uforth's FILE word set: the wrong INCLUDE, and a word missing in an included file

tests/_drv_file.fth is the only file left differing in uforth's driver suite (the other 10 are byte-identical). Both sides fail — at different points, for different reasons:

cpython: ...ERROR: filetest.fth:278: THROW -13 (line: '0 SI_INC !')
pxx:     ...ERROR: filetest.fth:217: INCLUDE expects a string filename on stack
                                     (line: '  INCLUDE required-helper1.fth')

Finding 1: the later INCLUDE redefinition does not win

uforth defines w_include TWICE, as two nested defs in two different enclosing functions:

• uforth.py:1707 — pops a string, else raises "INCLUDE expects a string filename on stack"; registered vm.define_word("INCLUDE", native=w_include);
• uforth.py:3041 — handles the stack form AND the ANS bare-name form via vm.next_token_strict(); registered later, with immediate=True.

pxx's error text is the FIRST one's, so the later registration did not replace it. CPython uses the second (it reads the bare name and includes the file).

Finding 2: 1+ is undefined inside an INCLUDEd file

Reached via the stack form, which both sides accept:

REQUIRE required-helper1.fth       pxx: THROW -13 at helper:3 ('1+')
S" required-helper1.fth" REQUIRED  cpython: Stack underflow at helper:3

CPython finds 1+ and fails on the empty stack (which is the test's business); pxx cannot find the word at all. 5 1+ . works fine at the REPL, so the dictionary lookup is failing only in the nested-inclusion context.

Isolated repros PASS — do not start from them

Both shapes were reduced and both match CPython:

• two same-named nested defs in two enclosing functions, registered into a dict under one key, second one winning — including with the native=/immediate= KEYWORD form and a @dataclass Word, i.e. uforth's exact spelling;
• 1+-style digit-leading word names, defined and called.

So the cause needs uforth's real state. Suggested route, given hand-instrumenting uforth produces a pxx binary that hangs (see [[bug-nilpy-uforth-dot-paren-prints-nothing]]): uforth's own --trace, diffed between the two runs, which is what cracked the .( case — look for the trace:call enter word=INCLUDE line and what body it names, and for how required-helper1.fth's source frame differs.

Gate

tests/_drv_file.fth byte-identical to the CPython run, the other 10 still identical, make test-uforth still PASS, plus the per-fix loop.

Log

• 2026-08-08 — resolved, commit 383f9c2f5.