Hi Sascha, Am Sun, Feb 25, 2024 at 08:32:52PM +0100 schrieb Sascha Steinbiss: > > nox > Creating virtual environment (virtualenv) using python3 in > > .nox/prepare > > nox > python -m pip install build numpy packaging PyYAML requests tomli > > nox > Command python -m pip install build numpy packaging PyYAML requests > > tomli failed with exit code 1: > > WARNING: The directory '/nonexistent/.cache/pip' or its parent directory is > > not owned or is not writable by the current user. The cache has been > > disabled. Check the permissions and owner of that directory. If executing > > pip with sudo, you should use sudo's -H flag. > > ... > Hmm, I never had that.
Sounds good. Maybe something is broken on my side? > The buildd builds also seem to pass this step with no > errors. Even better. > I am wondering if nox wants to pull anything from PyPI using pip and > either your DNS configuration is weird or your build container cannot access > the Internet (which than again it shouldn't)... but if buildds are not > supposed to access the Internet why don't the builds fail there? On the > buildds the issue there seems to be related to a script using an old(?) > Python YAML library syntax: > https://buildd.debian.org/status/package.php?p=python-awkward I admit I'm fine with the situation that the build works everywhere except for me as long as I have no reason to build it. ;-) > > Even worse: It requires pyarrow > > It does? The project metadata, where all other dependencies are listed, does > not mention it, and the dependencies that are listed there are all from > Debian: > > ... > dependencies = [ > "awkward_cpp==29", > "importlib_metadata>=4.13.0;python_version < \"3.12\"", > "numpy>=1.18.0", > "packaging", > "typing_extensions>=4.1.0; python_version < \"3.11\"", > "fsspec>=2022.11.0" > ] > ... > > Grepping through the test requirements though seems to indicate that the > tests in the awkward package need pyarrow: Its not only the tests: $ grep -Rl pyarrow src/* | wc -l 76 > ❯ grep pyarrow *requi* pyproject.toml > requirements-test-full.txt:pyarrow>=7.0.0;sys_platform != "win32" and > python_version < "3.12" > requirements-test-minimal.txt:pyarrow==7.0.0 > > But I disabled the Python unit tests (mostly because I couldn't get them to > work with the combined C++ and Python package built here) so I didn't catch > it. > I was also under the impression that awkward working with data from pyarrow > would just be one use case, but not a hard requirement for awkward to work. > Probably too optimistic given your observations :/ Seems so. Maybe we finally need to tackle pyarrow. > > ____________________ ERROR collecting tests/test_awkward.py > > ____________________ > > ImportError while importing test module > > '/tmp/autopkgtest.sD3d5G/autopkgtest_tmp/tests/test_awkward.py'. > > Hint: make sure your test modules/packages have valid Python names. > > Traceback: > > /usr/lib/python3.12/importlib/__init__.py:90: in import_module > > return _bootstrap._gcd_import(name[level:], package, level) > > tests/test_awkward.py:205: in <module> > > ak.str.to_categorical(ak.Array([["a", "b", "c"], ["a", "b"]])), > > /usr/lib/python3/dist-packages/awkward/_dispatch.py:62: in dispatch > > next(gen_or_result) > > /usr/lib/python3/dist-packages/awkward/operations/str/akstr_to_categorical.py:53: > > in to_categorical > > return _impl(array, highlevel, behavior, attrs) > > /usr/lib/python3/dist-packages/awkward/operations/str/akstr_to_categorical.py:60: > > in _impl > > pc = import_pyarrow_compute("ak.str.to_categorical") > > /usr/lib/python3/dist-packages/awkward/_connect/pyarrow.py:60: in > > import_pyarrow_compute > > raise ImportError(error_message.format(name)) > > E ImportError: to use ak.str.to_categorical, you must install pyarrow: > > In that case it looks like anndata needs a feature of awkward that also > needs pyarrow. Which is, yeah, a problem, given that we don't have anything > form Arrow in Debian (yet). Since pyarrow is just bindings that need the > rest of the library (https://arrow.apache.org/docs/python/index.html) then > this pulls in another large dependency. That's correct. > > Since python-anndata is the last package with Python3.12 issues there is > > some work > > left on this front. > > True. I am afraid that Arrow would be too much for me to take care of at the > moment, sorry. I started it a while ago but quickly found that that I bit > off more than I can chew without it becoming a full-time task. > (See my ITP and the thread on it) I'm aware of this. Thank you for your work on this anyway. Kind regards Andreas. -- http://fam-tille.de
