Re: [GSoC] blends-gen-control hints (Was: blends-dev, gsoc 2013)

[Andreas Tille <andreas@an3as.eu>]

Hi Emmanouil,

On Wed, Aug 21, 2013 at 09:22:07PM +0300, Emmanouil Kiagias wrote:
> > just to sum up our short live conversation:  I'd suggest to rather use a
> > '-q' option to silence the control file instead of needing to force the
> > output using '-w' - we really want the warning for developers notice.
> >
> This is done. The warning message is printed by default and added a -q
> argument which make the script silent.

I have noticed the commit.

> > So I think it is a very good idea to check whether our new method
> > compeares reasonably to the currently existing metapackages for
> > debian-edu, debian-med and debian-science to be really sure that our
> > method is really sensible.
> >
> In order to automate the comparisons for control/taskdesc files  between
> current and new blends-dev I wrote a script  "blendsdev_compare"(simple
> diff command was not working because packages where not in the same
> order):. I generated the orig.tar.gz for debian-med with the current(_cur)
> and the new blends-dev into  folders like:
> 
> debian-med-1.13.1_cur (current blends-dev)
>    and
> debian-med-1.13.1_new (new blends-dev)
> 
> And below I dump the differences between the control files from current and
> new blends-dev using the script:
> 
> ./blendsdev_compare -d
> ../debian-med-1.13.1_cur/debian/control,../debian-med-1.13.1_new/debian/control
> -c
> 
> For surprise there are a lot of differences in "Suggests" headers. There
> are a lot of missing "Suggests" packages from UDD so missing from the
> control file. I have attached a control.diff with the output of
> the blendsdev_compare script. The control.diff contains packages which are
> included in the current blends-dev and are not included in the new.
> 
> Example of the output comparing for debian-med:
> 
> * Comparing med-practice
>   --> diff in suggests
> [   u'mirrormed',
>     u'elexis',
>     u'openpms',
>     u'proteus',
>     u'ginkgocadx',

$ apt-cache show ginkgocadx

>     u'freeicd',
>     u'clearhealth',
>     u'freeshim',
>     u'medintux',
>     u'resmedicinae',
>     u'tinyheb',
>     u'sqlclinic',
>     u'remitt',
>     u'openrep',
>     u'freemed',
>     u'elementalclinic',
>     u'mirth',
>     u'freeb']

With the exception of ginkgocadx which is definitely in Debian this is
correct.  That the other packages are actually missing makes me aware
that we really need to finalise my plan to completely inject all data
inside the tasks files into UDD.  When parsing the tasks files and
finding some packages that are in no other table (not only in
blends_prospectivepackages) we either need to create another table or
append the data to blends_prospectivepackages and using a flag whether
these re coming from machine_readable gatherer or are just data from the
tasks files.

But for the metapackage creation for the moment this is OK (except
ginkgocadx).

> blends_dependencies_alternatives so they are also not included into the
> control file. These packages  are not found into UDD so they not included
> at all in blends_dependencies_alternatives table from the
> blends_metadata_gatherner.

Yes.
 
> >  Nice to have you met at DebConf
> >
> Same for me :-)

:-)


Kind regards

   Andreas.

-- 
http://fam-tille.de