repr() does not escape non-printables above U+007F

• Type: compat — Track N (Nil-Python frontend / RTL repr).
• Filed: 2026-08-30 by frankB, noticed while diffing the codecs UTF-8 decoder: the decoded VALUES matched CPython on all 81 cases and only their repr() rendering differed, which is how the two got separated.
• Measured at pin v395 (aa78a7faf63a).

Measured

So the escaping cutoff is U+0080 where CPython's is Unicode printability — it escapes categories Cc, Cf, Cs, Co, Cn and the non-space separators, at any code point.

test/test_nilpy_repr_escapes_non_printables passes and is not wrong; its cases are all below 0x80, which is the half that already works.

Why prio 15

CLAUDE.md's compat table: "our output formatting of a value differs → F if it's a float, else compat at low prio." This is formatting of a non-float. No program computes a different answer; repr is for humans, and the characters affected are ones that render as nothing useful either way.

Raise it if something starts comparing repr() output as data, or if a mimic_ differential needs to assert on strings in this range — the codecs one worked around it by comparing code points rather than repr, which is the better assertion anyway and is what a future differential should do.

What a fix costs, so nobody starts it expecting a one-liner

Doing this properly needs Unicode general categories — i.e. a unicodedata table — which is a real dependency and almost certainly not worth it at this prio.

The cheap 80% is the C1 range 0x80..0x9F, which is unambiguously non-printable, is what real text actually hits (mis-decoded latin-1, stray control bytes), and needs no table at all. 0xa0 (NBSP) is a defensible addition. Astral/unassigned escaping is the part that genuinely needs the tables; leaving it out is a smaller divergence than the C1 gap and can stay.

Whoever takes it: say in the commit which of those two tiers landed, so the ticket does not read as fully closed when only the cheap half is done.