Back in the day, Django didn’t have a built-in way to change model schemas. You either had to figure out and apply the changes yourself or use a third party tool like South.

After a successful Kickstarter campaign, Andrew Godwin, Django core member and the author of South, added a native migration tool to Django. It solved problems such as conflicting merge names and the ability to squash migrations once there are too many of them in an app.

One thing it also changed is introduce what I have dubbed ‘cosmetic migrations’: these are migrations that make no changes to the database schema, but only add internal Django changes, such as changing a field’s choices or the ordering of a model. I’m sure the change is for good reason, but it annoys me to no end, because the lack of schema impact means I’m unlikely to notice that I haven’t made them until at some point I do make a schema-altering change, and that migration is then flooded with an untold number of cosmetic changes. This is a problem because commits should be atomic.

Django will occasionally notify you that ‘[y]our models have changes that are not yet reflected in a migration’, but I found that I would only see those when it was too late.

The solution

Thankfully, it is possible to make these checks yourself, although I have never seen it advertised anywhere. Executing this command will give your what you need:

$ python manage.py makemigrations --check

This command seems wholly counter-intuitive to me, but it does what I want: Exit with a code 1 if there are unreflected changes.

You can plug the above in to your Continuous Integration system or possibly a pre-commit hook. If you do so, I recommend that you also make it --dry-run, like so:

$ python manage.py makemigrations --check --dry-run

This will ensure that no migrations are actually created, which just seems the saner option if you want to check.

From the department of ‘optimisations that are so premature that if you ever find yourself actually caring about it, you need to find better problems to solve’, I recently had this thought: What’s the cheapest/fastest iterable in Python?

Reminder: iterable and iterator are not the same thing.

The need for an empty iterable occasionally comes up, like when you need to provide a default value to a missing key in a dictionary, and you need it to be something you can iterate over without running the risk of a TypeError. The beautiful thing, of course, is that you can iterate over an empty iterable and just have nothing happen, so the actual type or contents don’t matter.

So I set out to test it. Again: you should never need to actually care about this. If you can live with the actual overhead of iterating over something, you can live with the overhead if that something is empty, no matter the actual type of iterable.

I evaluated strings, lists, tuples, dictionaries and sets. My hypothesis was that the fastest would be a string or maybe a tuple.

The test was performed on a late-2016 13″ MacBook Pro with a 3.3 GHz Intel Core i7 and I timed it using the timeit module.

First I tested out simply declaring the different types of iterables:

kweli:~ j$ python -m timeit -c '""'
100000000 loops, best of 3: 0.00672 usec per loop
kweli:~ j$ python -m timeit -c '[]'
100000000 loops, best of 3: 0.0187 usec per loop
kweli:~ j$ python -m timeit -c '()'
100000000 loops, best of 3: 0.0119 usec per loop
kweli:~ j$ python -m timeit -c '{}'
10000000 loops, best of 3: 0.0305 usec per loop
kweli:~ j$ python -m timeit -c 'set()'
10000000 loops, best of 3: 0.0924 usec per loop

So far so good: strings are the fastest, followed by tuples and dicts, with sets trailing far behind.

Then to actually iterating over them:

kweli:~ j$ python -m timeit -c 'for i in "": pass'
10000000 loops, best of 3: 0.0433 usec per loop
kweli:~ j$ python -m timeit -c 'for i in []: pass'
10000000 loops, best of 3: 0.0514 usec per loop
kweli:~ j$ python -m timeit -c 'for i in (): pass'
10000000 loops, best of 3: 0.0438 usec per loop
kweli:~ j$ python -m timeit -c 'for i in {}: pass'
10000000 loops, best of 3: 0.0707 usec per loop
kweli:~ j$ python -m timeit -c 'for i in set(): pass'
10000000 loops, best of 3: 0.136 usec per loop

And again, the hypothesis is confirmed, but interestingly the difference between lists and strings/tuples is much smaller when iterating compared to just declaring.

So in conclusion, use a string as an empty iterable, unless you have any reason at all not to. The difference is infinitesimal.

Over the last month or so, I’ve begun learning Clojure. I don’t do much blogging, let alone technical, but I realise I’ve actually had this blog for so long that my first impressions of Python (the language I spend most of my day job writing in), are documented.

I read through Kyle Kingsbury’s Clojure from the ground up series, and found it an easy learning process. Although I consider myself somewhat of a polyglot, I realised that I hadn’t actually learned any new programming languages in almost ten years, aside from various JavaScript type annotation supersets. (I’ve tried learning Haskell, but have been largely unsuccessful.)

All in all, I like it. I like its functional paradigm, making functions pure by default, but not having to ask permission to get side-effects (which is the feeling I get a bit with Haskell). Leiningen is a great tool to get up and running, and it takes care of a lot of the minutae, like installing dependencies and running tests. I really like being able to name functions almost anything, including non-ASCII characters and characters normally reserved (so now I can use possessive in a function name).

One thing I find about that I really dislike is the abbreviations. Now, this might simply be because it’s the first language I’ve picked up in a while, and I simply haven’t paid attention to it in other languages, but the incessant abbreviating every conceivable name drives me up the wall. Why in the world does it have to be conj and assoc, what’s wrong with conjoin and associate‽ The fact that abbreviations are applied so randomly proved a stumbling block for me, which I feel it really shouldn’t have to be. This is my first foray into Lisp, so I don’t know how much (if any) is simply convention, but space-saving concerns one might have had in the 50s can surely be ignored today. I get more riled up than I justifiably should be, but it irks me. (And again, I realise this might simply just be my internalisation of some abbreviations: I have no problem with str, concat and def.)

I find the destructuring syntax in a lot of cases to be greatly confusing and emanating magic. I have come to terms with let taking a vector of alternating key value pairs, but the sprinkling of keywords to imbue bindings with special properties means I’m still at a copy & paste–stage for some use cases. I’m not very far into macros yet (I have yet to write my first one), but from what I can sense, it leads to a lot of poorly designed APIs. But it might just take some getting used to. (I thought the self argument for Python methods was stupid at first, and now I don’t think about it.)

I also really miss Python’s named, any-order parameters. I realise something similar can be achieved in Clojure using keys destructuring, but that can’t be combined with arity overloading, which I also really like. (Yes, this might be a case of wanting to have a cake and eating it too.)

The lack of a good date and time library is also unfortunate (at least for the apps I tend to do). I’ve been using clj-time, which seems to be a pretty thin wrapper around Joda Time, and while it does its job, it has some odd shortcomings, the primary being its incapability of representing date-less times. I’ve resorted to vectors of hours, minutes, etc., but when you’re used to Python’s datetime library, specifically datetime.time in this instance, you find yourself wanting.

I have found one library that I really like, though: Enlive. It’s an unconventional templating library, in that it doesn’t make a DSL for templating (or, indeed, give access to the whole language, as in PHP), letting the templates instead be pure HTML, and doing the transformations in Clojure. It took me a little while to get the hang of doing things such as loops, but I think it makes for a clean separation of concerns, and I’ll definitely investigate the concept in Python. (There is a Python port, although it doesn’t seem to get much attention these days.)

All in all, I’m really excited about Clojure. For web development it lacks some of the maturity and cohesiveness that I’m used to with Django, but as a language it has a lot of interesting concepts and libraries.