These are chat archives for abranson/rockpool

25th
Jul 2016
Andrew Branson
@abranson
Jul 25 2016 07:26
So the app deletion works, but the deletion on the pebble.com side can't be working because the apps get restored later. They get launched on the watch when that happens too.
ruff
@rufferson
Jul 25 2016 07:36
Eh, sorry, i didn't get that, what is that deletion from pebble.com side?
Andrew Branson
@abranson
Jul 25 2016 08:00
if i remove a watchapp, it works, but reinstalls itself a few hours later
i have the app locker sync enabled, so i assumed it was deleting locally only
ruff
@rufferson
Jul 25 2016 08:03
Weird, the previous crash was during removal from locker, and that§s what i fixed, so it should be removed. Can zou check logs while removing the app - it should write a line regarding locker cleanup - first check apps missing lockally, then add apps to the locker which are missing there, and finally remove apps detected on first phase from the locker. so that last thing - does it produce any errors?
Andrew Branson
@abranson
Jul 25 2016 08:04
will do!
ruff
@rufferson
Jul 25 2016 08:05
These debug lines
qDebug() << "Removing" << id << m_locker.value(id).value("title").toString() << "from locker";
qWarning() << "Error deleting" << id << rpl->errorString();
Andrew Branson
@abranson
Jul 25 2016 08:05
do you know if I can check the locker contents some other way than in rockpool? on their website?
ruff
@rufferson
Jul 25 2016 08:05
yes you can using curl
Andrew Branson
@abranson
Jul 25 2016 08:06
ok thx :)
ruff
@rufferson
Jul 25 2016 08:06
curl -H "Authorization: Bearer <authToken>" "https://api2.getpebble.com/v2/locker/"
current authToken is stored at .local/share/rockpoold/<mac>/timeline/sync.ini
then you can use smth like http://jsonviewer.stack.hu/ to browse the json
Andrew Branson
@abranson
Jul 25 2016 08:15
cool thx
got to the bottom of the mime thing - one of coderus' utils.
ruff
@rufferson
Jul 25 2016 08:19
so is it ours? or system?
Andrew Branson
@abranson
Jul 25 2016 08:24
nothing to do with us. coderus's plugin is setting the browser mimetypes for every android app, which is confusing update-desktop-database.
ruff
@rufferson
Jul 25 2016 08:25
ah, so it's 3d party plugin installed on your phone, not even system? Or it was adopted and included to the new OS release?
Andrew Branson
@abranson
Jul 25 2016 08:25
nope, only in coderus' openrepo
it's really useful though, adds an 'android' option to the share menu, which passes the object to the android sharing UI. i used it to share photos on android telegram for example.
do you know him?
ruff
@rufferson
Jul 25 2016 08:27
ok, so it's not sneaky attempt from jolla to prevent sideloaded apps :)
Andrew Branson
@abranson
Jul 25 2016 08:27
No :)
anyway it happened on every install that triggered the db update, not just ours
ruff
@rufferson
Jul 25 2016 08:27
no, i don't know him, i only found his repo while searching for websockets
Andrew Branson
@abranson
Jul 25 2016 08:28
this was just the first thing i installed after i updated to the latest 2.0.2
ruff
@rufferson
Jul 25 2016 08:38
Hopefully I'll go for short vacation today, with limited inet access (just mobile), so will try to finish new pebble apps in the offline countryside :) I.e. won't be able to react promptly to any found issues in the current master
till friday that is
Andrew Branson
@abranson
Jul 25 2016 08:55
ok!
btw, here's that memory leak report:
ruff
@rufferson
Jul 25 2016 09:07
interesting, it's new timeline engine so it's from my beta, and also it seems it's already with end-call crash fix, so it's quite fresh
could be the call handler object is not properly released now, need to put a trace on it's destructor and see whether it actually fires up
ruff
@rufferson
Jul 25 2016 09:20
is it reproducable?
Andrew Branson
@abranson
Jul 25 2016 13:12
i haven't reproduced it yet. it's hard to say whether it's a version from before you finished tweaking the call handers though.
but it seems to be quite rapid, so I don't think it's a slow accumulation of memory
ruff
@rufferson
Jul 25 2016 13:16
yes, also there's quite a handful of pins - cca 70 - so i wonder how much memory it eats on fresh start. Mine with avg number of 20 pins a day is having quite low footprint, like 20M rss. there it was 1.5G rss... huge one
pin object is not big though, just a json + wrappers
of course there could be some malicious pin with json document of 1G size %)
but with dbus message limitation of 1MB i can hardly imagine that to happen
Andrew Branson
@abranson
Jul 25 2016 13:21
even with a duplicating pin it seems unreasonable, unless they're accumulating silently in the background
ruff
@rufferson
Jul 25 2016 13:22
and according to logs it started after some abnormal call activity
Andrew Branson
@abranson
Jul 25 2016 13:22
so you don't think this is fixed then?
ruff
@rufferson
Jul 25 2016 13:58
no, i think we've consumed some dangling pointer and that caused entire heap to be allocated
here: Jul 09 22:00:04 Sailfish [4219]: [D] VoiceCallManager::onVoiceCallsChanged:272 - removing "b4df7f437d1c4cab8bc221bda1e62fa1""Added action: 0 image://theme/icon-cover-dialer"
it should be qDebug() << "removing" << removeId << "which is" << handler->handlerId();
so the remaining part of the message is not showed so the handlerId was probably that dangling pointer
and after that it got oom
Jul 09 22:00:09 Sailfish kernel: misc memnotify: 0xf322ef40 - threshold updated to 0x1
Jul 09 22:00:11 Sailfish kernel: misc memnotify: 0xf322e040 - threshold updated to 0x1
Jul 09 22:00:13 Sailfish kernel: lowmemorykiller: Killing 'jolla-settings' (4281), adj 70,
but yea not sure whether it was before or after call crash fix. call crash was also due to dangling pointer but that was actually used in destructor so it asserted and segfaulted
ruff
@rufferson
Jul 25 2016 14:08
before crash fix debug looked like
qDebug() << "removing " << handler->handlerId();
so could be both. but i suspect rather previous, so now should be fixed
Andrew Branson
@abranson
Jul 25 2016 15:06
ok, i'll keep an eye out for it. have a good holiday!