Installing display profiles/interfacing with gcm

[Florian Höch]

Am 16.06.2010 12:41, schrieb Richard Hughes:
> On 15 June 2010 16:55, Florian Höch<lists+gnome-color-manager at hoech.org>  wrote:
>> I'm currently looking for a way to install display profiles in a similar way
>> as Argyll's dispwin does. So, I'm either looking for an utility
>> 'gcm-install-display-profile'<displayno>  <filename.icc>  that would copy the
>> file, do the mapping displayno->device and write the neccessary info to
>> device-profiles.conf (does gcm-install-systemwide do this?).
>
> No, gcm-install-systemwide just installs the profiles system-wide.
> It's also pretty fragile the way it works, and probably needs
> rethinking in the near future.

Ok.

>> Or for another way to achieve what I just described (DBus?).
>
> I think a DBus call would be okay, but it seems very specific to this
> use case. GCM also stores other data with the profile entry which
> would need to be formatted like the others by dispcalGUI.
>
>> Or is it safe/recommended to write to device-profiles.conf directly?
>
> I guess. If GCM is open at the time then it might cause problems, but
> I can always add an inotify watch on the config file and reload GCM's
> in-memory copy if it changes. I'm not sure this is a great idea, as
> the config file has no locking, and no ABI stability.

Hmm, sounds like I'll probably want to avoid it, then ;) Maybe as a 
"last resort".

>> Then, I'd just have to figure out how to do the Argyll displayno ->  GCM
>> device mapping (btw, what's the $(ascii) in
>> xrandr_$(vendor)_$(name)_$(ascii)_$(serial) mentioned in the GCM wiki?).
>
> It's the ASCII text returned in the EDID, e.g. "LTN154P2-L05".

Ah, ok. So then all the naming information is available through the 
EDID? That's fine with me.

> I'm
> fully open to changing the format of the device-profiles.conf file if
> required. That could mean adding a hash=${md5_of_edid} parameter to
> the display section, or it could be just making the identifier be
> xrandr_${md5_of_edid} -- but then we have all the problems of writing
> the junk data required to display a meaningful device icon when the
> display is not attached. If the user has already run gcm-prefs and the
> device was saved, and you just want to change the default, then
> searching on hash= is probably the right thing to do.

Agreed on your points. I think the format of the config file is fine, 
although addition of the EDID hash key sounds like a nice idea.

> Of course, the best option is probably for GCM to expose libgcm
> (better names welcome) which would allow you to just call
> gcm_device_save() trivially in C. It would mean a compile-time
> optional dep on GCM, but it probably the most correct thing to do.

That would be nice, then I could let GCM do "the right thing" for me. 
Dep on GCM is not a problem I think, as I'm probably going to use 
Python's ctypes then, which can tell me if the library is installed and 
I can also use it to wrap the C function call.

Thanks for the infos so far!

> Comments please. :-)
>
> Richard.
-- 
Florian Höch
http://hoech.net