wcc+w6c+w6c_ww: delete() range form delete(xs[lo:hi]) (fold-5a P2)

Hare's delete also takes a slicing place (harec check.c:1981-2027
EXPR_SLICE; Hare spells it delete(xs[i..j])): remove [lo, hi) — shift
[hi..len) down count = hi-lo strides, len -= count, cap unchanged; lo
defaults 0, hi defaults len, so delete(xs[:]) clears the slice with
storage retained. Checker accepts N_SLICE next to N_INDEX (object must
chase to a slice, harec :2024); the old range-unimplemented reject and
its #35 cite drop.

Lowering (both stages, converged byte-identical by construction) is the
single-element arm's same-slice whole-stride word-copy loop with a
DYNAMIC src offset (count*esz via a src register) instead of the
constant one-stride. Base shapes: local slice ident, deref-of-local,
plus NEW indexed local-slice base xs[g][lo:hi] — the fold-5a consumer
shape (regex.ha:333 delete(jump_idxs[group_level][..]); outer stride
off the type table). Bounds stay implicit, inheriting the documented
single-element posture (no index checks anywhere in cgen). Operands
evaluate left-to-right, exactly once, before the shift (harec order);
only the header ADDRESS is taken before operand eval, so a bound
expression's writes through the slice land before the copy.

test/809: 64 fixtures — full/explicit/re-clear/head/mid/tail/empty
a:a/end-boundary len:len/explicit 0:0 on a never-appended (nil-ptr)
slice, single-vs-range equivalence, cap preservation, esz 1/2/4/8/16/24
copy tails against the dynamic src, operand order-of-eval (lo/hi CALLs
fire once each, in order) + aliasing-visibility pins, the EXACT
[][]size regex consumer shape, deref base, 2 reject rows w/ diagnostic
text; every accept row cs==ww asm byte-id. test/804: reject_range row
retired (form now accepted), reject_nonindex text follows the widened
message.
This commit is contained in:
2026-06-04 23:18:38 +09:00
parent 60e61315bc
commit 1bcf2726cf
9 changed files with 1505 additions and 98 deletions

View File

@@ -1479,12 +1479,14 @@ cexpr(Checker *c, Node *n)
n->lhs->type = ty_err;
return n->type;
}
/* delete(xs[i]) — single-element slice removal, the
* delete-half of #35 (insert() is the twin arm below). harec
* ref/harec/src/check.c:1981-2027 also accepts the range
* form delete(xs[i..j]) (EXPR_SLICE) — out of scope here
* (regex fold-2b's consumers are all single-element);
* loud-rejected below, the range form stays filed on #35. */
/* delete(xs[i]) / delete(xs[lo:hi]) — slice removal, the
* delete-half of #35 (insert() is the twin arm below).
* harec ref/harec/src/check.c:1981-2027 accepts both an
* indexing place (EXPR_ACCESS/ACCESS_INDEX) and a slicing
* place (EXPR_SLICE — Hare spells it delete(xs[i..j]));
* either way the OBJECT must be a slice. The range form
* is the fold-5a prereq P2 (regex.ha:333
* delete(jump_idxs[group_level][..])). */
if (n->lhs && n->lhs->kind == N_IDENT &&
n->lhs->str && strcmp(n->lhs->str, "delete") == 0) {
Node *d = n->list;
@@ -1492,16 +1494,17 @@ cexpr(Checker *c, Node *n)
err(c, n->pos, "delete: takes exactly one argument");
if (d != NULL) {
(void)cexpr(c, d);
if (d->kind == N_SLICE)
err(c, n->pos, "delete: range form delete(xs[i..j]) unimplemented (task #35)");
else if (d->kind != N_INDEX)
err(c, n->pos, "delete: operand must be an indexing expression xs[i]");
/* harec check.c:2016's reject; wording
* adapted to ww's delete: prefix. */
if (d->kind != N_INDEX && d->kind != N_SLICE)
err(c, n->pos, "delete: operand must be an indexing or slicing expression");
else {
Type *bt = d->lhs ? d->lhs->type : NULL;
while (bt && bt->kind == TY_NAMED)
bt = bt->under;
/* harec check.c:2024 wording; a
* fixed-size [N]T base lands here. */
* fixed-size [N]T base and a str
* base land here. */
if (bt == NULL || bt->kind != TY_SLICE)
err(c, n->pos, "delete must operate on a slice");
}