Numpy 2 compatibility by A-CGray · Pull Request #87 · mdolab/openconcept

A-CGray · 2026-03-21T04:02:09Z

Purpose

Makes openconcept compatible with numpy 2.4+

Type of change

What types of change is it?
Select the appropriate type(s) that describe this PR

- Bugfix (non-breaking change which fixes an issue)
- New feature (non-breaking change which adds functionality)
- Breaking change (non-backwards-compatible fix or feature)
- Code style update (formatting, renaming)
- Refactoring (no functional changes, no API changes)
- Documentation update
- Other (please describe)

Testing

Explain the steps needed to test the new code to verify that it does indeed address the issue and produce the expected behavior.

Checklist

Put an x in the boxes that apply.

I have run unit and regression tests which pass locally with my changes
I have added new tests that prove my fix is effective or that my feature works
I have added necessary documentation

np.complex_ was removed in NumPy 2.0. np.complex128 is semantically identical and available in all supported NumPy versions. Also fixes a pre-existing bug in combine_split_comp.py where dtype=np.dtype (the class itself) was passed instead of the local variable dtype. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

NumPy 2.0 removed automatic coercion of 1-element arrays to Python scalars, so math.cos/sin/sqrt no longer accept numpy array inputs. Replace with np.cos/sin/sqrt which handle arrays natively. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

…enMDAO 3.43 Small numerical differences (~3e-6 relative) in surge_margin (N3) and battery SOC_final (HybridTwin) arise from library version changes and are not correctness regressions. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

During Newton solver warmup iterations, fltcond|Utrue can temporarily go negative, making Re negative and causing log(negative) = NaN which propagates through the Jacobian. Fix: compute Re = U*L/visc_kin and apply abs(Re) for real-valued arrays only, so complex-step derivative checking still propagates the imaginary part correctly (for complex arrays, negate Re where real(Re)<0). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

jmarette · 2026-07-30T15:01:46Z

Hi, thanks for working on this.

We ran into the numpy >=1.20, <2 cap in setup.py. It makes OpenConcept 1.2.6 incompatible with current AeroSandbox, which wants numpy >= 2.2.6. This PR looked like the fix so we tried it.

Tested at 10f644b on Python 3.12.13, with numpy 2.5.1, OpenMDAO 3.45.0 and AeroSandbox 4.2.10. Full suite from source, same result on three runs:

461 passed, 2 failed, 2 skipped, 48 errors.

The 48 errors are all ModuleNotFoundError: No module named 'openaerostruct', which we simply didn't install. Nothing to do with numpy.

The 2 failures are HybridTwinThermalTestCase::test_values_thermalhybridtwin and N3HybridTestCase::test_defaults. Both are value checks. We get -3.80263173e-05 where -3.80158704e-05 is expected, so roughly 2.7e-5 relative against a 1e-5 tolerance.

That one might be worth your attention. The branch already carries "Loosen test tolerances for minor numerical shifts with NumPy 2.x / OpenMDAO 3.43", and that commit puts the drift at about 3e-6. What we see is an order of magnitude larger. Our guess is that numpy moved again after you calibrated it.

The main thing we wanted to ask about is the history. These changes went into main on 2026-03-21 and were reverted the same day, by 9187a53 and 14b7176. Neither commit message says why, and there are no comments here.

Did you hit something concrete, or was it backed out so it could land through this PR instead? We'd rather not pin an unmerged branch if you already know something is wrong with it.

If it's the latter, happy to help. We can post the full test log, run other configurations, or send a PR for the tolerance calibration.

np.trapz and np.ComplexWarning were removed in NumPy 2, and NumPy 2.5 no longer allows assigning size-1 arrays to scalar array elements. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Exclude mdot_coolant from the propulsion input promotion with glob character classes and connect it explicitly, instead of listing every ac|propulsion input by hand. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Use U * sign(U.real), which is |U| for real inputs and remains complex-step safe, rather than branching on the input dtype. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

The SOC_final reference values were under-converged results whose last digits depend on the Newton iteration path, which changed in OpenMDAO 3.43. Use the tightly converged values with an absolute tolerance consistent with the solver tolerance, and revert the earlier tolerance loosening (the N3 discrepancy came from stale local engine deck caches, not NumPy). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

