Hi Jonas, On 12/03/2015 07:02 AM, Jonas H wrote: > Hi there, > > there seems to be something wrong with either the SQLite backend, the > SQLite library itself or Django's transaction handling. Whenever I try > to run the test suite of a specific project of mine > (https://github.com/jonashaag/fahrtkostenrechner), using > `fahrtkostenrechner/manage.py test events`, it shows the following behavior: > > * On Linux, it doesn't crash and shows actual test failures. > * On OSX, with SQLite's in-memory database as test database, a segfault > (!) occurs. (I'm not familiar with the lower-level Python SQLite > bindings, but I guess a segfault should *never* happen in the first place?!)
Certainly a segfault should not happen :-) The fact that one is even possible does point to a bug in Python's sqlite module (AFAIK any case where one can trigger a segfault from pure Python code is considered a bug in CPython). It may be that we can avoid triggering this particular segfault bug, if we can figure out how we are triggering it. I'd be extremely surprised to find _two_ segfault bugs in the SQLite bindings (that both occur only on OS X) so it seems almost certain to me that https://code.djangoproject.com/ticket/24080 is the same underlying issue one way or another. > * On OSX, with a on-disk test database, a > "django.db.utils.ProgrammingError: Cannot operate on a closed database." > (caused by `_cursor()`: > https://github.com/django/django/blob/4d0f8831a7498ab3b1ebcf4cafa2ee234503e4f8/django/db/backends/base/base.py#L206) > occurs. That's a useful clue! If it consistently tracks with the segfault, where the only difference is on-disk vs memory db, it seems likely that it's the same situation that's causing a segfault in the memory-db case. > It looks to me that the cause of this is some unfinished transaction > rollback, but I'm not sure, so I'm asking for guidance here. In the file > linked above in L193, is it legitimate for `needs_rollback` to be True? > In my debugging I found `connection != None` and `needs_rollback=True` > whenever `_cursor()` fails in L206, so I figured maybe there might be an > outstanding rollback for some reason? `connection != None` and `needs_rollback = True` is a legitimate state for the connection to be in when someone asks for a cursor, AFAIK. It would be most useful to trace back and find out a) what caused `needs_rollback` to be set to `True`, and b) when/how/why was the database closed? > I use Python 3.5, SQLite 3.9.2 and I can reproduce the crash on Django >>=1.8.4,<1.9. This might be useful information to add to https://code.djangoproject.com/ticket/24080, if it's not already present there. Does the version range you give imply that the issue goes away in Django 1.9, or just that you've never tried it there? Carl -- You received this message because you are subscribed to the Google Groups "Django developers (Contributions to Django itself)" group. To unsubscribe from this group and stop receiving emails from it, send an email to django-developers+unsubscr...@googlegroups.com. To post to this group, send email to django-developers@googlegroups.com. Visit this group at http://groups.google.com/group/django-developers. To view this discussion on the web visit https://groups.google.com/d/msgid/django-developers/56606820.3010900%40oddbird.net. For more options, visit https://groups.google.com/d/optout.
signature.asc
Description: OpenPGP digital signature
