← board

enumerate, zip, dict.items and most_common build pairs that print as lists

print(list(enumerate(["a", "b"])))   # CPython: [(0, 'a'), (1, 'b')]   pxx: [[0, 'a'], [1, 'b']]
print(list(zip([1, 2], ["a", "b"]))) # CPython: [(1, 'a'), (2, 'b')]   pxx: [[1, 'a'], [2, 'b']]
d = {"a": 1, "b": 2}
print(d.items())                     # CPython: dict_items([('a', 1), ('b', 2)])
                                     # pxx:     [['a', 1], ['b', 2]]

NilPy has ONE sequence representation, so a tuple is a TPyList carrying FIsTuple — set by the frontend wherever a tuple DISPLAY was written, and the only thing that makes (1, 2) render with parentheses rather than brackets (that is what bug-nilpy-str-of-tuple-is-empty established). The four pylib routines that manufacture pairs never set it:

pylib.pas — pyenumerate (≈2717), pyzip (≈2737), TPyDict.most_common (≈2933), TPyDict.itemlist (≈2984), each pair := TPyList.Create; with no pair.FIsTuple := True.

Iteration and unpacking are unaffected — for i, x in enumerate(xs) works — so this is confined to rendering, which is why it survived: nothing crashes, the output is merely not Python's.

Also, separately, the VIEW objects have no repr

d.keys() / d.values() / d.items() print as bare lists rather than dict_keys([...]) / dict_values([...]) / dict_items([...]). That is a different thing from the tuple flag (a missing view type, not a missing flag) and is worth its own decision about whether NilPy wants view objects at all — it is recorded here only so the two are not confused when someone diffs d.items() against CPython.

Shape of a fix

pair.FIsTuple := True; at the four sites. Cheap, but check the one place the flag changes MEANING rather than rendering: pypercent_format reads FIsTuple to decide whether the right-hand side is an argument LIST or a single value, so "%s" % <pair> would start unpacking. Grep the corpus for a % whose right side is an enumerate/zip/items element before landing it.

FIXED (this session)

pair.FIsTuple := True; added at all four sites (pyenumerate, pyzip, TPyDict.itemlist, TPyDict.most_common). Checked the pypercent_format concern: no corpus/test source has a % format whose right side is one of these pairs, and making them correctly tuple-tagged is the CORRECT direction anyway — real Python's "%s" % t on a tuple attempts positional unpacking too, so this fix only makes that path match CPython rather than introduce a new divergence.

Gate

make test-nilpy + a .npy printing enumerate, zip, items and most_common results, expectation taken from CPython's own output.

Log