Python task manager from scratch, part 46: Getting handlers under test

Now that handlers are explicitly passed in an environment, we can create a test environment. It's surprisingly easy to go wrong with test environments. Here are some failure modes I've seen:

Not including enough or diverse enough test data.
Continuing to mock out functions when modifying the test environment would be cleaner, more reliable, and more readable.
Hiding test data behind layers of "convenience" functions for making test data. The idea motivating such functions is that test data often have tons of fields, most of which aren't being tested at any given time.

This last is the most interesting to me. First, I think it's underappreciated that an object with too many fields or properties is often ill-designed. Sometimes the domain really is that complicated, but often this is a sign that one code object is representing many logical or real-world objects. That's a problem. Often, homonymy seems to be behind this: the "delivery" as understood in an order-processing system is (arguably) not the same thing as, or at least not best represented as identical to, the "delivery" that is a box you might be monitoring with a smart camera on your doorstop.

Second, lots of helper functions seem to be motivated by a desire to reduce lines of code. Avoiding repetition is nice, but test code is supposed to be a bit repetitive, and readers don't mind scrolling a bit. More fundamentally, sometimes a bespoke set of test data really needs to be created, with a certain amount of variation in various fields. Stuffing all of that variation in as values of a dozen parameters of a create_foo_test_data function does not make things more readable.

Finally, I've occasionally seen towers of helper functions, like:

make_test_foo()
make_test_foo_with_variation_in_bar()
make_test_foo_with_only_some_variation_in_bar()
make_test_foo_with_new_value_of_enum()

The underlying issue seems to be that whatever helper function was useful at a given moment was checked in and made immortal. I find that a few text fixtures, along with ad hoc test data creation, serves me well. I don't mind reading test code that requires a bit of copying and pasting. (Here is one discussion of the subject.)

Whether or not you like how I've done it, I've created a test environment, finally written some tests for those helpers, and fixed some bugs along the way.

Here's the current commit in the veery/ repository.