Sqlite segmentation fault regression when running tests (since 1.7.2)

[Steve Jalim]

A project has a reasonably large number of tests (~1600). As part of our dev workflow, we run tests against sqlite first, then against postgres before checkin.

On 1.7.1 all tests pass on sqlite, bar some deliberate DB-specific exclusions for tests that need specific postgres features. They also all pass on postgres.

On 1.7.2, all tests still pass on postgres (though the server needed a restart) but the sqlite test run fails at varying points with "segmentation fault" (and an exit code of 139, accordingly).

The segmentation fault occurs regardless of whether an in-memory or file-based database is used. When the file-based run crashes, the DB is around 495k in size.

There is very little output, even at max verbosity, because this is coming from something lower down than Django itself. However, because it only started happening with 1.7.2 and goes away when we revert to 1.7.1, I am assuming a change somewhere caused the regression.

Setup spec:

• Headless Virtualbox + Vagrant with 1GB allocated RAM
• Ubuntu 14.04 LTS, fully up to date
• libsqlite3-0 version: 3.8.2-1ubuntu2 amd64
• Python: 2.7.5 (up-to-date version for 14.04 - unclear which parts, if any of 2.7.8 have been backported)

Any steers here would be much appreciated. apollo13 on #django pointed out that I could git bisect the commits between 1.7.1 and 1.7.2, which I am open to doing if I can make the time.

Cheers
Steve