WITHDRAWN: the RC4 underflow was a harness bug, not a compiler bug

Filed 2026-08-08 on the strength of

$ printf 'INCLUDE testje.for\nBYE\n' | ./uforth
ERROR: Stack underflow

INCLUDE in uforth is ( "filename" -- ) — it pops the filename off the stack. INCLUDE testje.for therefore pops an empty stack, and the underflow is the CORRECT answer. The proper spelling is "testje.for" INCLUDE, and with it:

testje.for       0 7 11 15 11      IDENTICAL to CPython
testjefixed.for  8A 6A D9 02 6A    IDENTICAL
testjefix2.for   DB 3D D9 CD AF    IDENTICAL
testjefix3.for   2D D6 65 30 F2    IDENTICAL

Why it looked real

CPython appeared to accept the bad form and print the cipher output, which made it look like a pxx-only failure. It did not: the comparison was run before the select.select fix, so the two sides differed in prompt handling as well, and I read one difference as evidence for another.

What it cost, and the lesson

Nothing was changed on its account — the real bug in the same area ([[bug-nilpy-sequence-repeat-with-a-variant-count-falls-through-to-arithmetic]]) was found by bisecting the FILE, which immediately produced a failing line that reproduced with the correct INCLUDE spelling too.

The lesson is the one this arc keeps teaching from the other direction: when a corpus fails, check that the harness speaks the corpus's language before concluding anything about the implementation under test. A wrong invocation and a real defect look identical from the outside.

Superseded by [[bug-nilpy-uforth-drv-suite-last-line-and-file-words]] for what IS still red in uforth's own suite.