Also see: Truth API reference
Truth is a style of doing asserts that makes it easy to perform complex assertions that are easy to understand and give actionable error messages.
The basic way it works is wrapping a value in a type-specific object that provides type-specific assertion methods. This style provides several benefits:
Note that all examples assume usage of the rules_testing analysis_test framework, but truth itself does not require it.
def test_foo(env, target):
subject = env.expect.that_target(target)
subject.runfiles().contains_at_least(["foo.txt"])
subject.executable().equals("bar.exe")
subject = env.expect.that_action(...)
subject.contains_at_least_args(...)
Subjects are wrappers around a value that provide ways to assert on the value, access sub-values of it, or otherwise augment interacting with the wrapped value. For example, TargetSubject wraps Bazel Target objects and RunfilesSubject wraps Bazel runfiles objects. Normally accessing a target‘s runfiles and verifying the runfiles contents would require the verbose target[DefaultInfo].default_runfiles, plus additional code to convert a runfiles object’s files, symlinks, root_symlinks, and empty_filenames into a single list to verify. With subject classes, however, it can be concisely expressed as expect.that_target(target).runfiles().contains(path).
The Truth library provides subjects for types that are built into Bazel, but custom subjects can be implemented to handle custom providers or other objects.
Because Starlark‘s data model doesn’t allow customizing equality checking, some subjects allow matching values by using a predicate function. This makes it easier to, for example, ignore a platform-specific file extension.
This is implemented using the structural Matcher “interface”. This is a struct that contains the predicate function and a description of what the function does, which allows for more intelligible error messages.
A variety of matchers are in truth.bzl#matching, but custom matches can be implemented using matching.custom_matcher
Writing a new Subject involves two basic pieces:
Creating a constructor function, e.g. _foo_subject_new, that takes the actual value and an ExpectMeta object (see _expect_meta_new()).
Adding a method to expect or another Subject class to pass along state and instantiate the new subject; both may be modified if the actual object can be independenly created or obtained through another subject.
For top-level subjects, a method named that_foo() should be added to the expect class.
For child-subjects, an appropriately named method should be added to the parent subject, and the parent subject should call ExpectMeta.derive() to create a new set of meta data for the child subject.
The assert methods a subject provides are up to the subject, but try to follow the naming scheme of other subjects. The purpose of a custom subject is to make it easier to write tests that are correct and informative. It's common to have a combination of ergonomic asserts for common cases, and delegating to child-subjects for other cases.
Fundamentally, an assert method calls ExpectMeta.add_failure() to record when there is a failure. That method will wire together any surrounding context with the provided error message information. Otherwise an assertion is free to implement checks how it pleases.
The naming of functions should mostly read naturally, but doesn't need to be perfect grammatically. Be aware of ambiguous words like “contains” or “matches”. For example, contains_flag("--foo") -- does this check that “--flag” was specified at all (ignoring value), or that it was specified and has no value?
Assertion functions can make use of a variety of helper methods in processing values, comparing them, and generating error messages. Helpers of particular note are:
_check_*: These functions implement comparison, error formatting, and error reporting._compare_*: These functions implements comparison for different cases and take care of various edge cases._format_failure_*: These functions create human-friendly messages describing both the observed values and the problem with them._format_problem_*: These functions format only the problem identified._format_actual_*: These functions format only the observed values.