Repository navigation
t27b: @intCast with a result type (gen_fuzz, lane 4) #7412
Description
Part of #6063 (t27b lane 4).
specs/port/trinity/src/tri/gen_fuzz.t27 is blocked on ExprCall(@intCast). gen-zig prints @intCast(x) exactly as written, and the reference (t27c test-report) passes 8/8 with 0 vacuous tests. This is a backend gap, not a spec defect: t27b refuses every @intCast, and 36 corpus specs use it.
What Zig does, which t27b should match:
- The result type comes from the context:
@as(T, ..), a typed binding, a parameter, or areturn. With no result type it is a compile error. - The value is kept exactly. A lossless conversion widens. Any other conversion is checked: a value outside
Tis a safety panic in Debug, so the test fails. This includes unsigned to unsigned, unlikeas, which truncates there. - A literal operand that does not fit
Tis a compile error.
The slice:
- A conformance spec
specs/tri/t27b/conformance/int_cast.t27, whose names are unique in the corpus. It runs first undert27c test-report. @intCastwith an integer result type lowers to a checked cast.cli/t27b/tests/intcast.rscovers the trap cases, and each one was checked against the reference first.- gen_fuzz moves blocked -> pass in the ledger, together with the conformance row.
Done when gen_fuzz passes under t27b with the reference's verdicts, the corpus shows 0 mismatch and 0 jit_interp_mismatch, and no other spec changes verdict in a way the reference disagrees with.
Boundary
specs/port/trinity/src/tri/gen_fuzz.t27specs/tri/t27b/conformance/int_cast.t27cli/t27b/tests/intcast.rs
Activity
ledger: server_main.t27 passes once @intcast lowers; first row, pass (C…
gHashTag commentedon Oct 7, 2026
Foreign-line budget (owner rule #7371, merged in #7399): PR #7417 is over it, so it is now a draft with auto-merge off.
Master's
check_budget()run ongit diff --numstat --no-renames origin/master...claude/t27b-lane4-intcast:cli/t27b/src/lower.rscli/t27b/tests/intcast.rsThe
owner-approved-foreignlabel does not lift this cap, and splitting the work into smaller PRs to get under it would dodge the rule, so I am not doing that. The@intCastlowering (range check, literal fit, widen/wrap) has to move into.t27and come out of gen-rust. The trap cases now intests/intcast.rsneed a home that is not hand-written Rust. The conformance specspecs/tri/t27b/conformance/int_cast.t27and the ledger rows (gen_fuzz, server_main) stay valid for that rework.Part of #6063. PR: #7417.
gHashTag commentedon Oct 7, 2026
Reopened 2026-10-07: the bulk close of issues without a
## Boundarysection was reverted by the owner. Nothing to do here; this comment replaces the earlier close notice.t27b ledger: batch bless from a full lab run on master (Closes #7432) (…
spec(policy): drop a redundant @intcast in added_lines so t27b compil…
t27b: the Mach-O object layout and bytes are a t27 spec (Closes #7526) (
ledger: gen_fuzz and server_main pass once @intcast lowers (Closes #7412
t27b: @intcast with an integer result type lowers from a t27 plan (Cl…