shrinkplz 0.1.1

A command line tool to help you shrink failing test data

Shrinkplz

shrinkplz is a command line tool to help shrink failing test data, similar to git bisect.

shrinkplz can take some input data causing a test failure, and then attempt shrinking it by cutting out lines from the input data until it starts passing.

At the end of the shrinkplz session, you should have a (locally) minimal input data for your test that is still failing.

Installation

This command is available on PyPI

uv tool add shrinkplz

Use Cases

• Shrinking a failing SQL script, removing irrelevant statements
• Shrinking a failing CI test run, by removing elements from the test list
• Shrinking large input data to a program until it starts working again

Quick Start

Start a shrinking session by calling:

shrinkplz start some-test-data

shrinkplz expects a unicode file, and will attempt to shrink it by bisecting the lines themselves.

Once you've started a session, a file called current-input is created. This is the currently shrunk test data.

Once you have started your session, run your test with this test data or otherwise check your conditions:

./run_my_test.py current-input
# ... and then investigating the result

Once you have finished testing the data in current-input, you should mark it with the mark command

# if the test now passes
shrinkplz mark pass 

# if the test is still failing
shrinkplz mark fail

# if the test data is invalid (e.g. the data has become malformed)
shrinkplz mark invalid

Once you mark the result, current-input gets replaced with another candidate for testing.

Each mark gives shrinkplz more information about how it can shrink data.

• shrinkplz mark fail indicates that current-input is enough to cause the failure, so discards any test data from the original input that isn't in current-input

• shrinkplz mark pass indicates that current-input is missing the data that causes a failure, so we will restore the chunk of data that we removed

• shrinplz mark invalid similarly indicates that the chunk of data shouldn't be removed, though without inferring more

After shrinking as much as it can, shrinkplz will write out something like:

Completed our search!
The smallest known data is copied
to current-input

From there, the session is over and current-input represents the result of the search

Script

Just like with git bisect, shrinkplz can be scripted. This is especially valuable for larger pieces of test data, where a shrinking session might require hundreds or thousands of test runs.

shrinkplz script my-test-script my-test-input 

shrinkplz script takes two inputs:

• a test script, that will be run by shrinkplz to determine how to mark input data.

  The test script's error code determines what the marking is. A return code of 0 means to mark the data as pass, a return code of 125 indicates marking the data as invalid, and any other return code indicates a fail

• the test input to shrink

The test script is called without arguments, but just like when running a shrinking session manually, current-input will be available as the current input

#!/usr/bin/env bash
# an example shrink script
uv run my-prog.py < current-input

At the end of the shrinkplz script session, current-input will be the shrunk test data