The conformance test files originate from the https://github.com/google/cel-spec repository.
They are all Protobuf messages, serialized into textproto, like this:
test {
name: "self_eval_int_zero"
expr: "0"
value: { int64_value: 0 }
}The gherkinize.py script translates these into Gherkin scenarios.
Scenario: self_eval_zeroish/self_eval_int_zero
When CEL expression '0' is evaluated
Then value is celpy.celtypes.IntType(source=0)The behave tool expands the various Gherkin steps into Python functions to arrange a fixture, take an action on the fixture, and assert the actual result is the expected value.
Gherkin generation is controlled by a Makefile in the features directory which provides
two commands:
make allchecks the cel-spec repository for.textprotofiles, copies them to the localfeaturesdirectory, and converts them to.featuresfiles.make cleanremoves the.featureand.textprotofiles.
The underlying command runs the gherkinize.py script for each .textproto file:
% python ../tools/gherkinize.py basic.textproto -o basic.featureThis parses a source .textproto file and generates an equivalent .feature file.
A good way to use this is to do a checkout from https://github.com/google/cel-spec into an adjacent
directory. By default, the Makefile looks for <repo>/../../google/cel-spec but the location can
be overridden with the CEL_SPEC_PATH environment variable.
This script executes a number of git commands to pull the most recent tag for the google/cel-spec project.
Each .textproto file follows the SimpleTestFile schema.
These are deserialized into Python objects using an SDK generated from the schemas.
These Python objects that represent the Protobuf data and then loaded into classes that represented
Gherkin concepts (features with comment-delineated sections, and scenarios with given,
when and then clauses) and CEL types native to this library.
The classes representing CEL types overload __repr__() with the Python code needed to
instantiate the actual CEL types in src/celpy/celtypes.py — this code is used when rendering
the Gherkin clauses.
Finally, the classes representing features, sections, and scenarios are rendered to Gherkin tests
using the gherkinize.feature template.
Tests with unimplemented features (notably, enums) generate a warning but do not result in
scenarios. Tests which do not currently pass are listed in tags.toml. Presence in this file
results in a @wip tag being added to the scenario. In general, it's expected that scenarios
will be removed from this list over time and once passing, scenarios should never be added back to
this file.
Finally, the complete, generated .feature files are all tagged with @conformance.