[Pkg-utopia-maintainers] Bug#860262: flatpak: fails to download any platform etc (sticks at 0%)

[Simon McVittie]

Control: tags 860262 + help

On Thu, 13 Apr 2017 at 18:34:04 +0100, Jonathan Dowland wrote:
> Fresh stretch install via d-i rc3; then install flatpak. Attempt to
> flatpak install <some flatpakref, I chose Corebird from the flatpak
> website). This prompts the installation of a runtime. The installation
> of such stalls at 0%

... sometimes. I think this is a race condition, unfortunately.
I've occasionally seen it hang like this, but never when I was in a
position to spend a lot of time debugging it.

This might be related to the fact that libostree (which is what Flatpak
uses to get its data) usually (but not always!) fails test-local-pull.sh
at least once out of three attempts on ci.debian.net, but has never
failed this test repeatably on my laptop where I could debug it...

> I think I discussed this on IRC with smcv and he
> thinks it might possibly be a libsoup bug?

I thought it might be
<https://bugzilla.gnome.org/show_bug.cgi?id=778135>, but Philip Withnall
(who was looking into that bug for Endless) didn't think the version of
ostree we have in Debian should be susceptible to that particular bug,
because (unlike Endless at the time) we have a libostree version that
is believed to work around it by avoiding the problematic pattern
<https://github.com/ostreedev/ostree/commit/535033a4f01f772e696ad2201b34403203b98bea>.

If you can reproduce this, a gdb `thread apply all bt full`
from interrupting the stuck download with a signal would probably be
really useful.

Trying ostree from experimental would also be useful input, if you can
reproduce it reliably enough.

I suspect that bad things (hang, segfault, whatever) during download
are likely to be a bug in either libsoup or ostree, rather than actually
flatpak.

    S