nx_iterator.nx

nx_iterator.nx -- composable iteration; kills the nested-loops bug class. User cardinal 2026-05-15: "how many loops within loops type bullshit do we create that then caused memory leaks and infinite bloat etc, isnt it better just to build the system so its bulletproof there where a loop can call another loop like a function but with better structure i feel like javascript c etc have all created a shit paradigm not a usability paradagim." THE STRUCTURAL FIX: first-class iterators. A pipeline of N transformations expresses as ONE while loop at the consumption end instead of N nested loops. Zero raw nesting; zero per-iteration allocations; clear ownership. Example transformation -- 5 stages, 1 loop: let it: *Iterator = nx_iter_range(0, 100) it = nx_iter_filter(it, NX_FILTER_EVEN, 0) it = nx_iter_map(it, NX_MAP_SQUARE, 0) it = nx_iter_take(it, 5) let v: *nx_int = (sys_mmap(NX_SIZEOF_NX_INT)) as *nx_int while nx_iter_next(it, v) == 1 { // v[0] = next squared even number, up to 5 of them } Same logic in nested-loop style would be 4 deep with per-level state, easy to leak. Iterator style: linear, audit-able, the substrate's Captain Moroni doctrine applied to iteration. NO-CLOSURE-NEEDED DESIGN: filter / map use the same sealed-enum predicate/op kinds as nx_array_ops.nx (NX_FILTER_* and NX_MAP_*). Substrate dispatches internally. ~80% coverage; the remaining 20% custom predicates wait for closure support (queued nxc2 work). CONSUMERS use NX_REDUCE_* op kinds from nx_array_ops.nx (SUM / PRODUCT / MIN / MAX / COUNT / AND / OR / XOR) for fold semantics. MEMORY: each iterator is a fixed-size struct (8 nx_int fields + 1 array pointer + 2 inner pointers = ~88 bytes). Per-pipeline allocation is O(stages) and bounded; no per-iteration allocations.