This is surprising to me:

python_tests(
    name="tests",
    resolve="reqs",
    dependencies=[
        ":fetch_test_data_adhoc",  <-- A python script to fetch
        ":test_dataset",           <-- The tarball that gets fetched
        ...
    ]
)


adhoc_tool(
    name="fetch_test_data_adhoc",
    runnable=":fetch_test_data_py",
    output_files=[
        "results_dataset.tar",
    ],
)

resource(
    name="test_dataset",
    source="test_dataset.tar",
)

python_source(
    name="fetch_test_data_py",
    source="fetch_test_data.py",
)

This is incorrect because

:test_dataset

, if included as a dependency to

tests

, shows up as an empty file:

Engine traceback:
  in `test` goal

InvalidFieldException: The 'source' field in target test/python/mytest:test_dataset must have 1 file, but it had 0 files.

However if I remove the

:test_dataset

from the test dependencies, Pants is able to add the actual files into the sandbox, via adhoc_tool's

output_files

. Which seems to obviate the need for a

resource

target, and a test dependency... Thoughts? Am I doing this right?

That error suggests

test_dataset.tar

doesn’t exist as a file on disk adjacent to that BUILD file. Is that the case? Is that what you expect?

Yes, that's correct. The

python_source

runs a script that fetches test data from another repository... The reason I do this is to set a custom header on the request. It's for authentication.

Ah, a target like

resource

is required for files on disk, it’s not required for ones generated by other targets that aren’t on disk.