[espetro]

Closes #586

What changed

• Adds tests/test_stdlib_call_table.py, a table-driven differential sweep over the complete fixed-signature stdlib call table: 153 forms (18 MATH_FUNCTIONS, 23 WAVE_FUNCTIONS, 104 REST_FIXED, 8 SPECIAL_FUNCTIONS).
• The forms are batched into 12 tiny programs. Each program compiles once with --emit bin --call run, runs as a native binary, and runs the same source under CPython 3.14 (sys.executable). Comparison is stdout line-by-line plus exit status. Each probe emits exactly one stdout line, so a mismatch maps back to its table row. platform.python_version is compared on the 3.14 line rather than the patch, because the binary pins its declared target while the CI oracle can run a different 3.14 patch.
• A coverage test asserts that the table inventory, the probe set and the form count (153) agree. A new table row without a probe fails the test instead of being skipped silently.
• Three rows diverge and are xfail with their tracking issue: fix(unicodedata): mirrored returns bool where CPython 3.14 returns int #712 unicodedata.mirrored returns bool instead of int, fix(platform): platform.platform output differs from CPython 3.14 #713 platform.platform output shape, fix(posixpath): realpath raises on a missing target where CPython returns a path #714 os.path.realpath on a missing target raises RuntimeError instead of resolving lexically.
• Those are bugs against the parity contract, so they are filed as issues rather than added to docs/reference/divergences.md. The xfail references the issue, and the batch passes once the row is fixed.

Test output

uv run pytest tests/test_stdlib_call_table.py -n 2 -q --no-header -rxX:

XFAIL tests/test_stdlib_call_table.py::test_call_table_group[host] - host: recorded divergence: issue #713 (platform.platform output) (platform.platform: native='Darwin-25.6.0-arm64' cpython='macOS-26.6.2-arm64-arm-64bit-Mach-O')
XFAIL tests/test_stdlib_call_table.py::test_call_table_group[paths_divergent] - paths_divergent: recorded divergence, exit status native=1 cpython=0: issue #714 (os.path.realpath on a missing target)
XFAIL tests/test_stdlib_call_table.py::test_call_table_group[text] - text: recorded divergence: issue #712 (unicodedata.mirrored return kind) (unicodedata.mirrored: native='True' cpython='1')
10 passed, 3 xfailed in 28.50s

The nine passing batches cover 119 forms; the three xfail batches execute 34 more forms, with the three divergent rows named above. Total wall time stays under the 90 second budget with -n 2.

CI

The first run (ded3134) passed ruff and compat2 and test shards 1/3 and 2/3. Shard 3/3 failed because platform.python_version() prints the pinned target patch (3.14.9) while that runner's oracle is 3.14.7. be75ba9 compares the 3.14. line instead. In the follow-up run ruff and compat2 passed; the test shards were still queued when this body was updated, so the final CI result is not claimed here.

$ uv run ruff check
All checks passed!
$ uv run ruff format --check
809 files already formatted

Not done

• The three divergences are reported, not fixed (fix(unicodedata): mirrored returns bool where CPython 3.14 returns int #712, fix(platform): platform.platform output differs from CPython 3.14 #713, fix(posixpath): realpath raises on a missing target where CPython returns a path #714), to keep this PR to the test sweep.
• One out-of-scope observation has no issue because the request capped new issues at three: a module-level annotated global initialized from a call (d: str = tempfile.mkdtemp()) is re-evaluated on every read, so two prints of d differ. The sweep runs probes inside run() and does not depend on it.

[cloudflare-workers-and-pages]

Deploying pyscriptc with    Cloudflare Pages

Latest commit:	c6bf041
Status:	🚫  Build failed.

View logs