[Python-modules-team] Bug#747270: python-amqp unusable

[Thomas Goirand]

Hi,

Brian, thanks a lot for all the details that you're giving on the
message below.

Since python-kombu doesn't depend on anymore on python-librabbitmq, and
that it migrated to Jessie, we don't have any reverse dependencies for
python-librabbitmq anymore. So probably, it'd be a good idea to just ask
for the package removal now.

Brian, what do you think? Shall I proceed to open the bug against
ftp.debian.org so that the package python-librabbitmq is completely
removed from Sid and Jessie?

Michael, you're listed as uploader for python-librabbitmq, do you agree?

Thomas Goirand (zigo)

On 07/23/2014 07:50 AM, Brian May wrote:
> Hello,
> 
> Thank you for your email.
> 
> I assume you are reacting to the "is marked for autoremoval from
> testing" you would have received from Debian.
> 
> Unfortunately, as much as these are scary looking messages, I believe I
> had already resolved the issue. I sent a email to debian-python mailing
> list, however maybe you did not see it.
> 
> First an apology: like I already said
> in https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=747270#15, I used
> the wrong package names, and the confusion appears to have been carried
> across in your email.
> 
> python-librabbitmq != python-amqp
> 
> python-librabbitmq is the older library that uses C bindings, and only
> supports Python2. It has the bug as mentioned in this bug #747270. So
> far there doesn't appear to be any real interest in resolving the
> bug. https://github.com/celery/librabbitmq
> 
> This is not surprising, as there is a newer library that is entirely
> native Python code, that supports Python2 and Python3. In Debian
> unstable, this library is called python-amqp (not to be confused with
> python-amqplib in wheezy, which I believe is an earlier version of
> python-amqp). https://github.com/celery/py-amqp
> 
> Both python-librabbitmq and python-amqp are maintained upstream by the
> Celery project.
> 
> My understanding is that python-amqp can be used anywhere
> where python-librabbitmq was used, and it is API is compatible. This
> explains when python-kombu 3.0.19-1 has the following in
> its depends:
> 
> python-librabbitmq | python-amqp (>= 1.4.5)
> 
> Unfortunately, this means the older broken library gets installed
> automatically as the preferred version.
> 
> So I raised the severity of the bug against python-librabbitmq to Grave
> (***not*** python-amqp), either to raise interest in fixing the problem,
> or have the older broken library removed.
> 
> Unfortunately, the one issue I forgot is that python-kombu 3.0.19-1
> build depends against both python-librabbitmq and python-amqp. So it was
> marked as broken, and anything that build depends on python-kombu was
> also marked as broken, which resulted in a lot of scary looking emails
> being sent.
> 
> Note that no action would have been taken until August the 20th, there
> was no need to panic.
> 
> Yesterday, I rectified the situation by uploading python-kombu version
> 3.0.19-2 (as part of the Debian Python Maintainer team). This version no
> longer depends on python-librabbitmq. When this version hits testing,
> all problems should disappear. At the time I also sent a email to the
> debian-python mailing list explaining my solution to the "is marked for
> autoremoval from testing" automatic emails.
> 
> 
> On 23 July 2014 01:31, Thomas Goirand <zigo at debian.org
> <mailto:zigo at debian.org>> wrote:
> 
>     though you didn't give any justification on why python-amqp is
> 
>     "unusable". As much as I understand, there's a single use case (ie: with
>     celery) where there's a serious problem. Knowing that OpenStack makes
>     extensive use of python-amqp, and that Ubuntu is also using the same
>     version, I seriously doubt that it is unusable as you wrote. Also,
>     python-amqp doesn't depend on python-librabbitmq. However, python-kombu
>     does. So why are you even talking about python-amqp?
> 
> 
> 
> I didn't intend to say python-amqp was unusable. I intended to say
> python-librabbitmq was unusable, as per this bug report.
> 
> I posted a followup message when I realized I got this wrong.
>  https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=747270#15
> 
> Apologies again for the mistake.
> 
> python-amqp is the solution to this breakage, and I have no problems
> with it.
> 
> So when you say "Knowing that OpenStack makes extensive use of
> python-amqp, and that Ubuntu is also using the same version, I seriously
> doubt that it is unusable as you wrote.", you haven't given any
> information to suggest that python-librabbitmq is not broken. You
> said python-amqp is not broken, and I already know that.
> 
> 
>     All of the above are to me, signs that "severity grave" isn't correct.
> 
>     I'm therefore downgrading it to "important". Please don't switch it back
>     to grave unless you justify it accordingly.
> 
> 
> I feel you could have waited one day for a response from me...
> 
> The key question, as I see it: Is there any need to keep maintaining the
> old python-librabbitmq when we have a perfectly good replacement that is
> far better?
> 
> If python-librabbitmq still serves a useful purpose that cannot be
> replaced by python-amqp, and it still works for this purpose, then I
> agree, Grave was inappropriate for this bug report. We will need to come
> up with some other solution to the problems with celery.
> 
> However, as you yourself said "Knowing that OpenStack makes extensive
> use of python-amqp", there should be no problem if python-librabbitmq
> disappears, because python-librabbitmq is not python-amqp.
> 
> I discussed these issues on the debian-python team mailing lists, it
> appears maybe I should have included openstack lists too?
> 
> Thanks
> -- 
> Brian May <brian at microcomaustralia.com.au
> <mailto:brian at microcomaustralia.com.au>>