Plumbing Life's Depths

Just ran into a South migration issue.  Basically we had a field that looked like this (simplified):

class MyModel( models.Model ):
    hostname = models.CharField(
        default = random_unique_hostname,
    )

where random_unique_hostname() was written like so (again, simplified):

def random_unique_hostname():
    test = generate_random()
    try:
        MyModel.objects.get( hostname = test )
    except MyModel.DoesNotExist, err:
        return test
    else:
        return random_unique_hostname()

the actual code is significantly more complex, but that illustrates the problem.  When you try to generate a South migration to add some other field to this model (i.e. a few months after you've added this default), you'll see the migration creation fail because South is calling your default (ouch) with a database that doesn't have the new field (the pre-migration DB).

The solution is to use a minimal query, rather than calling .get, use .filter().count() or use .filter().values_list('id') or any other query that doesn't map to the whole object being loaded, just to check for unique-ness.

We use globalsub and MockProc extensively in our test-suite at work.  Today I wanted to look up the name of the argument in MockProc (we use globalsub in almost every test case, so I don't need to look that up).  So, look up the docs... wow, whoever wrote this library really sucks, there's just some stupid blog post rather than real documentation.  What a loser!  Oh, yeah...

So, suitably shamed, I sat down and wrote a bit of documentation this evening for both packages.  Cookies!

I've just done a bit of hacking to make the (Py)OpenGL.raw hierarchy a little more "normal" looking when you read the auto-generated code.  There are now functions declared, the functions are then annotated with decorators to make them into wrappers (which actually replaces the original functions).  Doesn't really affect how the code works, it just uses Python structures to define "functions" rather than raw data-structures.  It also reduces the distribution's size by about 5%, but I doubt the 44KB matters to anyone.

[Update] and then I did even more, so the unimportant package size is now down around 12%, but more importantly, the core library is being auto-generated from headers (again), which is what I really wanted to get to, as I want that change for future work.

Hey, google/android peoples. Here's a proposal: shame every licensee into releasing official unlockers for every Android phone ever released.  The unlocker should just install an open bootloader that can install whatever ROM the user wishes (it can scream loudly at the user that they are voiding their warranty, of course).

To be clear; not for upcoming, or recently released devices only.  For every phone ever.  Then we might begin to feel that a licensee cares an iota about keeping Android up to date.  I know they don't, but heck, at least that way users don't have to jump through a dozen hoops just to get a Cyanogen or similar ROM on their phone when the manufacturer abandons them.

Sure, making an iTunes-like financial incentive to keep users up to date would be nice too. But that takes a lot of engineering effort versus signing some image that can be uploaded to the phones.