SPEC 0: Bump minimum supported version to NumPy 2.0

[seisman]

Following SPEC0 policy, NumPy 1.26 should be dropped in 2025 quarter 3.

Bumps the minimum supported NumPy version to 2.0 in the following files:

• .github/workflows/ci_tests.yaml
• .github/workflows/ci_tests_legacy.yaml
• environment.yml
• pyproject.toml
• requirements.txt
• Remove workarounds for unsupported versions

Supersedes #3916.

[seisman]

https://github.com/GenericMappingTools/pygmt/actions/runs/17768889118/job/50499354183?pr=4090
The Python 3.11 CI job fails with the following error:

  error    libmamba Could not solve for environment specs
      The following packages are incompatible
      ├─ numpy =2.0 * is requested and can be installed;
      └─ xarray =2023.7 * is not installable because it requires
         └─ numpy >=1.21,<2.0a0 *, which conflicts with any installable versions previously reported.
  critical libmamba Could not solve for environment specs

Likely need to bump xarray first. See #4091.

The Python 3.11 CI job has two failing tests due to the removal of the conftest.py file. The failures are related to #2628. I think we should fix them to match the behavior of NumPy>=2.0

=================================== FAILURES ===================================
__________________ [doctest] pygmt.datatypes.image._GMT_IMAGE __________________
063 1 0
064 >>> x  # doctest: +NORMALIZE_WHITESPACE, +ELLIPSIS
065 array([-179.5, -178.5, ..., 178.5, 179.5])
066 >>> y  # doctest: +NORMALIZE_WHITESPACE, +ELLIPSIS
067 array([ 89.5,  88.5, ..., -88.5, -89.5])
068 >>> data.dtype
069 dtype('uint8')
070 >>> data.shape
071 (180, 360, 3)
072 >>> data.min(), data.max()
Expected:
    (10, 255)
Got:
    (np.uint8(10), np.uint8(255))

/home/runner/work/pygmt/pygmt/pygmt/datatypes/image.py:72: DocTestFailure
_____________ [doctest] pygmt.datatypes.image._GMT_IMAGE.to_xarray _____________
109 --------
110 >>> from pygmt.clib import Session
111 >>> with Session() as lib:
112 ...     with lib.virtualfile_out(kind="image") as voutimg:
113 ...         lib.call_module("read", ["@earth_day_01d", voutimg, "-Ti"])
114 ...         # Read the image from the virtual file
115 ...         image = lib.read_virtualfile(voutimg, kind="image")
116 ...         # Convert to xarray.DataArray and use it later
117 ...         da = image.contents.to_xarray()
118 >>> da  # doctest: +NORMALIZE_WHITESPACE, +ELLIPSIS
Differences (unified diff with -expected +actual):
    @@ -1,3 +1,3 @@
    -<xarray.DataArray 'z' (band: 3, y: 180, x: 360)>...
    +<xarray.DataArray 'z' (band: 3, y: 180, x: 360)> Size: 194kB
     array([[[ 10,  10,  10, ...,  10,  10,  10],
             [ 10,  10,  10, ...,  10,  10,  10],
    @@ -22,9 +22,10 @@
             [177, 179, 179, ..., 178, 177, 177],
             [185, 187, 187, ..., 187, 186, 185],
    -        [189, 191, 191, ..., 191, 191, 189]]], dtype=uint8)
    +        [189, 191, 191, ..., 191, 191, 189]]],
    +      shape=(3, 180, 360), dtype=uint8)
     Coordinates:
    -  * y        (y) float64... 89.5 88.5 87.5 86.5 ... -86.5 -87.5 -88.5 -89.5
    -  * x        (x) float64... -179.5 -178.5 -177.5 -176.5 ... 177.5 178.5 179.5
    -  * band     (band) uint8... 1 2 3
    +  * y        (y) float64 1kB 89.5 88.5 87.5 86.5 ... -86.5 -87.5 -88.5 -89.5
    +  * x        (x) float64 3kB -179.5 -178.5 -177.5 -176.5 ... 177.5 178.5 179.5
    +  * band     (band) uint8 3B 1 2 3
     Attributes:
    -    long_name:     z
    +    long_name:  z

[weiji14]

      The following packages are incompatible
      ├─ numpy =2.0 * is requested and can be installed;
      └─ xarray =2023.10 * is not installable because there are no viable options
         ├─ xarray 2023.10.0 would require
         │  └─ numpy >=1.21,<2.0a0 *, which conflicts with any installable versions previously reported;
         └─ xarray [2023.10.0|2023.10.1] would require
            └─ numpy >=1.22,<2.0a0 *, which conflicts with any installable versions previously reported.

