Improve Regression Test Reproducibility in CI

[harryswift01]

Summary

This PR improves the reliability and reproducibility of the regression test suite by running tests serially and reducing sources of ordering and numerical variability.

Changes

Regression execution:

• Removed pytest-xdist and its associated configuration so regression tests run serially within each system job.
• Replaced test-name filtering with the existing --system option.
• Replaced the custom pytest_generate_tests hook with standard pytest.mark.parametrize.

Subprocess configuration:

• Updated regression tests to run CodeEntropy using sys.executable.
• Set PYTHONHASHSEED and limited numerical libraries to a single thread within the regression subprocess.

Group indexing:

• Sorted group IDs before assigning accumulator indices so the mapping does not depend on dictionary insertion order.

Impact

• Regression tests now run in a more consistent environment without changing the underlying entropy calculations.
• Existing system-level GitHub Actions parallelism is retained.

[jimboid]

I can see instance level parallelism causing issues with regression testing. Removal of this unnecessary layer of parallel code since the performance improvements will yield a more reliable and reproducible regression test suite.

[harryswift01]

Summary

This PR improves the reliability and reproducibility of the regression test suite by running tests serially and reducing sources of ordering and numerical variability.

Changes

Regression execution:

• Removed pytest-xdist and its associated configuration so regression tests run serially within each system job.
• Replaced test-name filtering with the existing --system option.
• Replaced the custom pytest_generate_tests hook with standard pytest.mark.parametrize.

Subprocess configuration:

• Updated regression tests to run CodeEntropy using sys.executable.
• Set PYTHONHASHSEED and limited numerical libraries to a single thread within the regression subprocess.

Group indexing:

• Sorted group IDs before assigning accumulator indices so the mapping does not depend on dictionary insertion order.

Impact

• Regression tests now run in a more consistent environment without changing the underlying entropy calculations.
• Existing system-level GitHub Actions parallelism is retained.

[jimboid]

I can see instance level parallelism causing issues with regression testing. Removal of this unnecessary layer of parallel code since the performance improvements will yield a more reliable and reproducible regression test suite.

[harryswift01]

Summary

This PR improves the reliability and reproducibility of the regression test suite by running tests serially and reducing sources of ordering and numerical variability.

Changes

Regression execution:

• Removed pytest-xdist and its associated configuration so regression tests run serially within each system job.
• Replaced test-name filtering with the existing --system option.
• Replaced the custom pytest_generate_tests hook with standard pytest.mark.parametrize.

Subprocess configuration:

• Updated regression tests to run CodeEntropy using sys.executable.
• Set PYTHONHASHSEED and limited numerical libraries to a single thread within the regression subprocess.

Group indexing:

• Sorted group IDs before assigning accumulator indices so the mapping does not depend on dictionary insertion order.

Impact

• Regression tests now run in a more consistent environment without changing the underlying entropy calculations.
• Existing system-level GitHub Actions parallelism is retained.

[jimboid]

I can see instance level parallelism causing issues with regression testing. Removal of this unnecessary layer of parallel code since the performance improvements will yield a more reliable and reproducible regression test suite.

[harryswift01]

Summary

This PR improves the reliability and reproducibility of the regression test suite by running tests serially and reducing sources of ordering and numerical variability.

Changes

Regression execution:

• Removed pytest-xdist and its associated configuration so regression tests run serially within each system job.
• Replaced test-name filtering with the existing --system option.
• Replaced the custom pytest_generate_tests hook with standard pytest.mark.parametrize.

Subprocess configuration:

• Updated regression tests to run CodeEntropy using sys.executable.
• Set PYTHONHASHSEED and limited numerical libraries to a single thread within the regression subprocess.

Group indexing:

• Sorted group IDs before assigning accumulator indices so the mapping does not depend on dictionary insertion order.

Impact

• Regression tests now run in a more consistent environment without changing the underlying entropy calculations.
• Existing system-level GitHub Actions parallelism is retained.

[jimboid]

I can see instance level parallelism causing issues with regression testing. Removal of this unnecessary layer of parallel code since the performance improvements will yield a more reliable and reproducible regression test suite.

