[horizon] Horizon's openstack_auth is completely broken with Django 2.2, breaking all the dashboad stack in Debian Sid/Bullseye
I've started reviewing the patches and they mostly look good and easy to
merge. Just gotta get the core devs looking through them and merging them.
I think this is urgent since if we fail to add 2.2 support this cycle,
then Horizon will be out of LTS support for Django before the next
release cycle ends.
On 27/08/19 8:46 AM, Stephen Finucane wrote:
> On Sat, 2019-08-24 at 23:55 +0200, Thomas Goirand wrote:
>> Django 2.2 was uploaded to Debian Sid/Bullseye a few days after Buster
>> was released, removing at the same time any support for Python 2. This
>> broke lots of packages, but I managed to upload more than 40 times to
>> fix the situation. Though Horizon is still completely broken.
>> There's a bunch of issues related to Django 2.2 for Horizon. I managed
>> to patch Horizon for a few of them, but there's one bigger issue which I
>> can't solve myself, because I'm not skilled enough in Django, and don't
>> know enough about Horizon's internal. Let me explain.
>> In Django 1.11, the login() function in openstack_auth/views.py was
>> deprecated in the favor of a LoginView class. In Django 2.2, the login()
>> function is now not supported anymore. This means that, if
>> openstack_auth/views.py doesn't get rewritten completely, Horizon will
>> continue to be completely broken in Debian Sid/Bullseye.
>> Unfortunately, the Horizon team is understaffed, and the PTL told me
>> that they don't plan anything before Train+1.
>> As a consequence, Horizon is completely broken in Debian Sid/Bullseye,
>> and will stay as it is if nobody steps up for the work.
>> After a month and a half with this situation not being solved by anyone,
>> I'm hereby calling for help.
>> I could attempt to work on this, though I need help and pointers at some
>> example implementation of this. It'd be nicer if someone more skilled
>> than me in Django was working on this anyways.
> Fortunately, I do know Django, so I've gone and tackled this series
> All the unit tests are passing now, so I assume that's all that's
> needed. I didn't test anything manually.
> Probably a longer conversation to be had here regarding the long term
> viability of Horizon if issues like this are going to become more
> common (probably similar to what we're having in docs, oslo, etc.) but
> that's one for someone else to lead.