NilPy: f-strings

Hangs off [[feature-nilpy-corpus-uforth]]. 42 sites in uforth.py, and they are spread through every diagnostic and every word-listing path, so the corpus cannot produce comparable OUTPUT without them even once it parses.

Shape

Lex f"..." into its literal spans and {expr} holes, then lower to the concatenation NilPy already has: "a" + str(x) + "b". str() is already wired for every scalar and for variants (VariantToStr), so the lowering needs no new runtime.

Scope for v1 — exactly what the corpus uses:

• plain {expr} holes, expressions being names, attribute access, calls and subscripts;
• {{ / }} escapes;
• format specs {x:d} / {x:04x} / {x!r} are NOT in v1 unless the census says otherwise — check before writing the lexer, because a spec changes the hole grammar rather than extending it.

Why it is worth doing before the Track A blockers clear

It is entirely frontend: no variant reads, no RTTI, no ABI. That makes it one of the few uforth-blocking features Track N can land alone, alongside [[feature-nilpy-bytes-and-slices]].

Landed

v1 landed 2026-07-19 in f8f1231e: plain holes + escapes, expanded in the source before lexing, holes lowered through pylib's pystr_of overloads. test/test_nilpy_fstrings.npy is in make test-nilpy, diffed against CPython.

Completed the same session: !r / !s conversions (2fd05bda) and format specs (35e15c0f). The whole uforth f-string census is covered — 64 plain holes, 18 !r, 3 specs.

Spec grammar is [<|>] [0] [width] [d|x|X|o|b|s], interpreted in pylib from the spec TEXT rather than compiled in the lexer, so one place decides what "05x" means and adding a spec later needs no frontend change. Anything outside the grammar halts naming itself.

Still open, if a corpus ever needs them: nested same-quote f-strings (errors cleanly), triple-quoted f-strings, and = debugging specs.

Gate

test-nilpy green with a .npy case diffed against CPython + --tier quick

• self-host byte-identical + make fpc-check.