[KDE/Mac] Aligning the sockets in kdeinit5 and friends on OS X

[Jeremy Whiting]

Ian,


On Sun, Feb 1, 2015 at 9:24 PM, Ian Wadham <iandw.au at gmail.com> wrote:

> Hi guys,
>
> There are a few places in the Frameworks codebase where a UNIX socket
> (as opposed to a network socket) is used to communicate with kdeinit5.
> That
> socket has to have a file path and all the usages of the socket have to
> agree
> on that path name, otherwise the communication will fail and things that
> should
> happen, after the communication, will also fail.
>
> In Linux, the name has $DISPLAY appended to it, but on Apple OS X,
> depending on which version you use (out of several supported by
> MacPorts), $DISPLAY may have a UNIX-like value, no value at all
> or a variable value assigned by the XQuartz X11 server when it is
> needed by X11-based apps.
>
> Discussion took place last year on
> https://git.reviewboard.kde.org/r/119497/
>    "Report crashes of KDE apps in Apple OS X (1) (fix kcrash)"
> It indicated that use of $DISPLAY was not necessary on OS X, and David
> said:
>
>   "The reason we have $DISPLAY in the kdeinit socket name on X11, is to
>     allow the same user to have more than one graphical session.  I don't
>     mean two full KDE-workspace instances running (they'd overwrite config
>     files), but it happened to me sometimes that I would use ssh -X from
>     another machine, then start one app (which wasn't running in the main
>     session).   What happens then is that another kdeinit4 starts, in that
>     separate session (which has a separate dbus session). So it should all
>     be separate from the main session, hence $DISPLAY.
>
>    "AFAIK this use case doesn't apply to Mac, so it would be ok to have an
>     empty string in there."
>
> One problem at that time on KDE 4 and now on Frameworks is that pieces
> of code that use kdeinit5's socket do not agree, on Apple OS X, on how to
> compute the associated file path, so communication breaks down.
>
> The search
> http://lxr.kde.org/search?v=kf5-qt5&_filestring=&_string=kdeinit5_
> returns six hits in KF5 on the string "kdeinit5_", which is a fixed part
> of the
> socket's file name.  Three hits are not to do with the socket and three
> are,
> namely in:
>
>     frameworks/kcrash/src/kcrash.cpp
>     frameworks/kinit/src/kdeinit/kinit.cpp
>     frameworks/kinit/src/wrapper.cpp
>
> @David: Please can you confirm that these are the *only* relevant hits
> and that they will catch all the cases where the socket name is generated.
>
> Attached are three tentative patches to fix things so that the same name
> is generated everywhere in Frameworks on Apple OS X.  I am proposing
> to use "NODISPLAY", rather than the value of $DISPLAY or $MAC_DISPLAY.
> One reason is to make sure wrapper.cpp and kinit.cpp use the same name.
>
> Otherwise, on KDE 4, the kdeinit4_shutdown executable built from wrapper.c
> did not shut down kdenit4 in a controlled fashion on Apple OS X.  I am
> hoping
> that my patches, or something similar, will ensure that kdeinit5_shutdown
> works correctly on Apple OS X.
>
> Please have a look at the patches and let me know what you think and how
> they
> could be improved.
>
> @Jeremy: If/when they are looking good, please can you give them a quick
> test.
> Something like start kdeinit5, crash an app and see if KCrash asks
> kdeinit5 to
> start Dr Konqi and kdeinit5 receives the message.  Then see if the command
> kdeinit5_shutdown causes kdeinit5 to receive a message on the socket.
>

Yes, I can do that. I may need a reminder though. TBH I wonder if these
issues should be put into bugs.kde.org and tracked there with the kde-mac
list as cc so we can make sure we cover/fix all the issues that there are.

Another issue I found the other day and just took a quick glance in the
sources to find what's going on:
https://bugs.kde.org/show_bug.cgi?id=343707 It seems kdeinit5 is always
looking for three specific libraries to load, but their names are
linux/unix names in an #ifdef Q_OS_UNIX block. Do we need these libraries
and their names just need to be fixed or are they not needed on OS X ? I
would think at least the first two there are needed, I'll give that a shot
(changing the names it's looking for) and put a patch on reviewboard for
that today.

>
> It is quite possible that Dr Konqi will not start and kdeinit5 will not
> shut down,
> but those are topics for another email.  One step at a time.  I just want
> to make
> sure we can get the communication happening among kdeinit5 and friends
> on Apple OS X.
>
> Cheers, Ian W.
>
> P.S.
> @David: Some last questions.  Why does kinit.cpp use an indirect method
> called
> displayEnvVarName() to look for the socket name's suffix, but both
> crash.cpp and
> wrapper.cpp use a direct method called getDisplay()?  Should not all three
> be
> using identical code?  And could that be implemented by using a macro or an
> include file, rather than cloning code?
>
>
>
>
>
>
> _______________________________________________
> kde-mac at kde.org
> List Information: https://mail.kde.org/mailman/listinfo/kde-mac
> KDE/Mac Information: http://community.kde.org/Mac
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/kde-mac/attachments/20150202/4dfd70f7/attachment.html>