Bug: Using F() with mysql can leak an unwrapped OperationalError

[Chris Dary]

Hey folks - I'm using MySQL 5.7, python 3.6, Django 1.10.

I just ran into this one: It looks like with a PositiveIntegerField it's possible to get django to leak a raw OperationalError from mysql rather than a wrapped one as it's intended if you try to set it below zero. I assume this is a specific case of a more general issue.

Here's a sample model:

from django.db import models

class SampleModel(models.Model):
    num = models.PositiveIntegerField()

And here's some example code showing it leaking a mysql exception which is not trappable by the traditional django.db exceptions (in shell_plus to show the queries).

In [1]: import sys
   ...: from django.db import models
   ...:

In [2]: sample = SampleModel.objects.create(num=0)
   ...:
INSERT INTO `users_samplemodel` (`num`)
VALUES (0)

Execution time: 0.000492s [Database: default]


In [3]: try:
   ...:     sample.num = models.F('num') - 1
   ...:     sample.save()
   ...: except:
   ...:     err_type, error, traceback = sys.exc_info()
   ...:     print("Caught Exception of type: %s" % err_type)
   ...:     print("Error: %s" % error)
   ...:
UPDATE `users_samplemodel`
SET `num` = (`users_samplemodel`.`num` - 1)
WHERE `users_samplemodel`.`id` = 2

Execution time: 0.000409s [Database: default]

Caught Exception of type: <class '_mysql_exceptions.OperationalError'>
Error: (1690, "BIGINT UNSIGNED value is out of range in '(`limbo`.`users_samplemodel`.`num` - 1)'")

Looks like this just needs to be trapped and wrapped somewhere?

I was able to work around it just by using a transaction and checking exists() first, but thought it'd be helpful to report.

Thanks for all your work on Django!