[harryswift01]

Summary

This PR improves the reliability and reproducibility of the regression test suite by running tests serially and reducing sources of ordering and numerical variability.

Changes

Regression execution:

• Removed pytest-xdist and its associated configuration so regression tests run serially within each system job.
• Replaced test-name filtering with the existing --system option.
• Replaced the custom pytest_generate_tests hook with standard pytest.mark.parametrize.

Subprocess configuration:

• Updated regression tests to run CodeEntropy using sys.executable.
• Set PYTHONHASHSEED and limited numerical libraries to a single thread within the regression subprocess.

Group indexing:

• Sorted group IDs before assigning accumulator indices so the mapping does not depend on dictionary insertion order.

Impact

• Regression tests now run in a more consistent environment without changing the underlying entropy calculations.
• Existing system-level GitHub Actions parallelism is retained.

[jimboid]

I can see instance level parallelism causing issues with regression testing. Removal of this unnecessary layer of parallel code since the performance improvements will yield a more reliable and reproducible regression test suite.

[harryswift01]

Summary

This PR improves the reliability and reproducibility of the regression test suite by running tests serially and reducing sources of ordering and numerical variability.

Changes

Regression execution:

• Removed pytest-xdist and its associated configuration so regression tests run serially within each system job.
• Replaced test-name filtering with the existing --system option.
• Replaced the custom pytest_generate_tests hook with standard pytest.mark.parametrize.

Subprocess configuration:

• Updated regression tests to run CodeEntropy using sys.executable.
• Set PYTHONHASHSEED and limited numerical libraries to a single thread within the regression subprocess.

Group indexing:

• Sorted group IDs before assigning accumulator indices so the mapping does not depend on dictionary insertion order.

Impact

• Regression tests now run in a more consistent environment without changing the underlying entropy calculations.
• Existing system-level GitHub Actions parallelism is retained.

[jimboid]

I can see instance level parallelism causing issues with regression testing. Removal of this unnecessary layer of parallel code since the performance improvements will yield a more reliable and reproducible regression test suite.

[harryswift01]

Summary

This PR improves the reliability and reproducibility of the regression test suite by running tests serially and reducing sources of ordering and numerical variability.

Changes

Regression execution:

• Removed pytest-xdist and its associated configuration so regression tests run serially within each system job.
• Replaced test-name filtering with the existing --system option.
• Replaced the custom pytest_generate_tests hook with standard pytest.mark.parametrize.

Subprocess configuration:

• Updated regression tests to run CodeEntropy using sys.executable.
• Set PYTHONHASHSEED and limited numerical libraries to a single thread within the regression subprocess.

Group indexing:

• Sorted group IDs before assigning accumulator indices so the mapping does not depend on dictionary insertion order.

Impact

• Regression tests now run in a more consistent environment without changing the underlying entropy calculations.
• Existing system-level GitHub Actions parallelism is retained.

[jimboid]

I can see instance level parallelism causing issues with regression testing. Removal of this unnecessary layer of parallel code since the performance improvements will yield a more reliable and reproducible regression test suite.

[harryswift01]

Summary

This PR improves the reliability and reproducibility of the regression test suite by running tests serially and reducing sources of ordering and numerical variability.

Changes

Regression execution:

• Removed pytest-xdist and its associated configuration so regression tests run serially within each system job.
• Replaced test-name filtering with the existing --system option.
• Replaced the custom pytest_generate_tests hook with standard pytest.mark.parametrize.

Subprocess configuration:

• Updated regression tests to run CodeEntropy using sys.executable.
• Set PYTHONHASHSEED and limited numerical libraries to a single thread within the regression subprocess.

Group indexing:

• Sorted group IDs before assigning accumulator indices so the mapping does not depend on dictionary insertion order.

Impact

• Regression tests now run in a more consistent environment without changing the underlying entropy calculations.
• Existing system-level GitHub Actions parallelism is retained.

[jimboid]

I can see instance level parallelism causing issues with regression testing. Removal of this unnecessary layer of parallel code since the performance improvements will yield a more reliable and reproducible regression test suite.