version and the signal to be key attributes of a crash - so it won't
consider two bugs as dupes if they're for different releases or with
different signals, even if it thinks the traceback is the same. So we
get a few of these bugs per release and per signal. So I'm working
gradually through all the ones I've found. One big one was this one,
which had 81 dupes. I've got several more to look at next week.
The dupes turned out to be a mix of one-offs, Xwayland crashes without
sufficient data to identify them, and a few common cases.
1. One that cropped up a *hell* of a lot was this error:
"Finalizing proxy for an object that's scheduled to be unrooted:
I came across a few of these in the F25 and F27 bugs, but there were a
*lot* of cases reported against F26 around July to September last year.
For a while I wasn't duping these bugs as I wasn't sure they were
dupes, but there's *so* many with the same error it seems like there
probably is some common cause, so I started making them dupes of
after a while. Even
though the reports tailed off after September I'm not sure that means
the bug went away - it might just mean that all the people who
encountered it and are willing to file bug reports had already filed
their report, so they wouldn't create any more new ones. There *are*
some more recent reports, like these:
it'd be great if anyone could figure out what the hell's going on
2. There were also quite a lot more cases of "invalid global
wl_output", which I duped against
- it'd be good if
someone could apply the proposed fix for that to F26 as well as F27,
since it's clearly affecting F26 users too.
3. There were several cases of Xwayland crashes caused by two variants
on an intel driver error "intel_do_flush_locked failed", which I duped
4. There were several cases of errors along the lines of
"Settings schema (SOMESCHEMA) does not contain a key named (SOMEKEY)".
I *think* these are mostly misbehaving extensions, but I did wonder if
an error like that really has to cause the entire shell to crash?
is one example of
this type of error, there were several others for different extensions.
5. There were quite a few bugs which were basically cases where the
Shell or Xwayland crashed due to exhaustion of open file descriptors.
These seem to fall into a few different buckets. There's one which Ray
believes is caused by the Shell leaking timerfds on background change:
, and he's sent out
a proposed fix. There were a couple of others that seem related to
backgrounds, but had large numbers of fds open that were actually
related to wallpaper files and directories; I'm starting to suspect
these are caused by the "Random Walls" extension -
- and am working with the
reporters to confirm that. There were a couple of reports where most of
the open FDs were for /sys/class/power_supply ; we tentatively suspect
that may be a leak in upower and have assigned
to track that. And
there were a few reports where most of the open FDs were some sort of
. Finally, in
, most of the open
FDs appear to be some kind of *socket*.
6. The parent bugs for both F25 and F26 turned out to be interesting
cases in their own right. The error in the F25 parent,
, is "toggling down
object GSettings that's already queued to toggle up", which I came
across a few more reports of for F26 later: #1484728 , #1492312 and
#1502171 . kparal also apparently hit that one, per
7. The error in the F26 parent,
"Error in freeze/thaw accounting" (an error condition in mutter), and
again I found another case of that later,
. Also the first
commenter really *did* have the same bug as the OP there (piio and Igor
I'll carry on working through all of these tomorrow :) Thanks folks!
Fedora QA Community Monkey
IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net