Issues building from source

[dlakaplan]

Hi. I am trying to build this from source so I can explore nanograv/PINT#1944 without waiting for new pypi or conda releases. However, I cannot follow the instructions at https://github.com/numpy/numpy-user-dtypes/tree/main/quaddtype#installation-from-source. Some of the issues may just be missing steps in the instructions, others appear more challenging:

• The instructions do not say that meson is needed, although that is an easy fix
• The instructions do not say that sleef is needed. I tried to install that from conda, we'll see if that works
• I am not sure how to install the sleefquad library. I get:

  Run-time dependency sleef found: YES 3.9.0
  Library sleefquad found: NO
  ../meson.build:65: WARNING: Falling back to SLEEF subproject: sleefquad library is missing

and I don't know if this is a problem or not. I cannot find a separate source for sleefquad, so this may be a problem with how I installed sleef

• The instructions say both pip install numpy pytest and NumPy >= 2.4 (build from source) which seem incompatible. When I try to build I get many errors like:

  ../numpy_quaddtype/src/dtype.c:210:14: error: use of undeclared identifier 'NPY_CONSTANT_zero'
    210 |         case NPY_CONSTANT_zero:
        |              ^

which makes it look like it's missing a include file or some other source file.

Can you help with this? How should sleef and numpy be installed?

[dlakaplan]

Other info:

Using pip 25.2 from /Users/kaplan/miniforge3/envs/nptest/lib/python3.14/site-packages/pip (python 3.14)
Processing /Users/kaplan/pythonpackages/numpy-user-dtypes/quaddtype
  Preparing metadata (pyproject.toml): started
  Running command Preparing metadata (pyproject.toml)
  + meson setup /Users/kaplan/pythonpackages/numpy-user-dtypes/quaddtype /Users/kaplan/pythonpackages/numpy-user-dtypes/quaddtype/.mesonpy-wkn811uy -Dbuildtype=release -Db_ndebug=if-rel
ease -Db_vscrt=md --native-file=/Users/kaplan/pythonpackages/numpy-user-dtypes/quaddtype/.mesonpy-wkn811uy/meson-python-native-file.ini
  The Meson build system
  Version: 1.9.1
  Source dir: /Users/kaplan/pythonpackages/numpy-user-dtypes/quaddtype
  Build dir: /Users/kaplan/pythonpackages/numpy-user-dtypes/quaddtype/.mesonpy-wkn811uy
  Build type: native build
  Project name: numpy_quaddtype
  Project version: undefined
  C compiler for the host machine: cc (clang 17.0.0 "Apple clang version 17.0.0 (clang-1700.3.19.1)")
  C linker for the host machine: cc ld64 1221.4
  C++ compiler for the host machine: c++ (clang 17.0.0 "Apple clang version 17.0.0 (clang-1700.3.19.1)")
  C++ linker for the host machine: c++ ld64 1221.4
  Host machine cpu family: aarch64
  Host machine cpu: aarch64
  Program python found: YES (/Users/kaplan/miniforge3/envs/nptest/bin/python)
  Found pkg-config: YES (/opt/local/bin/pkg-config) 0.29.2
  Run-time dependency python found: YES 3.14
  Found CMake: /opt/local/bin/cmake (3.31.3)
  Run-time dependency qblas found: NO (tried pkgconfig, framework and cmake)

[ngoldbaum]

and I don't know if this is a problem or not

You don't need a system install of SLEEF. If one isn't available then the build will set one up using the vendored version set up as a meson subproject in the subprojects folder.

I think the bigger problem with the instructions is that right now quanddtype depends on an unreleased version of NumPy (NumPy 2.4), so you need to install NumPy from a checkout of the NumPy git repo before you build quaddtype.

So instead of pip install numpy pytest, you'd do:

git clone https://github.com/numpy/numpy
cd numpy
git submodule --init --recursive
pip install .
pip install pytest
cd ..

Also as you pointed out, the instructions are missing meson and meson-python, so you'll need pip install meson meson-python too.

And then follow the rest of the instructions.

Sorry for the trouble. I'll update the instructions to suggest installing NumPy from source and to install meson and meson-python.

