X11 usage in kdeinit/kinit.cpp

[Martin Gräßlin]

On Saturday 15 February 2014 20:49:39 David Faure wrote:
> On Thursday 13 February 2014 09:29:40 Martin Gräßlin wrote:
> > Hi,
> > 
> > on my quest to ensure that we can run all applications with -platform
> > wayland if compiled with X11 support I stumbled over
> > kinit/kdeinit/kinit.cpp which gets compiled into the kdeinit5 binary. To
> > quote the code:
> > 
> > // Can't use QGuiApplication::platformName() here, there is no app
> > instance
> > 
> > And that's kind of a global problem in that file. So I don't see how we
> > can
> > make a version for wayland without compiling it twice. Will we still need
> > kdeinit5 in the future or is that something which could be killed if we
> > improve the startup process (thus ignore the problem)?
> 
> Whether the speed improvement is still worth it for starting apps is still
> unknown, someone should profile it properly.
> However it's needed for sure for kioslaves, so that they can be started and
> passed between processes.
> 
> > A similar problem is present in wrapper.cpp which gets linked into the
> > binaries:
> > * kwrapper5
> > * kshell5
> > * kdeinit5_shutdown
> > * kdeinit5_wrapper
> > 
> > wrapper.cpp assumes there is a DISPLAY env variable and even if not it's
> > set to ":0".
> 
> Can't we detect that we're not on X11 simply from the fact that DISPLAY is
> not set?

No that doesn't work as I expect that at least during the next decade we will 
have both WAYLAND_DISPLAY and DISPLAY being set (DISPLAY for legacy 
applications which won't have a Wayland port any time soon - e.g. Qt4).

> 
> The current code is wrong on mac and windows too (but without a workspace,
> the functionality of passing slaves between apps is less used so I guess
> they just skip using kdeinit right now).
> 
> In kdeinit.cpp it seems it's mostly for startup notification (x11 only),
> and some old multihead display code that screams "remove me".
> 
> In the other files, it's about the name of the socket to talk to kdeinit. If
> e.g. wayland has no such thing as a display (i.e. the possibility to run
> multiple sessions on the same machine), then this can just be skipped.

I run into one such socket thingy and adjusted it to work on Wayland, too. 
It's one of the many review requests I created last week.

Cheers
Martin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 198 bytes
Desc: This is a digitally signed message part.
URL: <http://mail.kde.org/pipermail/kde-frameworks-devel/attachments/20140217/b4a48412/attachment-0001.sig>