Implement prompt consistency tests and harmonize all activate scripts #1330

bskinn · 2019-03-01T02:18:29Z

See history of bskinn/virtualenv:all-prompts branch, tip at
bskinn/virtualenv@67cda5f, for the actual sequence of
development of the prompt tests, leading to the first commit of the PR (e6a2b99)

This PR will NOT be ready to merge until the activate scripts for csh, xonsh, cmd and posh have been fixed. PR created now per discussion at #1316, downstream of #1168 and #1285.

At time PR was opened, closes #1294 and closes #1295. By the time PR is complete, will close #1296, close #1297, close #1298, and close #1299. Also resolves #1168, the initial discussion that prompted all this.

Planning also to fix the minor documentation error to close #1304 before merging.

Thanks for contributing a pull request, see checklist all is good!

wrote descriptive pull request text
added/updated test(s)
updated/extended the documentation
added news fragment in docs/changelog folder

See history of bskinn/virtualenv:all-prompts branch, tip at 67cda5f, for the actual sequence of development

setup.cfg

tests/activation/test_prompts.py

(1) Fix pathlib import for 2/3 compat (2) Refactor preamble_cmds to defaultdict (3) Convert shell info to namedtuples Also switch filename template string fields to autonumber. Also revert addition of black to setup.cfg

All cmd tests now pass (locally, CI pending). Closes pypa#1297.

All current PoSH tests now pass locally. Closes pypa#1298.

Path doesn't have .write_text() or .read_bytes() in 3.4. 2.7 complains if the environment variables are typed as unicode.

(1) Rearrange test_prompts.py to satisfy linters When function and variable definitions are interleaved at the module level, black and flake8 cannot be satisfied simultaneously, due to conflicting visions about how many blank lines should be present at certain spots. (2) Expand DISABLE_PROMPT tests for granularity VIRTUAL_ENV_DISABLE_PROMPT, in bash, disables the prompt if the env var is defined and of nonzero length, REGARDLESS of the truthiness of the nonzero-length value. These new tests bracket this specific behavior definition precisely.

tests/activation/test_prompts.py

virtualenv_embedded/activate.ps1

Switch to virtualenv.IS_WIN for platform checking. Refactor platform pytest.skip checks to per-shell "select case" sort of layout. Also refactor so that the actual skip occurs in the test function, where the fixture function just returns a skip message if skip is called for. Refactor the PoSH executability check to its own fixture. This speeds up the test runner by only carrying out the execution check once. Refactor the shell_info definitions to eliminate the .__new__.__defaults__ hack, since namedtuple(.., defaults=[...]) is only available on 3.7+. No more single-character function argument in env_compat.

It appears that for a completely non-colorized terminal fish's set_color command returns either a null, or an empty string, or something. This leads to incorrect string interpolation on the Azure test VMs. See: https://github.com/pypa/virtualenv/blob/94c6bcb4a3f8d7ff63e2215b18d76514d147a762/virtualenv_embedded/activate.fish#L92 The test in question fails, but the failing value of the prompt is "env()". This implies that the interpolation is not "seeing anything" from the output of the "(set_color normal)" invocation, and instead putting the environment name ahead of, rather than into, the parentheses. Hopefully this added env variable in the preamble of the script will coerce fish into always including colorizing ANSI codes even on the Azure VMs. --- Also, tweak clean_env fixture to ensure VIRTUAL_ENV_DISABLE_PROMPT is not present, as the tests all assume it is *unset* in the clean_env. Also also, buff a number of comments.

BEAR IN MIND, this only works for xonsh 0.8.9 and newer, and only as long as xonsh doesn't change its implementation. Probably the xonsh tests should be skipped on CI, since changes outside the virtualenv codebase could cause breakage. It would still be informative to have the xonsh tests in the codebase, though -- perhaps a pytest CLI argument could be added to enable them?

Since xonsh prompt configuration functionality is mostly handled in the xonsh codebase, don't want changes in xonsh to suddenly start causing virtualenv CI builds to fail.