[ngoldbaum]

I opened #230 - do those instructions help?

[dlakaplan]

Yes, definitely. And maybe reassure users that they don't need to separately install sleef. Note that I just tried this with the numpy source install and it seemed to work. So I think this can be closed by #230.

[dlakaplan]

OK, I tried installing the way you describe in #230, and that just gets 2.3.5.

[dlakaplan]

And after installing 2.4.0, I am finding incompatibilities with astropy (7.1.1). Still investigating if these are fixed in a dev version.

[ngoldbaum]

OK, I tried installing the way you describe in #230, and that just gets 2.3.5.

That's odd. It's not what I get:

goldbaum at Nathans-MBP in ~/Documents/srsly on master!
± pip install "numpy @ git+https://github.com/numpy/numpy.git"
Collecting numpy@ git+https://github.com/numpy/numpy.git
  Cloning https://github.com/numpy/numpy.git to /private/var/folders/nk/yds4mlh97kg9qdq745g715rw0000gn/T/pip-install-6teiybqv/numpy_3ffe2d5cfbaa4cefa7125c870ee5c230
  Running command git clone --filter=blob:none --quiet https://github.com/numpy/numpy.git /private/var/folders/nk/yds4mlh97kg9qdq745g715rw0000gn/T/pip-install-6teiybqv/numpy_3ffe2d5cfbaa4cefa7125c870ee5c230
  Resolved https://github.com/numpy/numpy.git to commit 598dbf69a66efafa310b6edc3abe634fba970c1f
  Running command git submodule update --init --recursive -q
  Installing build dependencies ... done
  Getting requirements to build wheel ... done
  Preparing metadata (pyproject.toml) ... done

[notice] A new release of pip is available: 25.2 -> 25.3
[notice] To update, run: pip install --upgrade pip

goldbaum at Nathans-MBP in ~/Documents/srsly on master!
± python
Python 3.14.0 free-threading build (main, Oct  8 2025, 10:32:29) [Clang 17.0.0 (clang-1700.3.19.1)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import numpy as np
>>> np.__version__
'2.4.0.dev0+git20251120.598dbf6'

[ngoldbaum]

@SwayamInSync would it be hard to restore compatibility with older versions of NumPy? Maybe with some preprocessor macros to disable functionality when an old version is found at build time?

[dlakaplan]

Going to numpy 2.4.0 is bringing up a lot of incompatibility issues with PINT more generally (implicit typecasts that now need to be explicit). I think we will have to deal with those eventually but it's also slowing down the testing.

[SwayamInSync]

@SwayamInSync would it be hard to restore compatibility with older versions of NumPy? Maybe with some preprocessor macros to disable functionality when an old version is found at build time?

Easily doable with conditional compilation and skipping corresponding tests, but this will lose a good amount of features and common operations like arange, linspace, etc won't work + if I remember PINT requires heavy use of np.finfo and this might not give the fair chance of testing

[SwayamInSync]

And after installing 2.4.0, I am finding incompatibilities with astropy (7.1.1). Still investigating if these are fixed in a dev version.

On inspecting I think this is from astropy as they are skipping tests on every numpy version but running on 2.4-dev (notice that NUMPY_LT_2_4, it expands to not minversion(np, "2.4.0.dev0"))

def test_all_arguments_reexposed(self, target, helper):
        try:
            sig_target = inspect.signature(target)
        except ValueError:
            if NUMPY_LT_2_4:
                pytest.skip("Non Python function cannot be inspected at runtime")
            elif target is np.fromstring:
                pytest.skip(f"known case of missing runtime signature ({target})")
            else:
                raise

        params_target = sig_target.parameters
        sig_helper = inspect.signature(helper)

They are fetching the signature but I think __text__signature__ is None this is in all versions of NumPy even in the current LTS but it won't trigger as these tests will get skipped on them.

Maybe they are assuming that we fixed this? or maybe it'll get updated when 2.4 becomes LTS?

[dlakaplan]

Note that I was able to get things to work using the dev (8.0) build of astropy.

[SwayamInSync]

Note that I was able to get things to work using the dev (8.0) build of astropy.

Awesome, we will be looking forward to your experience.