Spell a parameter name Python can take, ASCII or not

[CNSeniorious000]

paramName tested /^[A-Za-z_][A-Za-z0-9_]*$/. Python's identifier rule is XID, so a parameter named 路径 was folded into **kwargs — while the kernel, which validates the same name with inspect.Parameter, bound it as a real keyword parameter. The two halves already disagreed, and the block was the poorer one, for exactly the catalogues most likely to carry such a name.

 async def odd(
     *,
-    **kwargs: Any  # spell as dict keys: "路径", "Ω"
+    路径: str,
+    Ω: int = ...,
 )

def f(*, 路径: str) compiles, and inspect.Parameter('路径') is accepted — checked, not assumed.

The second condition, which is not optional

const isPyIdentifier = (name) => /^[\p{XID_Start}_]\p{XID_Continue}*$/u.test(name) && name.normalize('NFKC') === name

XID alone is not the rule the tokeniser applies. Source is NFKC-normalised before it is read, so a name that passes str.isidentifier() can still be un-spellable:

>>> 'fi'.isidentifier(), unicodedata.normalize('NFKC', 'fi')
(True, 'fi')
>>> exec("def f(*, fi=1, **kw): return locals()", ns); ns['f'](fi=2)
{'fi': 2, 'kw': {}}
>>> ast.parse("def g(*, fi: int = 0): pass").body[0].args.kwonlyargs[0].arg
'fi'

Spelled, fi would send the tool a key called fi — one it never declared — and the call would fail as though the tool were broken. Same for the fullwidth ｘ, which binds x. Those stay folded; 路径 and Ω are NFKC-stable and get spelled.

Verification

node test/smoke.js passes, and the new fixture's fence goes through python3 -c compile(...) so the emitted names are checked as source, not just as strings. Both conditions were mutation-checked separately:

mutation	which assertion fails
back to /^[A-Za-z_][A-Za-z0-9_]*$/	an NFKC-stable non-ASCII name is spelled
XID kept, .normalize('NFKC') guard dropped	an NFKC-unstable one folds even though Python calls it an identifier

Note

Scoped to the parameter path. isUsableName — tool names, imports, module names — stays ASCII: that one reaches import lines and dsh's class naming, so it deserves its own change rather than riding along. And the kernel's inspect.Parameter accepts fi today, which means <tool>? can still show a parameter no source can call; exotic enough that no live catalogue reaches it, but it is the same trap one level down.

[sourcery-ai]

Sorry @CNSeniorious000, you've used your own review budget of 250,000 diff characters for the last 7 days.

You can request another review in 4 days and 17 hours by commenting @sourcery-ai review. Upgrade to get a review now.