gen (Zig): signed and float % emits plain %, which Zig refuses to compile

[gHashTag]

Summary

t27c gen (Zig backend) emits a plain % for a remainder whose operands are signed integers (or floats). Zig refuses that at compile time, so any spec that takes % of an i32 parameter does not compile under zig test at all.

Found by the t27b differential test. Part of #5905.

Reproducer

module Sem;

fn inc(x: i32) -> i32 {
    return x + 1;
}

fn shr(x: i32, n: i32) -> i32 {
    return x >> n;
}

fn rem(a: i32, b: i32) -> i32 {
    return a % b;
}

test inc_max { assert_eq(inc(2147483647), 0 - 2147483647 - 1); }
test shr_big { assert_eq(shr(1024, 40), 0); }
test rem_neg { assert_eq(rem(0 - 7, 2), 0 - 1); }

endmodule

t27c gen sem.t27 > sem.zig && zig test sem.zig

Observed

error: remainder division with 'i32' and 'i32': signed integers and floats must use @rem or @mod

The whole file fails to compile, so inc_max and shr_big are never run either.

Expected

a % b on signed operands lowers to @rem(a, b): the truncated remainder, rem(-7, 2) == -1, which is what C, Rust and t27b compute. Unsigned % stays %.

/ already has this repair (W593 emits @divTrunc when an operand is known signed); % was never given the same rule. Zig refuses runtime float % with the same message, so floats need @rem too.