On Tue, 2016-04-19 at 07:43 -0400, Stephen Gallagher wrote:
On 04/18/2016 05:05 PM, Adam Williamson wrote:
- https://bugzilla.redhat.com/show_bug.cgi?id=1262556 - anaconda - MODIFIED
AttributeError: 'NoneType' object has no attribute 'get_data' (when wifi scan run and AP which does not broadcast SSID is in range)
This ought to be fixed in the new anaconda build which sbueno is just now sending out as an update. We still have the anaconda updates testing chicken/egg problem, but this one should be susceptible to testing by updating anaconda on a live image before running the installer, happily. You need to have a system with wifi and a wireless AP nearby which does not broadcast its SSID: boot a live image, update anaconda, disconnect any wired connection, do `virsh net-destroy default`, then run the installer and see if you can make it through the hub without a crash.
Given that Anaconda has a built-in mechanism for addressing exactly this (the updates.img), shouldn't we perhaps consider doing testing by providing updates.img shortlinks on the BZ (with the diff being from the last-known-good compose)?
updates.img testing is useful for some things, not for everything. For instance, my reproducer for this involves a system with no wired connection, so the standard way to load the updates.img is not so useful ;) You can load them from local media but it's a bit more work. For this specific case I figured using the 'update packages on a live image' approach would be a better test.
Also note that for karma purposes - as opposed to testing the fix before commit - we should really test the *entire* update, not just the specific change for the bug in question. Testing an entire Bodhi update via updates.img is not usually really possible because it's just not quite the same thing: you can't change *everything* with an updates.img.
The process for generating an updates.img isn't very onerous, IIRC; there's a `make updates` target in the anaconda repo. So probably we could put together a SOP for generating them (to avoid dumping this requirement perpetually on the anaconda folks, though I hope they would be willing to help produce that document).
Generally speaking we use updates.img to test the individual fixes before rolling an update, where necessary/appropriate. For testing the update we really should be testing the updated packages themselves.