svn commit: r364739 - in head: . sysutils sysutils/syslog-ng-devel sysutils/syslog-ng-devel/files

[John Marino]

On 8/12/2014 22:17, Cy Schubert wrote:
> In message <53EA7416.5080008 at marino.st>, John Marino writes:
>> On 8/12/2014 21:56, Bryan Drewery wrote:
>>> On 8/12/2014 2:44 PM, John Marino wrote:
>>>> On 8/12/2014 21:39, Cy Schubert wrote:
>>>>> Author: cy
>>>>> Date: Tue Aug 12 19:39:33 2014
>>>>> New Revision: 364739
>>>>> URL: http://svnweb.freebsd.org/changeset/ports/364739
>>>>> QAT: https://qat.redports.org/buildarchive/r364739/
>>>>>
>>>>> Log:
>>>>>   Reintroduce syslog-ng-devel for 3.6.0alpha2.
>>>>>   
>>>>>   Submitted by:	Peter Czanik <peter.czanik at balabit.com> (syslog-ng upli
>> ne)
>>>>
>>>>
>>>> Do Ports really need alpha quality -devel ports in the collection?
>>>>
>>>> If it were up to me I'd purge 90% of our -devel ports.  I tried to start
>>>> a conversation about a policy for these with portmgr, but as usual, only
>>>> one person responded.  I'd still like to have that conversation though.
>>>>  This -devel port trend is disturbing.
>>>>
>>>> John
>>>>
>>>
>>> Why? Devel ports need testing and there are many users willing to use
>>> them. Poudriere-devel probably has more users than the main port right
>>> now, judging from feedback I have received.
>>
>> As ports directly to improve FreeBSD infrastructure, poudriere-devel and
>> pkg-devel are included in my 10%.  (as in they are ok)
>>
>> As for why:
>> 1) They become a burden on everyone, even if they have a maintainer.
>> Sweeping changes have to be applied twice.
> 
> For the few -devel ports that there actually are I don't think this is a 
> biggie.


They aren't so few, and new -devel ports are being proposed in bugzilla
all the time for reasons that I consider weak.  Like I said, it's
trending upwards.


>> 2) What if every port had a -devel version?   Now we are taking 45k+
>> ports.
> 
> That's unrealistic.This would never happen.


It is unrealistic, but to illustrate a point.  The port count is
elevated and if this trend continues, it will go up.



>> 3) -devel versions are poor quality often
> 
> Not always. There are many GA versions of poorer quality than many -devel 
> ports.


Since the alpha or beta version has not been properly tested (by
definition), I don't see how this can be claimed or demonstrated with
certainty.  Obviously you hope they are fixing bugs but they could be
introducing bugs.  To use a beta version for a specific feature is the
definition of assuming risk.

It also says something (negative) about the release process of that s/w
if users are compelled to use the alpha/beta versions over designated
releases.



>> 4) -devel versions are often neglected and are often older than the
>> stable version
> 
> This can occur for many reasons. If they're neglected due to us, then 
> delete the port. If the port is in between releases, then it's part of the 
> release cycle.


It's always because of us.  Some of these -devel maintainers set the
port to IGNORE when it's the same as the stable, which is I guess the
best they can do (other than not have introduced the port in the first
place)



>> To me, they are more trouble than they are worth especially when the
>> ports are reset.  I think there should be a pretty high bar for devel
>> ports, and maintainer need to justify why they want to convert the
>> FreeBSD community into a testers for third party software (which is the
>> reason I've heard).
>>
>> With the exception of FreeBSD functionality, keep the testing out of
>> ports.  It will improve the quality and easy our collective maintenance
>> burder.
> 
> Let's agree to disagree.


Well, you'll never convince me alpha/beta s/w is better than a
(verified) release.  I can reasonable listen to your specific argument
about syslog-ng-devel and that other one and say, "ok, that makes sense
for this port" but I would like in general for the policy to be that
there has to be a solid benefit to *FreeBSD* for a -devel port to exist.
 Most reasons seem to benefit the 3rd party software because they get a
bigger test community.

(as a rhetorical question, do we really 3 version of syslog-ng, now 4
versions?  why can't this be reduced?  It's like have 5 versions of emacs)

John