[kixelated]

Problem

format!("{q:?}") of a VecDeque<T> refused with unsupported call to <VecDeque<T> as Debug>::fmt and a span (found by #303). Auditing the rest of std's collections showed that HashMap, HashSet, BTreeMap, and BTreeSet are refused the same way. Vec, BinaryHeap, slices, arrays, and Box<[T]> already translated.

Approach

• fmt.rs maps each collection's Debug to its iterator, as std's impls do: a deque to rt.fmtList(rt.dequeIter(q)), so items come from its head however the buffer wraps; a set to rt.fmtSet over its keys; a map to rt.fmtMap over its entries.
• rt.fmtSet is std's DebugSet, which is its DebugList in braces. rt.fmtMap writes each entry as a single set entry: key, ": ", then value, through one PadAdapter, as std's DebugMap does in pretty mode.
• A new fmt::collections case in the lang fixture runs differentially against native wasm32. It covers a deque driven by random push/pop at both ends; a deque wrapped by push_front and then popped from the back past the buffer's start; empty deques, maps, and sets; VecDeque<()>; nested deques; {:#?}, {:x?}, and width; elements with custom Debug (Hex) and Option<u8> elements; and B-tree maps of deques. To check that the new case is covered, I broke fmtMap's ": ": the differential failed with 212 differences.
• Hash collections are tested with at most one entry because std leaves their order unspecified. ts/README.md now says that a collection's Debug follows its iteration order.

Impact

• The translator's output: {:?} of VecDeque, HashMap, HashSet, BTreeMap, and BTreeSet now translates instead of being refused.
• Runtime: adds rt.fmtSet and rt.fmtMap.

Alternatives

• Defining rt.DebugSet/rt.DebugMap builders. Not done: no translated code calls f.debug_set() or f.debug_map() yet, so these are plain functions over DebugList.

Follow-ups

• std iterators' Debug: slice::Iter, vec::IntoIter, the B-tree and hash iterators, Chars, Bytes, and vec_deque::Iter, whose Debug writes its two ring-buffer slices (and VecDeque::as_slices, also refused). These are refused for a different reason: each writes its own wrapper text.
• LinkedList: the type itself is unsupported.

Choices made without asking: I fixed maps and sets in this PR because they are refused the same way, and recorded iterators and LinkedList as quests.

Review

Codex (quest review), gpt-6.1-sol, medium effort, round 1 on head ae2f61b7a9236402b0bba23f8f8a67b30602bfc7 against origin/main fba78a4a: no findings.

just ci passed locally on that head.

🤖 Generated with Claude Code

(written by Claude Opus 5.5)