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

[Andreas Tille <andreas@an3as.eu>]

Hi Emmanouil,

On Wed, Aug 28, 2013 at 08:07:22PM +0300, Emmanouil Kiagias wrote:
> Hello Andreas,
> 
> I have a question about the automatic changelog entry. As we said in some
> previous discussions the changelog entry is injected between some *flags*
> in the changelog like: * start automatically injected changes * [entry
> here] * end automatically injected changes *. For the moment the Makefile
> fails(printing a proper message) in case it does not find any of the 2
> flags(start/end). Also it fails in case it does not find a tasks.<json>
> containing the dependencies from the previous release(so it can not perform
> the comparison). Should I let it as is or make it optional? I mean if any
> of the above cases occurs, the Makefile should fail or just print a warning
> message and generate the orig.tar.gz anyway?

I think it is reasonable to fail with a warning.  I'm not fully sure
whether we finally really need the * start/end  automatically injected
changes * markers - we could agree upon making this list always the last
changelog entry and if it starts with some defined keyword it should be
possible to detect some previous entry which needs to be replaced or if
we should just add the entry.  For the moment I'll leave the decision
about using the markers or not to you in how far you consider this a
safe way of coding.

Regarding the missing JSON files which means we can not really create
the changes record at all I'd vote for a failure.  We really want this
feature, right.  Giving a reasonable hint what is missing and in case of
questions direct to the Blends list seems to be reasonable.  Please
remember that for a new Blend no such entry is possible and that the
check should pass if there is only one release of the Blends
metapackages.

> Also there is still one thing left(I forgot about that while I was at
> debconf). We should ask debian-edu people if they need the packages.txt /
> avoidpackages.txt to be stored in orig.tar.gz(should I send an email into
> this list?)

It might make sense to bring this topic up in a separate mail and also
CC debian-edu@lists.debian.org.

> Using debian-med/debian-science as examples I compared previous and current
> blends-dev and they have no differences(except the packages which do not
> exist in blend_dependencies_alternatives UDD table).

Sounds good.  This reminds me on my todo list ... :-)

Kind regards

       Andreas.

-- 
http://fam-tille.de