OpenAeroStruct 2.7.1 does not support NumPy 2. OAS 2.10 does, but it evaluates B-splines of panel quantities (t/c, skin and spar thickness) at panel centres, which changes the aerostructural results; with uniform distributions the results are unchanged. Update the reference values, make run_OAS set the panel t/c directly, and bump the oldest tested versions to OpenMDAO 3.35 (required by OAS 2.10), NumPy 1.25 and SciPy 1.11. Install pyCycle from GitHub in the latest build because its NumPy 2 fixes are unreleased. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

codecov · 2026-10-27T07:17:15Z

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.06%. Comparing base (0d2adeb) to head (5e47f96).

Additional details and impacted files

@@           Coverage Diff           @@
##             main      #87   +/-   ##
=======================================
  Coverage   82.05%   82.06%           
=======================================
  Files         103      103           
  Lines       10730    10733    +3     
=======================================
+ Hits         8805     8808    +3     
  Misses       1925     1925

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

With Python 3.11, installing pyCycle pulled in the latest OpenMDAO, which requires NumPy 2 and silently replaced the oldest NumPy and SciPy. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

The glob character classes also excluded any other input starting with "t" or "m". Promote everything with "ac|propulsion|*" and rename just mdot_coolant with an explicit tuple, which takes precedence over the wildcard. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Promote each ac|propulsion input by its full name, leaving out only the per-node mdot_coolant, rather than renaming it out of the ac| namespace. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Since OpenMDAO 3.35 every problem creates a <name>_out directory in the working directory, so running the tests left hundreds of them in the repo root. Point OPENMDAO_WORKDIR at a temporary directory for the pytest session unless the user has already set it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

OpenMDAO is now pinned when pyCycle is installed, so the later step that reinstalled the oldest OpenMDAO had nothing left to do. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

- Use a single Python setup step that picks the oldest or latest version - Let pyCycle and the testing and docs extras install in one pip call, rather than installing pyCycle, NumPy, SciPy and OpenMDAO separately - Drop the conda always_yes config, which the latest build never needed, and pass -y to the one conda install instead Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Give each dep-versions value its own versions through matrix include, so the job names list the OS and the Python, NumPy, SciPy and OpenMDAO versions instead of every version setting for both builds. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Writing outputs to a temporary directory broke the Windows build: the temporary directory and the checkout are on different drives, and OpenMDAO's inputs report calls os.path.relpath on its output file. The reports are what create the <name>_out directories, so turn them off. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

A-CGray · 2026-10-27T17:44:55Z

@sabakhshi can you post the details of the environment you're having test failures with?

sabakhshi · 2026-10-27T17:58:46Z

@sabakhshi can you post the details of the environment you're having test failures with?

certifi==2026.10.23 charset-normalizer==3.5.2 contourpy==1.4.0 cycler==0.12.1 fonttools==4.68.0 idna==3.20 iniconfig==2.3.1 kiwisolver==1.5.1 matplotlib==3.11.2 networkx==3.7 numpy==2.5.4 -e git+ssh://git@github.com/OpenMDAO/pyCycle.git@ee7e16126a7a54b5fe35292c53b7eb4e9572620b#egg=om_pycycle -e git+ssh://git@github.com/sabakhshi/OpenAeroStruct.git@0a4d614f9494cbadaba4a6fa32a33d6e54e14f34#egg=openaerostruct -e git+ssh://git@github.com/mdolab/openconcept.git@5e47f965c6b34f3f7d8dd057b504a47896f7b16a#egg=openconcept openmdao==3.45.1 packaging==26.3 parameterized==0.9.0 pillow==13.0.0 pluggy==1.7.0 Pygments==2.21.0 pyparsing==3.3.3 pytest==9.1.1 python-dateutil==3.0.0 requests==2.34.2 scipy==1.18.1 six==1.17.0 tzdata==2026.5 urllib3==2.8.0

A-CGray · 2026-10-27T18:07:53Z

@sabakhshi which python version?

sabakhshi · 2026-10-27T18:38:55Z

@sabakhshi which python version?

