catala/tests/README.md
2021-05-26 21:57:51 +02:00

44 lines
2.0 KiB
Markdown

# Catala test suite
This folder contains Catala source files designed to test the features of the
language.
It uses `make pass_tests` to launch tests and compare the test terminal output
with an expected output.
Expected outputs are stored using the convention
`<name_of_test>.catala_<language>.<name_of_scope>.out` in the corresponding test output folder.
For both workflows: use `CATALA_OPTS="..." make ...` to pass in Catala compiler
options when debugging.
## Workflow for adding new tests
1. Create a new test file in `foo/{good,bad}/bar.catala_<language>` (pick the right directory and
an informative name for your test)
2. Write your test, and pick a toplevel scope `A` to run.
3. From this directory, launch `make foo/{good,bad}/bar.catala_<language>.A.run` to get the output of
your test.
4. When you're happy with the output, launch `make foo/{good,bad}/bar.catala_<language>.A.in`. This
will record the content of the output of your test into a file.
5. Check that your test pass with `make foo/{good,bad}/output/bar.catala_<language>.A.out`.
6. That's it, you've added a new test for the Catala language!
## Workflow for fixing regressions
1. Run `make`, if a test fails you should see something like
`[FAIL foo/{good,bad}/output/bar.catala_<language>.A]`.
2. Compare the computed and expected output with `make foo/{good,bad}/output/bar.catala_<language>.A.out`.
3. Debug the compiler and/or the test, running `make foo/{good,bad}/bar.catala_<language>.A.run`
periodically to check the output of Catala on the test case.
4. When you're finished debugging, record the new test output with
`make foo/{good,bad}/bar.catala_<language>.A.in`.
5. Re-run `make` to check that everything passes.
6. That's it, you've fixed the Catala test suite to adapt for changes in the
language.
If a compiler change causes a lot of regressions (error message formatting changes
for instance), you can mass-reset the expected the outputs with `make reset_tests`.
**Caution**: use at your own risk, regressions should be fixed one by one in
general.