There is a numpy<2.0 pin set at conda-forge/conda-forge-repodata-patches-feedstock#728, we might need to bump xarray higher, to 2024.5.0 I think?

[seisman]

https://github.com/GenericMappingTools/pygmt/actions/runs/17768889118/job/50499354183?pr=4090 The Python 3.11 CI job fails with the following error:

  error    libmamba Could not solve for environment specs
      The following packages are incompatible
      ├─ numpy =2.0 * is requested and can be installed;
      └─ xarray =2023.7 * is not installable because it requires
         └─ numpy >=1.21,<2.0a0 *, which conflicts with any installable versions previously reported.
  critical libmamba Could not solve for environment specs

Likely need to bump xarray first. See #4091.

Still failing. The minimum xarray version compatible with numpy 2.0 is xarray 2024.5:

$ mamba search xarray=2024.5 --info                                                                                                                                                         Loading channels: done
xarray 2024.5.0 pyhd8ed1ab_0
----------------------------
file name   : xarray-2024.5.0-pyhd8ed1ab_0.conda
name        : xarray
version     : 2024.5.0
build       : pyhd8ed1ab_0
build number: 0
size        : 768 KB
license     : Apache-2.0
subdir      : noarch
url         : https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/noarch/xarray-2024.5.0-pyhd8ed1ab_0.conda
md5         : e839fd0ae78a368c930f0b1feafa6736
timestamp   : 2024-05-13 13:49:27 UTC
constraints :
  - hdf5 >=1.12
  - distributed >=2023.4
  - nc-time-axis >=1.4
  - cartopy >=0.21
  - h5py >=3.8
  - h5netcdf >=1.1
  - toolz >=0.12
  - dask-core >=2023.4
  - pint >=0.22
  - scipy >=1.10
  - numba >=0.56
  - cftime >=1.6
  - seaborn-base >=0.12
  - iris >=3.4
  - zarr >=2.14
  - flox >=0.7
  - bottleneck >=1.3
  - sparse >=0.14
  - matplotlib-base >=3.7
  - netcdf4 >=1.6.0
dependencies:
  - numpy >=1.23
  - packaging >=23.1
  - pandas >=2.0
  - python >=3.9


$ mamba search xarray=2024.3 --info
Loading channels: done
xarray 2024.3.0 pyhd8ed1ab_0
----------------------------
file name   : xarray-2024.3.0-pyhd8ed1ab_0.conda
name        : xarray
version     : 2024.3.0
build       : pyhd8ed1ab_0
build number: 0
size        : 747 KB
license     : Apache-2.0
subdir      : noarch
url         : https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/noarch/xarray-2024.3.0-pyhd8ed1ab_0.conda
md5         : 772d7ee42b65d0840130eabd5bd3fc17
timestamp   : 2024-03-29 19:57:37 UTC
constraints :
  - bottleneck >=1.3
  - sparse >=0.13
  - nc-time-axis >=1.4
  - scipy >=1.8
  - zarr >=2.12
  - flox >=0.5
  - netcdf4 >=1.6.0
  - cartopy >=0.20
  - h5netcdf >=1.0
  - dask-core >=2022.7
  - cftime >=1.6
  - numba >=0.55
  - hdf5 >=1.12
  - iris >=3.2
  - toolz >=0.12
  - h5py >=3.6
  - distributed >=2022.7
  - matplotlib-base >=3.5
  - seaborn-base >=0.11
  - pint >=0.19
dependencies:
  - numpy >=1.23,<2.0a0
  - packaging >=22
  - pandas >=1.5
  - python >=3.9

I guess we need to either bump to xarray 2024.5 now or wait until 2026 quarter 2.

[weiji14]

I guess we need to either bump to xarray 2024.5 now or wait until 2026 quarter 2.

xarray=2024.06.0 according to numpy/numpy#26191 (comment) (unsure if it's the repodata patch that's wrong, or that linked thread is wrong, but I did find pydata/xarray#8946).

How about we delay the numpy 2.0+ bump for another quarter, i.e. leave this PR in draft until November/December 2025?