3.12.3

A-CGray · 2026-10-27T20:38:25Z

I cannot recreate the failure you're seeing. Here's the full list of commands I ran (in the root dir of openconcept):

pyenv install 3.12.3
pyenv virtualenv 3.12.3 openconceptTest
pyenv activate openconceptTest
pip install -e .[testing]
pip install git+https://github.com/OpenMDAO/pyCycle.git
pytest -v

A-CGray and others added 7 commits March 20, 2026 23:57

Remove numpy upper bound

ecbe263

Update testing versions

30d53e1

Fix promotion issue

10f644b

A-CGray requested a review from a team as a code owner March 21, 2026 04:02

A-CGray requested a review from sabakhshi March 21, 2026 04:02

A-CGray marked this pull request as draft June 11, 2026 15:25

A-CGray and others added 7 commits August 6, 2026 12:14

Merge branch 'main' into feature-numpy2-compat

ee0c3d2

Fix remaining NumPy 2 incompatibilities in OpenAeroStruct wrappers

1d2b4cf

np.trapz and np.ComplexWarning were removed in NumPy 2, and NumPy 2.5 no longer allows assigning size-1 arrays to scalar array elements. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Simplify HybridTwin_thermal mdot_coolant promotion fix

e070fbf

Exclude mdot_coolant from the propulsion input promotion with glob character classes and connect it explicitly, instead of listing every ac|propulsion input by hand. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Simplify negative Reynolds number guard

b829453

Use U * sign(U.real), which is |U| for real inputs and remains complex-step safe, rather than branching on the input dtype. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Leave commented-out formulas in weights_turboprop unchanged

130c7e4

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Stop the oldest CI build from upgrading NumPy

b3006c6

With Python 3.11, installing pyCycle pulled in the latest OpenMDAO, which requires NumPy 2 and silently replaced the oldest NumPy and SciPy. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

A-CGray marked this pull request as ready for review October 27, 2026 14:30

A-CGray and others added 6 commits October 27, 2026 10:37

List HybridTwin_thermal ac|propulsion inputs explicitly

ffca936

Promote each ac|propulsion input by its full name, leaving out only the per-node mdot_coolant, rather than renaming it out of the ac| namespace. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Remove redundant OpenMDAO reinstall from the oldest CI build

1078d6c

OpenMDAO is now pinned when pyCycle is installed, so the later step that reinstalled the oldest OpenMDAO had nothing left to do. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

A-CGray changed the title ~~Numpy 2 comatibility~~ Numpy 2 compatibility Oct 27, 2026

A-CGray and others added 2 commits October 27, 2026 11:10

Update GHA names

761dd5a

A-CGray and others added 3 commits October 27, 2026 11:59

Bump version to 1.3.0

0847417

minor code style tweak to make it clear trapazoid came from a package

954e6ce

Tighten solver tolerance in problematic test

5e47f96

A-CGray merged commit 6351bf4 into main Oct 27, 2026
11 checks passed

A-CGray deleted the feature-numpy2-compat branch October 27, 2026 22:32

Navigation Menu

Repository navigation

Numpy 2 compatibility - #87

Numpy 2 compatibility#87
A-CGray merged 26 commits into
mainfrom
feature-numpy2-compat

A-CGray commented Mar 21, 2026 •

edited

Loading

Uh oh!

jmarette commented Jul 30, 2026

Uh oh!

codecov Bot commented Oct 27, 2026 •

edited

Loading

Uh oh!

A-CGray commented Oct 27, 2026

Uh oh!

sabakhshi commented Oct 27, 2026

Uh oh!

A-CGray commented Oct 27, 2026

Uh oh!

sabakhshi commented Oct 27, 2026

Uh oh!

A-CGray commented Oct 27, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

A-CGray commented Mar 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Purpose

Type of change

Testing

Checklist

Uh oh!

jmarette commented Jul 30, 2026

Uh oh!

codecov Bot commented Oct 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

A-CGray commented Oct 27, 2026

Uh oh!

sabakhshi commented Oct 27, 2026

Uh oh!

A-CGray commented Oct 27, 2026

Uh oh!

sabakhshi commented Oct 27, 2026

Uh oh!

A-CGray commented Oct 27, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

A-CGray commented Mar 21, 2026 •

edited

Loading

codecov Bot commented Oct 27, 2026 •

edited

Loading