Hojun-Cho
993da52333
wwstage's nodeisstr (cgenutil) didn't recognize N_INDEX-of-[N]str. cgindex emitted only the ptr-half MOVQ when the result was used as a str arg (call, .len access, str streq), so the .len half read stack residue. Surfaced by worker-21 during #21 dev — pre-#21 slotsize=24B masked the read-side defect; post-#21 (16B stride) exposed it. cstage's typed-AST node_isstr handles this naturally; wwstage's untyped pattern walks the base ident's tnode shape. Added N_INDEX arm to nodeisstr: walk the indexed base's tnode through N_TARRAY / N_TSLICE / N_TPTR.lhs, return isstrtype on the element. Mirrors cgindex's own base-type walk byte-for-byte in shape so the two now agree on load-shape decisions. Not covered (separate bugs, separately filed): - N_UN(TK_STAR) of *str — cgun itself never loads .len into BX. - tuple `.1` of str — N_TTUPLE path has its own load shape. - alias-typed base (`type a = [N]str`) — N_TNAME isn't peeled; cgindex doesn't peel it either, so agreement holds. Outside #34 scope. Test 711: 3 new rows — barelet_index_call_arg (streq direct arg), nested_call_index_arg (f(g(argv[i])) — nested-call recursion), barelet_index_len_arg (sister regression-pin for cgindex element stride in bare-let context; pins a different code path that was already correct post-#21). The pre-existing wwstage `..findflag(SB)` symbol-mangling bug in getopttest wwstage build is filed as task #37, not in this commit's scope.