[kixelated]

Problem

format!("{it:?}") of std's iterators refused with unsupported call to <... as Debug>::fmt and a span (found by #312), while their collections' Debug translated. Each iterator writes its own text: Iter([1, 2]), Chars(['a']), Bytes(Copied { it: Iter([97]) }), a bare list of entries, or Range { iter: [(2, SetValZST)] }.

Reproduced on lang: 34 refusals of Debug across the iterators below.

Approach

• The runtime's iterators behind these types gain remaining(): the items each has left, by value, without taking them (a place-lending iterator's values). rt.fmtIter writes them as a list inside the wrappers std writes, rt.fmtChars writes Chars(...), and rt.fmtSetVal writes a BTreeSet entry as std's internal map entry, (key, SetValZST).
• fmt.rs maps each iterator's Debug by its std path to its wrapper and item type, as read from the pinned std source and confirmed natively:
  • Iter([..]): slice::Iter, binary_heap::Iter, btree_set::Iter; IterMut([..]): slice::IterMut; IntoIter([..]): vec, binary_heap, and vec_deque's; Drain([..]): vec::Drain; Drain { iter: Drain([..]) }: binary_heap::Drain.
  • Chars([..]) and Bytes(Copied { it: Iter([..]) }).
  • A bare list: btree_map's Iter, IterMut, Keys, Values, ValuesMut, Range, RangeMut, IntoIter, IntoKeys, IntoValues; hash_map's Iter, IterMut, Keys, Values, ValuesMut, IntoIter, IntoKeys, IntoValues, Drain; hash_set's Iter, IntoIter, Drain.
  • Range { iter: [(k, SetValZST), ..] } and IntoIter { iter: [..] }: btree_set::Range and IntoIter.
• An iterator lending &[u8; N] as rt.ByteSlices was an iterMap; it is now rt.iterLend, whose remaining() reads through to the items as the collection holds them, so a slice's or collection's iterator of byte arrays prints too. values_mut of copied values is now its own pick of HashIter and BTreeIter rather than an iterMap over iter_mut, so it has remaining() too.
• A hash iterator's remaining() walks a fresh iterator of the same map past what it took; the map cannot change while the iterator borrows or owns it.
• A new fmt::iters case in lang runs differentially against native wasm32: every iterator above consumed from the front by nth and from the back by next_back (B-tree ones included), empty, {:#?}, {:x?}, width, Chars and Bytes over one- to four-byte chars and escaped chars, byte-array elements and keys, and elements with custom Debug. Hash iterators hold at most one entry, consumed or not, since std leaves their order unspecified. Breaking Chars::remaining and the SetValZST text made the differential fail with 186 differences.

Kept refused, with their span:

• vec_deque::Iter and IterMut, and VecDeque::as_slices: they write the ring buffer's two halves, Iter([3], [1, 2]), and the runtime's deque splits elsewhere than std's (power-of-two growth that moves the head to 0, with_capacity ignored, item-by-item shifts in insert and remove). Recorded as A deque's two slices, after VecDeque capacity in std-rest.
• vec_deque::Drain, which writes its internal counts, Drain(drain_len, idx, tail_len, remaining): same quest.
• CharIndices: char_indices itself has no mapping, so nothing makes one.

Impact

• The translator's output: {:?} of the iterators above translates instead of being refused. Iterators lending byte arrays are made by rt.iterLend instead of rt.iterMap.
• Runtime: adds rt.fmtIter, rt.IterWrap, rt.fmtChars, rt.fmtSetVal, rt.iterLend, and Iter.remaining() on std's iterators that have it.

Alternatives

• Cloning the iterator, as std's Debug impls do: the runtime's iterators have no clone, and a JS Map iterator can't be copied, so each kind reads its own state instead.
• Unwrapping iterMap in fmtIter for byte arrays: iterLend keeps that read-through to the one wrapper it is right for.

Follow-ups

• A deque's two slices: as_slices, and Debug of a deque's Iter, IterMut, and Drain.
• Cloning std's iterators (from #319): noted there that remaining() reads what std's Debug reads through a clone, so clones may replace it.

Choices made without asking: the quest named the listed iterators; I also covered each sibling the same mapping refused (IterMut, IntoIter, Drain, Range, Into*, *Mut). The quest's plan said btree_set::Iter writes a bare list; std writes Iter([..]), which this follows.

Review

Codex (quest review), gpt-6.1-sol, medium effort, against origin/main d1a076ef:

• Round 1 on ac3946de: one finding, "Preserve bounded zip folds for borrowed iterators" (ts/src/lower/iter.rs:877). Refuted: it was #314's zip fold, on origin/main but not yet merged into that head, so the review saw it as reversed; this PR's only change to iter.rs is by_ref's iterMap to iterLend. Fixed by merging origin/main.
• Round 2 on 517d782da7479f7ef6001de254ef6712bfd7df3f: no findings.

just ci passed locally on 517d782d.

🤖 Generated with Claude Code

(written by Claude Opus 5.5)