wrong type inference

[dimdin]

Zig Version

0.12.0-dev.3381+7057bffc1

Steps to Reproduce and Observed Behavior

To reproduce run:

pub fn main() !void {
    const olength: u32 = 1;
    const precision: ?usize = null;
    const exponent: i32 = -1; 
    const req_bytes = 2 + @max(@abs(exponent) + olength, precision orelse 0);
    _ = req_bytes;
}

❯ zig run test.zig
test.zig:5:25: error: overflow of integer type 'u2' with value '4'
    const req_bytes = 2 + @max(@abs(exponent) + olength, precision orelse 0);
                      ~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

This is an actual library line:

zig/lib/std/fmt/ryu128.zig

Line 275 in e831313

2 + @max(@abs(f.exponent) + olength, precision orelse 0);

It happens only in comptime

Expected Behavior

result type must be u32 or usize instead of u2

[xdBronch]

since all the arguments to @max are known at comptime it returns the smallest fitting integer type that can represent its value, that being a u2 in this case
see related issues #18365 #16139 #14039

[dimdin]

It behaves differently in comptime and in runtime.
Code that is perfect in runtime cannot be used in comptime.
I came to this while trying to format a float at comptime, and zig float format currently is broken.

[xdBronch]

i understand the problem statement. i believe the fix here is to provide an intermediate type coercion in the line that errors e.g. 2 + @as(u32, @max(@abs(f.exponent) + olength, precision orelse 0))

[dimdin]

I agree. The line came from this merge: #19229