$ ./python Tools/clinic/clinic_test.py
Error:
Destination does not exist: 'file'

It works in 3.4.

The clinic tests should probably be integrated into the regrtest suite, since clinic is integrated into the build process.

I will admit that even during development I rarely used Clinic's test suite.  My test suite is: make a change, then run Clinic over all of CPython and do a diff.  If nothing changed then Clinic is still working.

Any chance to fix Argument Clinic test? Without this we can't add new features, fix bugs and add tests for searching new bugs. This is critical for further converting to Argument Clinic.

I have done some diving and git bisecting into this. There are at least two issues:

1. `Clinic.__init__` no longer works without passing in a filename. Because soon after `if filename` in line 1726 [1] it calls `d('file')` which only works if there is a `file` buffer.
This makes the odd line 38 [2] in clinic_test.py to fail miserably with sys.exit. Deleting this line results in a proper test failure (22 test fail out of 54), which is already much better than the sys exit message.

2. `Clinic` has shifted from using `field_destinations` to `destination_buffers` in issue 23500. The test code did not change accordingly.


This means that we need to adapt the test code to the use `destination_buffers`, and add it to the regular test suite. The whole situation could only have happened because the test were not run regularly.

I would like to tackle this, but I am not entirely sure yet what should the differences between FakeClinic and Clinic be. So I am studying it. Hints are welcome.




[1] https://github.com/python/cpython/blob/6969eaf4682beb01bc95eeb14f5ce6c01312e297/Tools/clinic/clinic.py#L1726

[2] https://github.com/python/cpython/blob/6969eaf4682beb01bc95eeb14f5ce6c01312e297/Tools/clinic/clinic_test.py#L38

Pablo, Serhiy: there are now 2 PR (one written by Pablo, one by me). Would you mind to have a look and tell me what is your preferred PR? My PR is based on Pablo's PR, I just changed how the tests are run to make sure that they are run on *all* CIs, not just Travis CI.

New changeset 65fc98e7b1f62c2e621f04780a3a77c3498cc195 by Victor Stinner in branch 'master':
bpo-26901: Fix the Argument Clinic test suite (GH-8879)
https://github.com/python/cpython/commit/65fc98e7b1f62c2e621f04780a3a77c3498cc195

> Any chance to fix Argument Clinic test? Without this we can't add new features, fix bugs and add tests for searching new bugs. This is critical for further converting to Argument Clinic.

Done in master branch. But test_clinic.py is really basic, it should really be extended to add more tests!

I also proposed to backport test_clinic.py to 3.7: PR 9048.

Thanks Pablo for the initial fix and for your review of my PR!

New changeset 1e921236d7ecc77299d34380d6a2159e9db05a1a by Victor Stinner (Miss Islington (bot)) in branch '3.7':
bpo-26901: Fix the Argument Clinic test suite (GH-8879) (GH-9048)
https://github.com/python/cpython/commit/1e921236d7ecc77299d34380d6a2159e9db05a1a

We decided to only fix 3.7 and master branches, not 3.6 and older: https://github.com/python/cpython/pull/9048

The initial issue has been fixed, so I close the issue.

A new issue might be opened if someone wants to extend the existing test suite (test_clinic).

Thank you Pablo and Victor!

I do not mind backporting to 3.6. Argument Clinic still can take fixes, and they can add new tests.