[Aschenbath]

Expose blocktime_ms in OpenMP information from threadpool_info(), controller info(), and the CLI. LLVM/Intel OpenMP values are read from kmp_get_blocktime on every query and converted to milliseconds, preserving fractional milliseconds. Runtimes without that symbol, including GNU OpenMP, report None (null in JSON).

The dynamic property and OpenMPController.get_blocktime() reflect the calling thread's current value after kmp_set_blocktime, including when a controller is reused. The unit is cached per loaded runtime so later environment changes do not change the conversion.

Unit-detection limitation: the common LLVM/Intel kmp API does not export a unit getter. Conversion therefore uses a valid KMP_BLOCKTIME startup setting, which must remain available until the runtime is first inspected. Microsecond settings require a runtime that supports them. Changing units through kmp_set_defaults is unsupported. These conditions are documented in the README and API docstrings; this implementation does not provide unconditional native unit introspection.

Validation on this revision:

• Fresh Python subprocesses load real OpenMP with OMP_DISPLAY_ENV=VERBOSE and KMP_SETTINGS=1, parse the effective runtime setting, and compare it with threadpool_info(). Cases cover the default, 37, 37ms, 35001us, and 35 us, then change the environment and call kmp_set_blocktime to check cached units with a fresh value. The two microsecond cases failed before conversion on Intel OpenMP.
• Windows, CPython 3.12.12, Intel OpenMP 2026.2.0: focused OpenMP tests passed; full suite 64 passed, 52 skipped, 2 failed. Both failures are test_get_original_num_threads warning expectations with this NumPy/SciPy setup, and both also fail against unchanged master 13b1a129901f353ee4820a3fb7e9a6f7d5fcf4f4 in the same environment.
• Native glibc Linux, CPython 3.13.18, LLVM OpenMP 19.1.7, NumPy/SciPy and compiled GNU OpenMP helpers: full suite 83 passed, 35 skipped with PYTHON_CPU_COUNT=4. The initial unrestricted run reached 82 passed and 35 skipped but timed out in the existing 10-second controller concurrency stress test, which creates four threads per reported CPU. No Linux baseline run is claimed.
• All repository pre-commit hooks passed on an ext4 copy with tracked Git file modes, and all 43 tracked files matched the working tree bytes. git diff --check passed.

macOS and hosted CI results are not claimed here. AI assistance was used for investigation, implementation, testing, and this description.

Closes #270.

[ogrisel]

Thanks for the PR. Here is a first pass of feedback.

[ogrisel]

I would rather always report "ms" to avoid any ambiguity on how to correctly leverage this value. Is it possible to programmatically inspect the runtime and do the conversion in get_blocktime itself so that it always returns ms values?

[ogrisel]

We could even rename 'blocktime' to 'blocktime_ms' if we ensure that we always report this value in miliseconds.

[ogrisel]

I think we should include a test that imports openmp in a Python subprocess and the verbose env variable set and then check that the value reported by threadpool_info is consistent with the value parsed from stdout (including when manually set to non-ms units).