The TF program code implements a combinatorial explosion. There are many environments in which the core engine operates, there are several apps that make use of it, and there is a built in query language.
One way of mastering the complexity and safeguarding the correctness of it all, is by employing unit tests.
To my embarrassment, most parts of TF are not covered by unit tests.
Here we describe the parts that are covered. But first a few words about the machinery of unit testing.
We build a
which contains only 10 slots, with node type
sign and one other node type
The code to build it is in makeTestTf.py.
Probably the corpus will be enlarged when new tests are being implemented.
We use the
framework that comes with Python.
It is just a few lines to setup a test class, hence we have not yet separated the code for setting up the tests and the data for the specific test suites. The fact that we currently have just one test suite does not give much incentive to factor the framework out right now.
basic spatial relations (
that are being used in search deserve thorough unit testing.
There are quite a few of them, they describe refined but abstract relations between nodes and the slots they are linked to, and they are optimised for performance, which make their implementation error-prone.
For every relationship we test whether it holds or holds not between a bunch of particular nodes.
The implementation of the relationships tries to determine on before hand whether its operands are guaranteed to be slots or guaranteed to be non-slots, or whether no guarantee can be given. For each particular test, all these cases will be triggered.
All in all we defined 1000 pairs of nodes leading to 2500 queries, which will all be tested against expected answers.
Expand source code Browse git
""" .. include:: ../docs/about/tests.md """