remove upper bound from numpy version

[jkittner]

dask/crick#53 was resolved and tests pass locally for me.

This is causing issues for us since other packages started requiring numpy>=2.0

[jkittner]

#346 should resolve those 3.11 test failures - it looks like it captures the rasterio warnings?

[j08lue]

We can check once that one is merged and we rebase here.

[jkittner]

There are no 3.9 wheels for crick on pypi. It tries to install version 0.0.6 which is compiled with numpy < 2 and hence incompatible with numpy >= 2.

2024-10-24T10:32:57.6608536Z   Downloading crick-0.0.6-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (1.1 kB)

However, version 0.0.7 is available - but only with wheels starting at 3.10 since crick 0.0.7 dropped support for 3.9. https://github.com/dask/crick/blob/5b7f0c779ac873dfc895f39d41a6682e519764b6/pyproject.toml#L17 (see: dask/crick#57)

The only way to do this would be this:

diff --git a/setup.py b/setup.py
index 5b16c7e..be42428 100644
--- a/setup.py
+++ b/setup.py
@@ -79,6 +79,7 @@ setup(
         "tqdm",
     ],
     extras_require={
+        ':python_version == "3.9"': ['numpy<2.0.0'],
         "test": [
             "pytest",
             "pytest-cov",

which would mean terracotta would use numpy < 2 and crick 0.0.6 on 3.9 and numpy > 2 and crick > 0.0.6 on > 3.9 - I think that would be fine.

Tbh, quite an annoying mess these dependencies - sorry...

[j08lue]

Dask and other surrounding projects have already done this

• Drop Python 3.9 dask/crick#57

Dask references NEP 29 — Recommend Python and NumPy version support as a community policy standard for motivation to drop Python 3.9 support (in their changelog).

Should we follow, @dionhaefner?

[dionhaefner]

Hmm if #347 (comment) works out of the box I think I prefer that

[jkittner]

macos is missing the postgres binaries: pg_config (or they are not on PATH) that's why it fails compiling psycopg2 and building a wheel.
At least widnows and 3.9 works now...