Skip to main content

Main API

Our previous is_prime example was too minimal to show how Touca can help us track regressions in the behavior and performance of real-world software workflows. Let's use a Profile Lookup software that takes the username of a student and returns basic information about them including their name, date of birth, and GPA.

class Student:
username: str
fullname: str
gpa: float

def find_student(username: str) -> Student:
# ...

Here's a Touca test we can write for our code under test:

import touca
from students import find_student

def students_test(username: str):
with touca.scoped_timer("find_student"):
student = find_student(username)
touca.assume("username", student.username)
touca.check("fullname", student.fullname)
touca.check("birth_date", student.dob)
touca.check("gpa", student.gpa)

We are using the same test framework as before but we are tracking more data about the behavior and performance of our software using various data capturing functions. Let's see how these functions work.

Describing the Behavior

For any given username, we can call our find_student function and capture the important properties of its output that we expect to remain the same in future versions of our software.

We can start small and capture the entire returned object as a Touca result:

touca.check("student", student)

But what if we decided to add a field to the return value of find_student that reported whether the profile was fetched from the cache? Since this information may change every time we run our tests, we can choose to capture different fields as separate entities.

touca.assume("username", student.username)
touca.check("fullname", student.fullname)
touca.check("birth_date", student.dob)
touca.check("gpa", student.gpa)

This approach allows Touca to report differences in a more helpful format, providing analytics for different fields. If we changed our find_student implementation to always capitalize student names, we could better visualize the differences to make sure that only the value associated with key fullname changes across our test cases.

Note that we used Touca function assume to track the username. Touca does not visualize the values captured as assertion unless they are different.

We can capture the value of any number of variables, including the ones that are not exposed by the interface of our code under test. In our example, let us imagine that our software calculates GPA of students based on their courses. If we are just relying on the output of our function, it may be difficult to trace a reported difference in GPA to its root cause. Assuming that the courses enrolled by a student are not expected to change, we can track them without redesigning our API:

def calculate_gpa(courses: List[Course]):
touca.check("courses", courses)
return sum(k.grade for k in courses) / len(courses) if courses else 0

Touca data capturing functions remain no-op in production environments. They are only activated when running in the context of a @touca.Workflow test function.

Describing the Performance

Just as we can capture values of variables to describe the behavior of different parts of our software, we can capture the runtime of different functions to describe their performance. Touca can notify us when future changes to our implementation result in significant changes in the measured runtime values.

student = find_student(username)

The two functions start_timer and stop_timer provide fine-grained control for runtime measurement. If they feel too verbose, we can opt to use scoped_timer instead:

with touca.scoped_timer("find_student"):
student = find_student(username)

It is also possible to add measurements obtained by other performance benchmarking tools.

touca.add_metric("external_source", 1500)

In addition to these data capturing functions, the test framework automatically tracks the wall-clock runtime of every test case and reports it to the Touca server.

Like other data capturing functions, we can use Touca performance logging functions in production code, to track runtime of internal functions for different test cases. The functions introduced above remain no-op in production environments.