MacOS (t)csh apparently prepends "prompt\t" to the content printed to stdout, precluding the stricter equality test of the prompt contents.

Closes pypa#1304.

Closes pypa#1284.

conftest.py

tests/activation/test_prompts.py

virtualenv_embedded/activate.ps1

Per discussion in pypa#1330

(1) WIP: Convert partially to info class model Still midstream, so lots still broken. (2) WIP: More conversion to shell-info class model (3) Finish shell-info class conversion & cleanup. Needs testing on Windows before merge to main PR branch. (4) Fix posh provisioning test glitches

Force push to re-trigger Azure pipeline x2

gaborbernat · 2019-04-10T14:13:52Z

Thanks, @bskinn, however once merged pypy3 tests started failing - https://dev.azure.com/pypa/virtualenv/_build/results?buildId=6588 can you investigate?

bskinn · 2019-04-10T14:51:32Z

@gaborbernat I checked it...the failing test isn't related to the new prompt test machinery. Was something changed/fixed relating to pypy3 recently? I realize now, I forgot to merge/rebase master, so the merged PR may have obliterated a recent change?

bskinn · 2019-04-10T15:10:31Z

Hm, no -- none of the recent commits on the history seem at all related to this... if anything, it looks like a behavior change in pypy3... though it's really odd timing that it would've been changed between my committing 677bbc1 and your merging the PR.

It looks like the test was expecting pypy3 to error on an attempt to import distutils, due to some pypy3 oddity, but it did not error.

        def _check_no_warnings(module):
            subprocess.check_call((exe, "-Werror", "-c", "import {}".format(module)))
    
        # pypy3's `distutils.sysconfig_pypy` imports `imp`
        # https://bitbucket.org/pypy/pypy/pull-requests/634/remove-unused-and-deprecated-import-of-imp/diff
        if virtualenv.IS_PYPY and sys.version_info > (3,):
            with pytest.raises(subprocess.CalledProcessError):
>               _check_no_warnings("distutils")
E               Failed: DID NOT RAISE <class 'subprocess.CalledProcessError'>

I don't think this is something that I'm equipped to fix -- I don't know anything about the internals of pypy3, and I have no idea if/how this distutils behavior affects virtualenv.

bskinn · 2019-04-11T01:04:44Z

I looked at it some more and... Huh. That pypy PR mentioned in the comment in the test function got merged.

Maybe this fixed the error in pypy3, and there's no reason any more for the pypy3 special-case pytest.raises(...) check?

gaborbernat · 2019-04-11T12:06:52Z

I believe so too 😃

SQUASH: Implement prompt consistency tests

e6a2b99

See history of bskinn/virtualenv:all-prompts branch, tip at 67cda5f, for the actual sequence of development

gaborbernat reviewed Mar 1, 2019

View reviewed changes

setup.cfg Outdated Show resolved Hide resolved

tests/activation/test_prompts.py Outdated Show resolved Hide resolved

tests/activation/test_prompts.py Outdated Show resolved Hide resolved

bskinn added 4 commits March 1, 2019 10:41

SQUASH (3): Various fixes and responses to review

e0f7394

(1) Fix pathlib import for 2/3 compat (2) Refactor preamble_cmds to defaultdict (3) Convert shell info to namedtuples Also switch filename template string fields to autonumber. Also revert addition of black to setup.cfg

Harmonize activate.bat.

7626e55

All cmd tests now pass (locally, CI pending). Closes pypa#1297.

Harmonize activate.ps1

bd2c924

All current PoSH tests now pass locally. Closes pypa#1298.

Revise for 3.4, 2.7 compat

01106ea

Path doesn't have .write_text() or .read_bytes() in 3.4. 2.7 complains if the environment variables are typed as unicode.

bskinn force-pushed the pr-all-prompts branch from 2e584ba to 01106ea Compare March 1, 2019 18:14