Hide Forgot
Description of problem: We have hit sqlalchemy bug http://www.sqlalchemy.org/trac/ticket/1493 several times in production. It is fixed in newer versions of sqlalchemy 0.5.x. Please, update the package to the newer version. Version-Release number of selected component (if applicable): python-sqlalchemy0.5-0.5.5-1.el5 How reproducible: random (due to the nature of the bug) Actual results: application crash Expected results: application working
I took a look at performing this upgrade over the past week but 0.5.6 and newer are failing their unittests. In 0.5.6 there's only one failing unittest -- I haven't confirmed that it's only a single unittest failure for 0.5.8 but the build for 0.5.8 failed in the same place. Failing build: https://koji.fedoraproject.org/koji/taskinfo?taskID=3098567 Build with just that test patched out to show that we'd otherwise be fine: https://koji.fedoraproject.org/koji/taskinfo?taskID=3098569 @Lev: are you abbot? If so, I'll approve your pkgdb request if you'd like to work on upgrading. We'd want to see if we can get this unittest to not use empty inserts to perform its test. If it can then we can patch and update. If we can't then we could consider adding the decorator to exclude running this test if the sqlite is old enough to not support empty inserts.
Hi, yes I'm abbot, sorry, forgot to mention that I've left the access request. I saw this test failing, and currently I just built a package locally which patches this failing test away. I'm not sure if it does worth it to re-implement this failing test with non-empty inserts, I would consider simply removing it from the known to fail platforms (which is only EL5 to my best knowledge). Same with the a decorator to exclude the test: this is the only failing test, it is easier to remove it, and consider a decorator/more complex solution if this situation repeats with another sqlalchem version (but this is unlikely since 0.5.8 is the last release in the 0.5.x series and there should be no more releases in this branch).
AFAIK, the decorator is already implemented which is why I bring it up. The other way would be to use the -e commandline option to nosetests to exclude the test from running. Fixing the test would be better in that we would know that we aren't breaking a feature in our quest to fix something else :-) but yeah, I bring up the methods to exclude since, if we can tell that the test is only failing b/c of the version of sqlite but we could use the feature fine with other db's; we could exclude the test rather than fix the test. Anyhow, you're now approved so go ahead and update the package to 0.5.8 :-)
Toshio pushed python-sqlalchemy0.5-0.5.8-9.el5 to the epel5 stable repo to fix security Bug #783305. https://admin.fedoraproject.org/updates/EL-5/FEDORA-EPEL-2012-0715