Re: [cross-project-issues-dev] Is there a reason to have org.apache.comm

[Thomas Hallgren <thomas@xxxxxxx>]

I think this discussion highlights another problem. All Helios 
contributors are dependent on the Eclipse platform. The versions chosen 
by the platform are therefore rather important hints to what versions 
the contributors should use.

The general rule of thumb for a contributor should be to resolve 
dependencies against the Eclipse platform first and if no viable version 
can be found there, only then consult what's in Orbit. In other words, a 
Helios contributor should be satisified with what the platform provides 
even if a later version is found in Orbit.

For this particular bundle we are lucky since it marked as 
singleton='false' which means that there is no install conflict. It will 
be very interesting to see what happens when more then one version of 
the org.apache.commons.logging bundle is installed. I wouldn't bet on it 
working flawlessly, but it might.

- thomas



On 05/20/2010 04:16 PM, Stéphane Bouchet wrote:
> FYI,
>
> Looking in the httpclient page :
> "Commons HttpClient 3.x codeline is nearing the end of life. All users 
> of Commons HttpClient 3.x are strongly encouraged to upgrade to 
> HttpClient 4.0."
>
> looking at the dependencies, the last 4.0.1 comes with commons-logging 
> 1.1.1 :)
>
> Scott Lewis a écrit :
> > FWIW,
> >
> > Yes...as Pascal and John A indicated, ECF is using the apache 
> > httpclient (3.1) and this dictates a dependency on commons logging 
> > 1.0.4.
> > We (ECF, etc) would be completely willing to move some other mutually 
> > agreed upon version of commons logging, but there are some important 
> > subtleties:  Since it's httpclient that depends upon 1.0.4, once a 
> > replacement version of commons logging was selected, someone would 
> > have to look for/test for regression in httpclient 3.1 in order to do 
> > so...since there's unfortunately no guarantee that httpclient 3.1 
> > doesn't have a hard dependency on 1.0.4 (I'm not saying there is or 
> > anything...just that we don't know, and discovering something like 
> > that could be dramatically bad if the change was done 
> > cavalierly...especially this late in the game).
> >
> > All this by way of saying...I think aligning on a version would be 
> > fine (I mean...other than the version upon which we're aligned 
> > now)...but it's unfortunately not trivial for us to safely change 
> > this dependency...particularly without some available resource to 
> > check/test for possible regression.
> >
> > Scott