pcp
[Top] [All Lists]

pcp builds ... are we on the right track?

To: PCP <pcp@xxxxxxxxxxx>
Subject: pcp builds ... are we on the right track?
From: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Tue, 02 Sep 2014 08:27:31 +1000
Delivered-to: pcp@xxxxxxxxxxx
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:31.0) Gecko/20100101 Thunderbird/31.0
I think this not a new issue, but I am visiting some QA hosts I have not used 
for some time.

I just tried to build new packages on Debian and the build aborted because 
libqt4-dev is not installed.

To fix this involves installing between 50 and 100 new packages (different 
results on different systems) and does not work at all on some debian-based 
platforms because the repos don't seem to hold consistent versions of all of 
the required packages.

Now this is not our problem per se, but I think there is a basic philosophical 
issue we need to sort out ...

I believe these are important principles:

1. Backwards compatibility is really important ... we cannot afford to move PCP 
to a place where it only works on the latest and greatest bits-n-pieces.  
Client-server considerations (split and lagging upgrades) and SOE rules mean we 
need to not only work in down-rev run time environments, but we have to able to 
rebuilt the latest PCP on older platforms.

2. If we are to improve the penetration of PCP, we should always make decisions 
that expedite server installation, even if this means some reduction in desktop 
functionality ... in PCP nirvana, the ratio of server:desktop installs is 
likely to be 100:1 or greater.

So, I think the build should be really tolerant of missing pre-requisites that 
do not compromise the core server builds ... I'd put the development package 
for libqt4 squarely in this camp.

And QA should be really tolerant of PCP functionality that is not built or not 
enabled or not installed on a particular host.  We already do a pretty good job 
here.

In the case of development package for libqt4, the Debian-like builds seem to 
be the only ones that mandate this as a build prerequisite, so I am proposing 
to remove this mandate.

I will be on the lookout for other similar candidates as I move through the QA 
minefield, but would welcome comments in the mean time.

<Prev in Thread] Current Thread [Next in Thread>
  • pcp builds ... are we on the right track?, Ken McDonell <=