… posix compatibility
SGE job arrays allow one job id to be associated with a set of processes on an SGE cluster. modified SGEEngineSet to be a subclass PBSEngineSet ipcluster will now generate a default SGE job script if --sge-script is not provided. most folks should ignore the --sge-script option unless they know they need it. if --sge-script is passed, check that the script exists and that the user has defined a "#$ -t" setting within the script. if not, add the setting for them by copying the script to a temp file and launching the job array using the modified temp file. ipengines terminate cleanly now when the ipcluster command exits. i think we still need to handle furl files when engines are assigned to different hosts via SGE. without using ssh or nfs, the only other way is to put the contents of the furl in the job script before submission but this is less secure. need to discuss this.
It turns out that chaining too much info into the same deferred produces rather weird side-effects. So I broke up the test into several independent ones with separate deferreds. Note that the original bug fix was OK, the fix is just for the Twisted tests themselves. As Min said, with Twisted code sometimes getting the tests right is harder than the code itself.
Reported by Chris Filo Gorgolewski on the list: http://mail.scipy.org/pipermail/ipython-user/2010-July/006993.html
Thread with details here: http://mail.scipy.org/pipermail/ipython-user/2010-July/006993.html While it seems the original problem wasn't really a bug, I'd already written an extra custom test, so might as well keep it.
Now we import foolscap components according to their api. I think these four files contain all the naked 'from foolscap import' statements we had, so all such warnings should be gone now.
See https://bugs.launchpad.net/ubuntu/+source/wxbanker/+bug/394394 for a similar report
This fix was backported from Min's commit to trunk with minimal changes.
Patch by David Warde-Farley https://bugs.launchpad.net/ipython/+bug/409566