t27b: a typed list local of numeric literals that the reference prints as a tuple (formats.t27:628) #8175

Refs #6063

Boundary

• In scope: t27b lowering of a local declared with an explicit array type whose elements are all bare numeric literals, for example specs/numeric/formats.t27:628, const values = [3]f32{ -1.0, 0.0, 1.0 };. t27b blocks it as ExprArrayLiteral ("array literal with no result type"). The plan lives in specs/tri/t27b/tuple_local_plan.t27 (t27b: an untyped list local that the code indexes is a Zig tuple: constant indices, and run-time ones only at compile time #8050 / t27b: an untyped list local that the code indexes lowers as a tuple, from a t27 plan (Closes #8050) #8055) and is extended there, gen-rust'd.
• Out of scope: std.math.inf in the same file (t27b: std.math.inf, std.math.log with base e, std.math.log1p and an f64 @round, from libm_plan.t27 (Closes #7923) #7925); any array literal whose elements are not literals.

User Scenarios

• An author writes a typed list of float constants inside an invariant and loops over it. The reference prints it as .{ -1.0, 0.0, 1.0 }, a tuple of comptime_float. t27b should either run it with the same result or refuse it with a named reason, and never guess.

Requirements

• The reference drops the declared element type, so t27b may lower the local as [N]T only when every element literal is exactly representable in T. Then reading an element as T and reading it as comptime_float give the same value. Any element that is not exactly representable stays refused with its own reason.
• t27b: an untyped list local that the code indexes lowers as a tuple, from a t27 plan (Closes #8050) #8055's plan currently answers NOT_MINE for bare comptime_int / comptime_float elements. This issue changes that answer only for the exactly-representable case.
• A conformance spec under specs/tri/t27b/conformance/ covers the representable case (-1.0, 0.0, 1.0, 0.5) and a refused non-representable case (0.1 in f16 or f32). It passes t27c test-report, and t27b passes it with runtime asserts.

Success Criteria

• t27c test-report on the new conformance spec is all pass with 0 vacuous; t27b test --check gives the same count with runtime asserts > 0.
• Assert-flip mutants of the new spec are all killed.
• The lab receipt compare against master is IMPROVED_ONLY, EQUIVALENT or NEUTRAL, with 0 mismatch.
• formats.t27's ExprArrayLiteral row is gone from the ledger (the file may still block on t27b: std.math.inf, std.math.log with base e, std.math.log1p and an f64 @round, from libm_plan.t27 (Closes #7923) #7925).

Requirements

FR-001: t27b MUST lower a typed list local declared with an explicit array type whose elements are all bare numeric literals as [N]T only when every element literal is exactly representable in T; reading an element as T and as comptime_float MUST give the same value.

FR-002: Any element that is not exactly representable in the declared type T MUST be refused with its own named reason.

FR-003: The plan in specs/tri/t27b/tuple_local_plan.t27 MUST answer the exactly-representable case for bare comptime_int / comptime_float elements instead of NOT_MINE.

FR-004: A conformance spec under specs/tri/t27b/conformance/ MUST cover the representable case (-1.0, 0.0, 1.0, 0.5) and a refused non-representable case (0.1 in f16 or f32); it MUST pass t27c test-report, and t27b MUST pass it with runtime asserts.

Success Criteria

• t27c test-report on the new conformance spec exits 0 with all tests passing and 0 vacuous; t27b test --check on the same spec exits 0 with the same test count and runtime asserts > 0.
• Assert-flip mutants of the new conformance spec are all killed (exit 0 on mutation run).
• Lab receipt compare against master yields IMPROVED_ONLY, EQUIVALENT, or NEUTRAL with 0 mismatch.
• The ExprArrayLiteral row for specs/numeric/formats.t27 is absent from the ledger.