unittests fail with python-3.3beta1 for 0.7.8, 0.7-snap and 0.8-snap

[Former user]

Fedora 18 (hitting freeze in October) will ship with python3-3.3. We're getting ready for that by building against the python3-3.3beta1 to see what's broken. The SQLAlchemy 0.7.8 build failed its unittests.

Tested builds with the snapshots for 0.7-snap and 0.8-snap and had fewer failures but still weren't clean. I'll attach build output for each of these.

[Former user]

Build log for 0.7.8

[Former user]

Build log for the snapshot of the 0.7 branch

[Former user]

build log for the current HEAD

[Mike Bayer]

somepackage.somemodule.<locals>.SomeClass, well look at that. The kind of improvement nobody needs eh ? :) those tests should just be loosened up to not compare that exactly. then there's one weird conjunction switch, and then as usual the function call count tests, and those are the real PITA since they change on every platform.

For that last one, I think I might try googling for an approach that will report a consistent callcount across platforms since I'm super-tired of dealing with those tests.

[Mike Bayer]

OK after trying two completely opposite techniques, I have an approach for those "callcount" tests that will resolve the issue for 3.3 - basically the profiling results are maintained in a datafile, and if no values are available for compare on a particular platform, it will be skipped. I'll be merging this to 0.7 soon but if you try 0.8 right now, all the profiling tests should be skipped because "3.3" won't be in the profiling file. The tests which look for "<locals>" I still have to do something with.

[Mike Bayer]

7a81ecc82c37beae6de0e99ad07ba524b6006c8e 734d081ce6fdd970a96e43c139626519eb6125b8 1018e3b1d634f082dfed76299117b99f994c77cf 32cafe65b7fee6ad55ee22d8133d78a8228a57a1

[Mike Bayer]

seems to work for me, 0.8 and 0.7 tip, give it a try

[Former user]

Snapshots for 0.8 and 0.7 work here too. Thanks!

[Mike Bayer]

Removing milestone: 0.7.9 (automated comment)