[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [p2-dev] Projector weight function

Hi Daniel,

thanks for reply.
Just for clarification. I do not assume the current behavior as a bug.
But I failed to find information regarding the objective function definition used.
So I raised the question here really out of curiosity.
Unfortunately the provided documentation link does not work for me. (404)

The reported Tycho problem should be fixed in Tycho anyway.
0.0.0 versions used in the feature.xml should be strictly resolved to the latest available version. 

Best regards, 
Eckart

On Behalf Of Daniel Le Berre
Sent: Mittwoch, 20. Juli 2011 15:44

Hi Eckart,

Thanks for the bug report. We will take a closer look at your particular case.

The way the objective function is defined in p2 is described here:
http://www.cril.univ-artois.fr/spip/publications/lash2010.pdf

The behavior you observe might be expectable.

Suppose that you have package x version 1 and x version 2

Version 1 has no dependencies and version 2 has dependencies on y z u and v.

Suppose we have a minimal penalty of 1 for installing a package (to avoid installing a non necessary package).

In that case, if the penalty for installing an "old" package for x1 is 4 for instance, then the solver prefers to install x1 for a penalty of 4 instead of x2 for a penalty of 5 (1 for each x2, y,,z, u, v).

Note that with a bigger penalty for old versions, we could still install x2.

I will check in your bug report if this is really what happens in your case.

Cheers,

Daniel
Le 20 juil. 2011 à 13:22, Langhuth, Eckart a écrit :

> Hi,
>  
> As requested I created a p2 bug for this issue
> https://bugs.eclipse.org/bugs/show_bug.cgi?id=352560
>  
> Best regards,
> Eckart
>  
> From: p2-dev-bounces@xxxxxxxxxxx [mailto:p2-dev-bounces@xxxxxxxxxxx] On Behalf Of Pascal Rapicault
> Sent: Montag, 18. Juli 2011 18:52
> To: P2 developer discussions
> Subject: Re: [p2-dev] Projector weight function
>  
> For a new install (which is the case in a tycho build), the intent is for the resolver to return the highest version available.
> For addition on top of an install or update, the algorithm is a bit more complex but should favour the highest version of the things being in the transitive closure of what is being changed.
>  
> Could you please open a bug (https://bugs.eclipse.org/bugs/enter_bug.cgi?product=Equinox) with the necessary IUs to recreate the problem.
>  
> Thx
>  
> On 2011-07-14, at 4:46 PM, Langhuth, Eckart wrote:
> 
> 
> Hi,
>  
> while investigating a problem resulting in bug
> https://bugs.eclipse.org/bugs/show_bug.cgi?id=352081
> I stumbled over the behavior of org.eclipse.equinox.internal.p2.director.Projector.
>  
> Out of curiosity. Is there a description of the weight function used to determine a solution?
> In my example a feature includes a single plugin without a version constraint. (0.0.0)
> The search scope contains two version of the required plugin.
> The solution provided by the projector contains the plugin with the smaller version.
> My assumption would have been to prefer a solution with latest plugin version.
> The difference I see is that the higher version plugin defines more dependencies.
> Is it intended to prefer a small installation over a new one?
>  
> Best regards,
> Eckart
> _______________________________________________
> p2-dev mailing list
> p2-dev@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/p2-dev
>  
> _______________________________________________
> p2-dev mailing list
> p2-dev@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/p2-dev

_______________________________________________
p2-dev mailing list
p2-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/p2-dev