LinuxRegistry in Freedesktop & KDE

[George]

On Sat, Apr 17, 2004 at 11:37:48AM -0300, Avi Alkalay wrote:
> > Because it requires people to recode for absolutely no real-world gain
> > in features or ease or use.  In fact, it's a disservice to users,
> > because it would cause all existing knowledge and tools to become
> > invalidated.  With no other gain in usability in the new config
> > system, that's bloody pointless.
> 
> Current tools are inconsistent. Current knowledge is poor. I want my
> mother to use Linux as she uses Windows. You'll find good knowledge in
> experienced sysadmins. (A free desktop isn't only for geeks). Please
> think in the big picture.

Will your mother care about the on-disk implementation of the configuration?
My mother won't.  Actually my mother will probably bitch at me if some
configuration got lost because LR doesn't have any way to avoid races
and has no notification (though she won't know that she'll just know that
stuff breaks).

> Someday we'll have to set some standards in this area. It is inevitable
> for interoperability. We can speak here because we all agreed in the
> "english" standard, even if it isn't the best for everybody.

Interoperability?  Why does gdm have to 'interoperate' with apache on it's
configuration?  I learned english to speak to you, but I don't need to learn
which brand of toiletpaper you use because that is your implementation detail
and not mine.  Plus it's very unlikely I'll be at your house being forced to
use your toiletpaper which may be incompatible with mine.  Standardizing on
one toiletpaper brand will not make much difference except waste a lot of
effort.  And some people may actually use different and more elaborate
devices for the same purpose that we use toilet paper for and would argue
that toilet paper doesn't have all the features they require after they have
taken a dump.

Ideally all non-obscure configuration should be easy to do with a GUI easily,
and thus the actual configuration is an implementation detail.

A configuration GUI needs to be geared towards the application anyway, so you
will have GUI code for each application.

> TCP/IP is not big deal also. But before him, computers had to set up a
> raw communication. TCP/IP was a very simple, stupid, and straight forward
> solution. So simple that it works for everything.

But gdm doesn't need to set up any comunication with apache's configuration.
If it ever will need to do that I think it will be a sign that something went
wrong in gdm design.  Same for vast majority of other apps.

> Take care here. It isn't so inefficient. It was not refined yet, and
> effieiency is not the only think you must look. Configuration gathering
> is a small part of what programs do at startup. And LR nature lets you
> get them on demand. So it isn't big deal. And I still want to see GConf
> benchmark numbers. So please open your mind and look at the benefits
> also.

But LR is also unusable for desktop (fam is not the solution for notification
as such a solution will be incredibly race prone).  Also what is the benefit
of LR?  Why should we accept a possible efficency hit and break existing
configurations, APIs and spend a lot of effort changing existing code to use
it?  In essence, why should we care?

> Sometimes I think how much disk space is wasted with the compiled code
> into each program that handles configuration parsing..... Smaba, Apache,
> XFree, init, mount, modprobe, etc, etc. Why not remove it from there and
> put all in a shared lib?

ARE YOU ARGUING FOR ADDING DEPENDENCIES?  I may have overheard you.  You know
putting common code in shared libraries and then using them from other
places.  You know ... depending on them.

> Thanks for the comments. Comments like these made me change a lot of LR's
> design in the begining. But this is past, so please open your mind also
> for the benefits first.

So what are the benefits?  I have not heard anything except ineroperability
which I think is not a very huge benefit to begin with.  There is no race
avoidance, not much for lockdown features, description (comment) of the key
is not localizable, no notification of change (fam is NOT the solution here),
not easily used for shared (network) storage (perhaps using nfs, but that's
not that useful), inefficencies on most filesystems for applications having
lots of keys, further to be really useful would have to have buyin from most
projects, a completely different format from almost anything currently in
use, very hard to edit by hand (must edit lots of small files)

I see only problems but no benefits.  For example I see no benefit something
like gdm would get out of using LR, except that it would break existing
installations, make people have to learn yet another configuration format,
make it much harder to edit with a text editor and generally make many users
unhappy.

George

PS: Note the sarcasm sprinkled throughout.  Sometimes it seems that people
take seriously every word.

-- 
George <jirka at 5z.com>
   Never interrupt your enemy when he is making a mistake.
                       -- Napoleon Bonaparte