Wow. That's kind of obnoxious of PyQt and PySide ::.

This should definitely be tested on MS Windows: subprocess can be more picky there given the more limited forking on that platform. (Though on the other hand, it's harder to not have a GUI environment on Windows, so practically we could just skip this new test if on Windows).

I'm not crazy about solving this problem with subprocess: it will increase startup time and memory by a whole second Python interpreter, plus most of matplotlib since it imports qt_compat. Particularly for people who run plot-creating batch scripts this is a problem -- I know they should switch to the Agg backend for that, but many users don't know or remember to do that.

For Linux, at least, we could just check to the existence and validity of the DISPLAY environment variable. Obviously on a Mac, X11 isn't required, but is there a similar failure case there? I just to want to solve a small problem (which I agree is a problem) with a jackhammer if a philips head screwdriver might do.

I agree, starting a process is a bit heavy.
I modified the code so as to just check that DISPLAY exists and contains a ":".
For non-linux platforms I assumed the problem existed but I did not verify.

This is much better. Thanks for your patience on this -- I have one more suggestion.

I think that the sys.platform == 'linux' is not exactly the right check. You could be using the framebuffer build of Qt on Linux, for example, in which case there would be no DISPLAY, but we would expect it to work. Likewise, we could be using the X11 build of Qt on a Mac. I think what we really want to check for is what build of Qt is installed. I couldn't find any way to obtain this information, but there are certain members that only exist in an X11 build, for example:

https://deptinfo-ensip.univ-poitiers.fr/ENS/pyside-docs/PySide/QtGui/QX11Info.html

Maybe we check for the existence of that, and then do the DISPLAY env variable check? (Or maybe there's a better way and my Google-foo isn't strong enough today).

Hi,

I tried this;

            # check for DISPLAY env variable on X11 build of Qt
            if hasattr(QtGui, "QX11Info"):
                display = os.environ.get('DISPLAY')
                if (display is None) or (not ':' in display):
                    raise RuntimeError('Invalid DISPLAY variable')

I also removed the subprocess import.
It still needs some testing on non-linux platforms
What do you think ?

J.

Looks great. @mdehoon, @pelson, @cgohlke: Any takers to test this on Windows and Mac OS-X?

Hi,

I used a regex to verify DISPLAY:

if display is None or not re.search(':\d', display):

since:
http://unixhelp.ed.ac.uk/CGI/man-cgi?ssh+1

J.

Sorry, I only have a non-X11 build of Qt4 on my Mac, so I can't test this.

@mdehoon: It actually would be more valuable if you could test this with the native build of Qt4 on your Mac. In particular, it's an open question as to whether the method used here to determine if we're using an X11 Qt4 is correct or not. So if you wouldn't mind trying it with a native Qt4 build, that would be very helpful.

It seems to work fine: With the native Qt4 build on my Mac,

from PyQt4 import QtGui
hasattr(QtGui, "QX11Info")

returns False.

Hi,

I just tried hasattr(QtGui, "QX11Info") on windows with PyQt4, and it's OK too.

J.

Great. Merging. Thanks for the patch.