As I first mentioned in my post[1] on comp.lang.python, it seems possible to modify a dict while a lookup is ongoing, without causing the lookup to restart.  This requires the ma_table be reallocated twice (first to a new address, then a second time back to the original address).  This could also happen if the constants or behaviour relating to ma_smalltable were changed, but the current implementation is safe.

To fix this, I suggest two more tests be added along side the ep0 and startkey tests that currently restart the lookup process.

1) ma_mask.  If ma_table was reallocated, and the new allocation is smaller, memory corruption would result.  Checking ma_mask for changes would ensure this doesn't happen.

2) reallocation counter.  Increment every time ma_table is reallocated.  It would be technically possible to overflow this, but the circumstances would be exceedingly unusual.


[1] http://groups.google.com/group/comp.lang.python/browse_thread/thread/7f5b8fc59aadbdea/d77bdbcf5b20c056

Is this something that should be taken up on python-dev?

I don't believe there's anything to debate on this, so all it really needs is a patch, followed by getting someone to review and commit it.