"Ant" <Ant_member@pathlink.com> wrote in message news:blg9ll$ktr$1@digitaldaemon.com...
> In article <blg6mv$gf0$1@digitaldaemon.com>, Matthew Wilson says...
> >
> >Ant
> >
> >I'm afraid I know nothing about GConf, so can't comment.
>
> A tree struct that holds key/value pairs,
> (if it's not that ignore the rest of this)
> and more (a value can be a list) but the common interface would
> leave out the "uncommon" features (wouldn't you know...)

Sounds great. I'd be very happy if the D.win32.registry module could serve as an exemplar for such a thing, as well as being used in its Win32 implementation

> There is a GConf client that looks a lot like the regedit.exe
> (for win9x I didn't check latelly)

Where does it store its stuff? Is it system global, or can each process have its own reg file?

> >As regards the registry library, as soon as Walter's finished some
essential
> >compiler-waltering on the C++ compiler we'll be putting the registry API into Phobos, and releasing the next version, so it'll be best if you can hang fire a little while, and then it'll all be available in its majesty, ready for digestion and potential regurgitation. :)
> >
> >If the reg and GConf are in any way similar, I'd certainly like to investigate a potential merge. I'd probaby prefer that there was a
separate
> >common module, perhaps DConf, which could be implemented in terms of the
reg
> >module on Win32, and the appropriate technology on Linux. Does that sound sensible?
>
> That's my idea.
> I'm coding and experimental version that I named DConf :)
> (copy/paste of course no fancy ,def files yet ;)

Great minds, and all that !

On Thu, 02 Oct 2003 02:31:52 +0000, Sir Ant scribed thus:
> I'm gonna look at GConf now.
> "GConf is a system for storing application preferences."
> (Took me what? 1 hour just to find that it exist)

I would not recommend making GConf and the registry the same API - they look the same but really are quite different.

For instance:

* GConf entries must (?) have schemas that give key documentation,
possible values etc
* GConf key types are different
* GConf notifications are used a lot more than win32 registry
notifications are
* GConf is not guaranteed to be available on any given Linux system in the
same way the registry is.
* GConf is not meant to be used to store arbitrary data like the registry
is, it's supposed to be for app preferences only (hence the powerful
application defaults/lockdown settings)

> Do you think we could integrate GConf and your win32.reg in a common interface? Is that desirable?

I would generally recommend that D bindings have the same name as what they bind to (so import gtk rather than import dui) and are direct mappings rather than abstractions.

If somebody wants to use the registry and GConf in the same way
they can always write their own simple abstraction class on top of them,
IMHO, and use the versioning stuff in D to let it figure itself out.

In article <pan.2003.10.02.10.23.14.738016@theoretic.com>, Mike Hearn says...
>
>On Thu, 02 Oct 2003 02:31:52 +0000, Sir Ant scribed thus:
>> I'm gonna look at GConf now.
>> "GConf is a system for storing application preferences."
>> (Took me what? 1 hour just to find that it exist)
>
>I would not recommend making GConf and the registry the same API - they look the same but really are quite different.
>
>For instance:
>
>* GConf entries must (?) have schemas that give key documentation,
>possible values etc

??

>* GConf key types are different

That's why we need an abstraction layer

>* GConf notifications are used a lot more than win32 registry notifications are

Don't see it as a problem

>* GConf is not guaranteed to be available on any given Linux system in the same way the registry is.

That's a real problems.

>* GConf is not meant to be used to store arbitrary data like the registry is, it's supposed to be for app preferences only (hence the powerful application defaults/lockdown settings)
> 
>> Do you think we could integrate GConf and your win32.reg in a common interface? Is that desirable?
>
>I would generally recommend that D bindings have the same name as what they bind to (so import gtk rather than import dui) and are direct mappings rather than abstractions.

I'm even thinking of remove all the gtk tokens from DUI.
(probably a mistake(?) I'll take another look at it)

>
>If somebody wants to use the registry and GConf in the same way
>they can always write their own simple abstraction class on top of them,
>IMHO, and use the versioning stuff in D to let it figure itself out.

My first idea was to have that.
except for the "their own" part.
Integrating it into phobos might be too ambitious...

Thanks for your insights!

Ant