[PRINCE2-AI]

Description

Fixes the two Windows failures left in #1460. Both changes are test-only.

1. test_expired_cache_entry_is_refetched (Windows, Python 3.12 and older)

The test patched ELECTRICITYMAPS_CACHE_TTL to 0 and expected the second call to refetch. The cache check is time.monotonic() - cached[0] <= ELECTRICITYMAPS_CACHE_TTL, and on Windows Python 3.12 time.monotonic() uses GetTickCount64() (15.6 ms resolution): two calls inside one tick give a difference of exactly 0, the entry still looks fresh, and the test observes one request instead of two.

The test now moves the module's clock past the TTL instead of shrinking the TTL, so it checks the same "entry older than TTL" contract without depending on clock resolution. The patch replaces the name time in electricitymaps_api's namespace, which leaves the real clock in place for requests and responses.

2. test_task_energy_with_live_update_interference (any machine with an NVIDIA driver)

TestCarbonTracker patched codecarbon.core.gpu.pynvml, but AllGPUDevices and NvidiaGPUDevice read codecarbon.core.gpu_nvidia.pynvml (codecarbon/core/gpu.py:35-38, codecarbon/core/gpu_nvidia.py:54). gpu.pynvml is only a backwards-compatible alias, so on a machine with a driver the test queried real GPU energy and overshot the expected total. The patch now targets the module that is actually read.

Group 1 of #1460 no longer applies: #1462 skips the RAPL tests off Linux.

Fixes #1460.

Verification

• Before, Windows 11 / Python 3.12 / master ab8627a1, -m "not integ_test", tests/test_viz_data.py excluded because dash is not installed: 871 passed, 1 failed — the failure was test_expired_cache_entry_is_refetched. It also fails on Python 3.11 here.
• After: tests/test_electricitymaps_cache.py tests/test_emissions_tracker.py → 51 passed.
• ruff format --check clean on both files. ruff check reports the same 5 pre-existing findings as master, none on a changed line.

I could not reproduce the GPU test locally: this machine has no NVIDIA GPU, so test_task_energy_with_live_update_interference passes either way here. That half follows hifarrer's reproduction in #1460, where patching codecarbon.core.gpu_nvidia.pynvml made it pass on a driver-equipped machine.

Types of changes

• Bug fix (non-breaking change which fixes an issue)

AI Usage Disclosure

• 🟠 AI-generated: prepared with an AI assistant under my direction. I ran the commands above on my own machine and reviewed every changed line. The GPU half is verified only through the reproduction linked in Unit tests on Windows: 13 failures besides the CLI ones in #1459 (one also on any machine with an NVIDIA GPU) #1460, as stated above.

Checklist

• My code follows the code style of this project.
• My change requires a change to the documentation.
• I have updated the documentation accordingly.
• I have read the docs/how-to/contributing.md document.
• I have added tests to cover my changes.
• All new and existing tests passed.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.