Hello Andreas,

On Tue, Jul 30, 2013 at 1:01 PM, Andreas Tille <[email protected]> wrote:

>
> The patch needed some additions to the importer (sure, if we prevent
> something by a constraint we need to make sure the data breaking the
> constraint will not be included).  So while we now have a clean database
> I'm a bit concerned about the way how we avoid the duplicates.  Strictly
> speaking it is the fault of the metapackage designer but IMHO we should
> try to be error tolerant.  What might happen is that we get some sequence
> inside the tasks file say
>
> Ignore: foo
>
> Depends: foo
>
> Suggests: bar
>
> Depends: bar
>

> In the database we would get
>
> udd=# SELECT package, dependency FROM blends_dependencies WHERE package in
> ('foo', 'bar') ;
>  package | dependency
> ---------+------------
>  foo     | i
>  bar     | s
>
> even if both should get a dependency 'd'.  Currently I issue error lines
> into
> the logfile.  Just check
>
>    grep -iw error blends_metadata_gatherer.log
>
> However, it seems to be clear that this logfile is rarely visited and so
> I wonder what to do.  Should we check for the strongest possible
> dependency?  Should we just accept what comes first (as we do now).
>

I got your point, you are right, when that happens it is a problem. To be
honest I am not quite sure what is more *correct* in this case. Maybe
storing the strongest possible dependency seems better but if a package
appears in Depends and Suggests(or other cases) then there is no hint what
is *better* cause basically is an error in the task files.


> Currently those cases are only in debian-edu and I will talk to the
> debian-edu people at DebConf anyway, but I wanted to mention this
> problem here.
>
Yeap that will be helpful, also thanks for mentioning because I did not
think of that yesterday when we had the talk about the duplicates.


> Perhaps we include an example of the problem below into Debian Fun task
> to make sure we will not forget.
>
> I will add them.


> Once I was using the database including constraint the resulting taskdesc
> files were OK.
>
> :-)

 Also I just made a commit where I added Makefile and rules files in the
project. For the moment I commented out the packages.txt/avoidpackages.txt
from Makefile. I tried running "make dist" inside a blend and it seems to
work(orig.tar.gz file is generated containing the needed control/taskdesc
files). I generate a control file and a taskdesc.template. In the rules,
for testing, I generate a taskdesc.<arch> from the template using the
host's arch. This also seems to work. For local testing I have added custom
local path into the Makefile/rules. Except from  Makefile/rules file what
else we need at this moment?


Kind regards

Emmanouil

Reply via email to