Configuration

It’s all automatic

Wherever possible, HypoFuzz is designed to do the right thing without configuration.

hypothesis fuzz accepts arguments to determine parameters like the number of processes to use, and the port on which to serve the dashboard - though every one of them is optional - and you can select which tests to fuzz in the same way you would select them via pytest.

Once that’s all set up though, the fuzzer is not configurable: manual prioritization of a fuzzing process is considerably more error-prone than allowing the adaptive scheduler to do its thing, and computer time is much cheaper than yours.

If the fuzzer is missing something, please get in touch so we can improve the heuristics and feeback mechanisms - and benefit everyone, automatically.

Custom coverage events

HypoFuzz runs coverage on every input to observe the branch coverage of your code - but we know that there are many things code can do that aren’t captured by the set of executed branches. This blog post gives a good overview of coverage-driven verification workflows.

We therefore treat each hypothesis.event() as a “virtual” branch - while it’s not part of the control-flow graph, we keep track of inputs which produced each observed event in the same way that we track the inputs which produce each branch.

You can therefore use the event() function in your tests to mark out categories of behaviour, boundary conditions, and so on, and then let the fuzzer exploit that to generate more diverse and better-targeted inputs. And as a bonus, you’ll get useful summary statistics when running Hypothesis!

The Hypothesis database

The Hypothesis database forms the basis of HypoFuzz workflows: failing examples can be reproduced automatically just by running the tests - because those inputs are added to the database and replayed by Hypothesis itself.

It is therefore critical that the fuzzer is using the same database as Hypothesis - regardless of how or where you run it.

Hypothesis’ default database is designed to persist failing examples on a single developer’s machine, and does so via the filesystem. If you want to share the database between your team members and your CI server though, this isn’t going to work so well - either set the database setting to an ExampleDatabase backed by a network datastore, or use a DirectoryBasedExampleDatabase pointed to a shared filesystem.