[Debtags-devel] Re: Recent progress

[Hervé Eychenne]

On Wed, Feb 23, 2005 at 10:30:41AM -0800, Erich Schubert wrote:

> Hi Hervé,

> I believe that once we reach some "policy" status for debtags, the
> responsiveness to bug reports due to incomplete tagging etc. will
> increase. ;-)

Probably increase, but almost certainly never ensure some kind of
completeness. :-/

> But we need something nice to show for then.

> > - even if the tagging of a package is complete at a given moment,
> >   there is currently no way to know it _is_ complete. And even if we =
knew
> >   it was, there is currently no way of knowing which tags were introd=
uced
> >   after that, that might make the tagging (given the whole new set)
> >   incomplete again.

> Actually, there is a "carefully tagged" service tag, which is to be
> used for that.

Ok, I was not aware of its existence... 
I think the "completely tagged" or not question is deeply related to
the existence of a timestamp.
If you admit that, there are several kinds of issues:
- When should such a tag be added or removed?
  When a complete tag review occurs, the package is marked as complete.
  But when new tags are introduced, the review should be marked obsolete
  until a new review (with just the new tags) takes place.
  As there are chances that new tags may be added regularly (the list of
  tags will never be complete, by essence), we probably cannot afford
  marking all "carefully reviewed packages" as uncompletely reviewed each
  time a single new tag is introduced.
- How to ensure the visibility of this tag?
  If the package is not completely tagged (yet), no "carefully tagged"
  mention will appear in the list of current tags, and there are
  chances people don't even know that such a tag exists (like I did).
  I see 2 solutions:
  - add a special case in the user interface, asking is the package is
    "carefully tagged" or not. This ensures visibility.
  - reverse the tag: a "uncompletely tagged" is more striking in the user
    interface. But this doesn't give the timestamp of the review. That
    is not an issue, as I think the "carefully tagged" timestamp should
    be an attribute of the package.

Therefore, I think that we should store the "carefully tagged" information
as a package attribute, not a tag. If the package is not "carefully
tagged" yet, the timestamp is 0. If it's carefully tagged", the value
is the date of the last review (an integer, i.e. number of seconds since
epoch). So we would get rid of the "carefully tagged" tag, which
doesn't fit all the needs.

> And the time when this tag was added can be reconstructed using my back=
ups.

Good. :-)

> > - So, as a maintainer responsible for the tagging of a package
> >   wanting to ensure that the tagging of my package is up-to-date, I r=
eally
> >   wouldn't want to wander through the list of tags from time to time,
> >   trying to find some new dispersed tags that may apply to my package.
> >   It's in no way efficient (long and repetitive process), and you mig=
ht
> >   still miss some new interesting tags.

> As soon as the maintainers will do this, our vocabulary will need to
> be mostly frozen;

I think it's utopian. It will always evolve. Not hugely, but it will,
and that's normal. Let's deal with that instead of requiring a
restrictive freeze.

> then we can either go through the packages list ourselves

Completely utopian:
- the number of packages is huge, and will keep increasing
- the people who should decide if a tag must be added or not to a
  package must have a good knowledge of the package itself.
  It's absolutely not the case of even a small team. I insist: only
  the maintainers or people who know the package well can do that.
  Not "ourselves" in any way.

> when we add a new tag, or publish it with changelogs,
> similar to the way policy changes are handeled now.

Garanteed uncompleteness. ;-)

> We can then maybe write a lintian tool which will spit out a message
> when new tags have been introduced in the meantime and the maintainer
> needs to review them.

Yes, it's a must.

> > - the current interface (http://debian.vitavonni.de/packagebrowser/in=
dex.cgi)
> >   is (IMHO) hardly usable for any of the two important tasks it
> >   should help with:

> Well, editing is currently better done using debtags-edit.

I think a X11 GUI will always be superior (in terms of functionnality)
to a Web interface. Yet, I think that a Web interface is much more
accessible... I mean that if we require that every people who want to
edit tags for a package has to install a X11 application (like
debtags-edit), there will (unfortunately) be much less updates.
Believe my experience. :-(

> >     tagging of a package I know the name of. For the moment, I have t=
o
> >     first click in any subgroup so he shows me a list of packages,
> >     click on the name of a package, and then change the name of the
> >     package for the one I want in the URL. Oh my...

> Yeah, that is the same way I do it. Havn't gotten around to placing a
> search form somewhere.

That would be... nice. ;-)

> >     Ok... is the tagging considered complete for now? I don't know. :=
-(
> >     So I look to the "tags that can be attached to this package". But

> Well, when is it complete? I don't know either. Maybe when you go over
> the list and don't want to add any more tags

Absolutely. When someone autoritative (experienced with the package
functionnalities) does that and ensure it's "complete".

> >     what do some tags exactly mean? Mystery... For example, I see

> Yes, vocabulary descriptions are not complete either.

Even if they were, it would be good to provide an access to them. :-)

> >     convinced by things like "Service tags" (what does it mean?),

> Well, they are for service purposes. ;-)

Of course, but they appear in the interface, and that is confusing.
:-(

> >     "Role of the package in the system" (my sound converter doesn't
> >     play any particular role in the system, right?), "Purpose of the

> Sure, it is an application for you, not a server for example.

That is the tag "application". I was just saying that reading "Role
of the package in the system" in the list of tags is confusing.

> >     Well, I'm sorry to have to say that again (we've had this
> >     conversation before, Erich), but I think this way of presenting
> >     things is really unusable in practice for average users... :-(

> Yes, but mostly because tagging is just *SOOO* different. Packages are
> lacking lots and lots of tags, so the subgroups don't make sense
> currently.

I have the intuition that even if it was complete, that wouldn't be so
nice. Of course, I cannot really prove it until it is. And as it won't
be tomorrow, I guess you win. :-/

 Hervé

-- 
 _
(°=  Hervé Eychenne
//)  Homepage:          http://www.eychenne.org/
v_/_ WallFire project:  http://www.wallfire.org/