Very Mysterious Unit Test Failure in ReactorTest.Py

[sevyharris]

Bug Description

The Continuous Integration tests started failing last night.
https://github.com/ReactionMechanismGenerator/RMG-Py/actions/runs/21060625788/job/60565859664

For some reason, the Cantera reaction object (made using the .from_yaml call) is slightly altering the rate constant:

FAILED reactionTest.py::TestReactionToCantera::test_arrhenius - AssertionError: assert {'rate-constant': {'A': 660000.0, 'b': 1.62, 'Ea': 45354560.0}} == {'rate-constant': {'A': 660000.0000000001, 'b': 1.62, 'Ea': 45354560.0}}

  Differing items:
  {'rate-constant': {'A': 660000.0, 'Ea': 45354560.0, 'b': 1.62}} != {'rate-constant': {'A': 660000.0000000001, 'Ea': 45354560.0, 'b': 1.62}}

  Full diff:
    {
        'rate-constant': {
  -         'A': 660000.0000000001,
  ?                      ---------
  +         'A': 660000.0,
            'Ea': 45354560.0,
            'b': 1.62,
        },
    }
FAILED reactionTest.py::TestReactionToCantera::test_multi_arrhenius - assert 0.001 == 0
 +  where 0.001 = round(0.0009765625, 3)
 +    where 0.0009765625 = abs((5000000000000.0 - 5000000000000.001))
 +      where 5000000000000.0 = <ArrheniusRate at 7f9185b7c7b0>.pre_exponential_factor
 +        where <ArrheniusRate at 7f9185b7c7b0> = HO2(5) + OH(4) <=> H2O(27) + O2(6)    <Reaction(Arrhenius)>.rate
 +      and   5000000000000.001 = <ArrheniusRate at 7f9185b7cd70>.pre_exponential_factor
 +        where <ArrheniusRate at 7f9185b7cd70> = HO2(5) + OH(4) <=> H2O(27) + O2(6)    <Reaction(Arrhenius)>.rate
FAILED reactionTest.py::TestChargeTransferReaction::test_get_rate_coeff - assert 0.0078125 < 1e-06
 +  where 0.0078125 = abs((43870506959778.99 - 43870506959779.0))
=========================================================== 3 failed, 55 passed, 2 skipped, 7 warnings in 9.43s ============================================================

What's strange is that I can't for the life of me see why the test started failing now. The last PR passed these tests and didn't mess with any of the testing, cantera, or conda environment code.

How To Reproduce

Run pytest reactionTest.py in the RMG-Py/test/rmgpy directory on the main branch with the latest version of RMG.

Expected Behavior

CI unit tests should be passing so we can easily evaluate new code suggestions.

One other mysterious aspect of this is that I tried rolling back the version on the main branch by 20+ commits and it still fails. 4a6e9fc is what I tried. I would expect that if the problem existed back in July, CI would've caught it by now.

My hunch is that it has something to do with conda environments changing how ct.Reaction.from_yaml() gets executed. There's some conversion on the Cantera side that's multiplying the reaction rate by something that mangles some bits of the float.

Installation Information

Describe your installation method and system information.

• OS: Ubuntu
  • RMG-Py: e848daa

Additional Context

I have a medium-out-of-date branch on the Northeastern Explorer cluster where this test passes: https://github.com/sevyharris/RMG-Py/tree/autoscience_uncertainties_2025

[sevyharris]

The conda versions of Cantera seem to be identical between the failed and successful runs... (copied from the Github Actions logs)

January 16th failure:
2026-01-16T08:38:59.0385762Z + cantera 2.6.0 py39hcdf224b_5 conda-forge 3MB
2026-01-16T08:38:59.0433448Z + libcantera 2.6.0 h587f4e0_5 conda-forge 2MB

January 15th success:
2026-01-15T18:17:02.4576601Z + cantera 2.6.0 py39hcdf224b_5 conda-forge 3MB
2026-01-15T18:17:02.4667587Z + libcantera 2.6.0 h587f4e0_5 conda-forge 2MB

[JacksonBurns]

Could you do a full diff of the conda environment?

[sevyharris]

There is no difference 😯

I saved the diff here:
https://www.diffchecker.com/pEIA8BUn/

The first column with the timestamp is deleted from each of the original files:

conda_20260115.txt
conda_20260116.txt

awk '{$1=""}1' conda_20260115.txt | awk '{$1=$1}1' > conda_env_20260115.txt for my own future reference

[sevyharris]

It looks like some of the daily CI runs since 1/15/2026 have passed and some failed, and Python 3.9 ubuntu-latest Build and Test with RMS is always the job at fault.

On 20260118, Python 3.9 ubuntu-latest Build and Test passes test_arrhenius

2026-01-18T09:09:09.1056051Z test/rmgpy/reactionTest.py::TestReactionToCantera::test_arrhenius PASSED
2026-01-18T09:09:09.1066529Z test/rmgpy/reactionTest.py::TestReactionToCantera::test_multi_arrhenius PASSED
2026-01-18T09:09:09.1089400Z test/rmgpy/reactionTest.py::TestReactionToCantera::test_pdep_arrhenius PASSED
2026-01-18T09:09:09.1101065Z test/rmgpy/reactionTest.py::TestReactionToCantera::test_multi_pdep_arrhenius PASSED
2026-01-18T09:09:09.1111246Z test/rmgpy/reactionTest.py::TestReactionToCantera::test_chebyshev PASSED
2026-01-18T09:09:09.1126010Z test/rmgpy/reactionTest.py::TestReactionToCantera::test_falloff PASSED

While Python 3.9 ubuntu-latest Build and Test with RMS fails test_arrhenius:

2026-01-18T09:24:52.6026626Z test/rmgpy/reactionTest.py::TestReactionToCantera::test_arrhenius FAILED
2026-01-18T09:24:52.6427976Z test/rmgpy/reactionTest.py::TestReactionToCantera::test_multi_arrhenius FAILED
2026-01-18T09:24:52.6441576Z test/rmgpy/reactionTest.py::TestReactionToCantera::test_pdep_arrhenius PASSED
2026-01-18T09:24:52.6453458Z test/rmgpy/reactionTest.py::TestReactionToCantera::test_multi_pdep_arrhenius PASSED
2026-01-18T09:24:52.6463717Z test/rmgpy/reactionTest.py::TestReactionToCantera::test_chebyshev PASSED
2026-01-18T09:24:52.6482640Z test/rmgpy/reactionTest.py::TestReactionToCantera::test_falloff PASSED

The conda environments are identical: https://www.diffchecker.com/AZUwVvpm/ but I would expect this since the RMS installation happens after displaying the conda environment.