Not able to build on debian11 arm Python 3.10 about levenshtein HOT 18 CLOSED

I have reinstalled again and it is ok:

Python 3.10.7 (main, Sep 25 2022, 13:32:17) [GCC 10.2.1 20210110] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from rapidfuzz import fuzz_cpp
>>> 

sorry for fake claims, apparently at the first reinstallation maybe I took some previous wheel or maybe even accidentally installed rapidfuzz-2.10.3-py3-none-any.whl instead of rapidfuzz-2.10.3-cp310-cp310-linux_armv7l.whl . Not sure. Anyhow all good now. Thanks.
Not sure if I should close this or you want to close it as soon as the scikit-build regression is fixed, which was the original cause of this bug for Levenshtein.

from levenshtein.

Ah yes that is a bug in scikit-build: scikit-build/scikit-build#718, so you can workaround this by hardcoding the version in the pyproject.toml: "scikit-build==0.14.1". I am experiencing this when developing on Fedora as well.

I still need to test whether the build passes with 0.14.1 everywhere. If it does I I will hardcode it until there is a fix in scikit-build.

from levenshtein.

Uh this requires more work, since the piwheels build fails.
Seems like either their compiler does not support C++17 or there is a compiler bug I need to workaround ...

from levenshtein.

well, I don't observe anymore fails after forcing scikit-build to use 0.14.1 version:

$ pip3 install .
Defaulting to user installation because normal site-packages is not writeable
Looking in indexes: https://pypi.org/simple, https://www.piwheels.org/simple
Processing /home/juris/Levenshtein
  Installing build dependencies ... done
  Getting requirements to build wheel ... done
  Preparing metadata (pyproject.toml) ... done
Requirement already satisfied: rapidfuzz<3.0.0,>=2.3.0 in /home/juris/.local/lib/python3.10/site-packages (from Levenshtein==0.20.5) (2.10.2)
Requirement already satisfied: jarowinkler<2.0.0,>=1.2.2 in /home/juris/.local/lib/python3.10/site-packages (from rapidfuzz<3.0.0,>=2.3.0->Levenshtein==0.20.5) (1.2.3)
Building wheels for collected packages: Levenshtein
  Building wheel for Levenshtein (pyproject.toml) ... done
  Created wheel for Levenshtein: filename=Levenshtein-0.20.5-cp310-cp310-linux_armv7l.whl size=165564 sha256=5b6eb87aed1223ee3b78db4ffff1f6c334b37a888b426aadf822da60b3b0b129
  Stored in directory: /tmp/pip-ephem-wheel-cache-m9xx6z2k/wheels/b3/27/31/93160a7a7432fa29e69bfdcd8d0b87fe1c99a6a6b5950d1ab5
Successfully built Levenshtein
Installing collected packages: Levenshtein
  Attempting uninstall: Levenshtein
    Found existing installation: Levenshtein 0.20.5
    Uninstalling Levenshtein-0.20.5:
      Successfully uninstalled Levenshtein-0.20.5
Successfully installed Levenshtein-0.20.5

from levenshtein.

You should probably build rapidfuzz from source as well, since the build on piwheels is broken right now due to this bug in scikit-build + a bug in gcc ...
Otherwise you only have the pure Python version, which is quite a bit slower.

from levenshtein.

I think I got from pip some prebuilt rapidfuzz, so I have everything allright now:

Python 3.10.7 (main, Sep 25 2022, 13:32:17) [GCC 10.2.1 20210110] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from rapidfuzz import process, fuzz
>>> choices = ["Atlanta Falcons", "New York Jets", "New York Giants", "Dallas Cowboys"]
>>> process.extract("new york jets", choices, scorer=fuzz.WRatio, limit=2)
[('New York Jets', 100.0, 1), ('New York Giants', 78.57142857142857, 2)]

from levenshtein.

Can you try:

from rapidfuzz import fuzz_cpp

from levenshtein.

ah well this then means I am using the slow version:

Python 3.10.7 (main, Sep 25 2022, 13:32:17) [GCC 10.2.1 20210110] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from rapidfuzz import fuzz_cpp
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ImportError: cannot import name 'fuzz_cpp' from 'rapidfuzz' 

from levenshtein.

Yes I expected this, since the arm build on https://www.piwheels.org/project/rapidfuzz/ was silently broken. I hope this will be fixed by the changes in rapidfuzz/RapidFuzz@24bf798

from levenshtein.

i took your latest commit from rapidfuzz and I am afraid the problem is still there. This is what I got after build in rapidfuzz repo:

WARNING: The C extension could not be compiled, speedups are not enabled.
Plain-Python build succeeded.

from levenshtein.

Could you try:

RAPIDFUZZ_BUILD_EXTENSION="1" pip install . -v

to check why the c++ build fails?

from levenshtein.

this would be the full result: dump.txt

from levenshtein.

It appears you did not clone the submodules. You should be able to get them using:

git submodule update --init

from levenshtein.

yes sorry those submodules were culprit, it built the thing, not sure though why this would still complain:

>>> from rapidfuzz import fuzz_cpp
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ImportError: cannot import name 'fuzz_cpp' from 'rapidfuzz'

from levenshtein.

Hm can you check whether the fuzz_cpp.*.so file is present in the corresponding site-packages directory (site-packages/rapidfuzz/fuzz)?
If it is you can try to cd into the directory and try to import fuzz_cpp in there.

from levenshtein.

if the file is present, what is the file name? (to check whether the platform tag is correct)

from levenshtein.

I think we should keep this open, for anyone else running into this issue until it is resolved in scikit-build.

from levenshtein.

This should be fixed with the latest version of scikit-build

from levenshtein.