improve error message for unable to evaluate constant expression in inline while

[oconnor0]

The following code works as I would expect.

pub fn inclusive_range(comptime Int: type, comptime lower: Int, comptime upper: Int) [upper-lower+1]Int {
  var values: [upper-lower+1]Int = undefined;
  var i: Int = lower;
  while (i <= upper) : (i += 1) {
    values[i - lower] = i;
  }
  return values;
}

pub fn main() void {
    for (inclusive_range(u64, 2, 72)) |i| {
        warn("i {}\n", i);
    }
}

However changing line 6 to inline while (i <= upper) : (i += 1) { produces the following error:

E:\Source\langdev\inclusive_range.zig:6:19: error: unable to evaluate constant expression
  inline while (i <= upper) : (i += 1) {
                  ^
E:\Source\langdev\inclusive_range.zig:13:25: note: called from here
    for (inclusive_range(u64, 2, 72)) |i| {
                        ^

2 thoughts:

1. Is this a bug in zig?
2. If not an error messages that involves the inline while vs while or whatever the underlying issue is would be greatly appreciated.

[andrewrk]

You need to make i a comptime var.

That said I think we can improve the error message.

[SimonN]

Zig's C++ codebase calls ir_resolve_const from 98 places. Each time, ir_resolve_const generates the error message itself, and the call site does not add its own error. It will be hard to change the error message for a single specific error. What should the high-level strategy be to improve this error message throughout the codebase?

Example: Last week, in usercode, I mistakenly wrote std.debug.warn(s) with a runtime string s, even though warn needs its first argument fmt to be comptime. The error again was "unable to evaluate constant expression".

• My first interpretation: s is constant, but cannot be evaluated.
• Desired interpretation: s is not constant.

The inline while example from November 2018 should get a very different error message than my example here; my example should get e.g.: "argument fmt of warn requires a comptime parameter, but s is not comptime".

[BarabasGitHub]

Ran into this today. It's very confusing.

    var i : u32 = 0;
    inline while(i < 10) : (i += 1) {
    }

will say:

error: unable to evaluate constant expression
    inline while(i < 10) : (i += 1) {
                   ^

And my first thought is: "What?! Why? If the expression is constant, then why can't you evaluate it? Should be easy, right?"

Better would be something like: "Runtime variable used in comptime expression."

[jsyrjala]

I ran into this error message with this bit of code:

const value: f32 = 1. + get_value();
_ = std.math.clamp(value, 0.0, 255.0);

Here the problem is that value is type f32 and constants comptime_float. Casting constants to f32 fixes it, but it was also confusing.

[jibal]

I believe that the error message actually means "expression cannot be evaluated at comptime" or something to that effect. The wording of the message says something completely different (and wrong).

[Vexu]

Now results in:

a.zig:4:21: error: unable to resolve comptime value
    inline while (i <= upper) : (i += 1) {
                  ~~^~~~~~~~
a.zig:4:21: note: condition in comptime branch must be comptime-known

Tested in:

$ git grep "condition in comptime branch must be comptime-known" test   
test/cases/compile_errors/branch_in_comptime_only_scope_uses_condbr_inline.zig:// :4:15: note: condition in comptime branch must be comptime-known
test/cases/compile_errors/branch_in_comptime_only_scope_uses_condbr_inline.zig:// :11:11: note: condition in comptime branch must be comptime-known
test/cases/compile_errors/condition_comptime_reason_explained.zig:// :8:9: note: condition in comptime branch must be comptime-known
test/cases/x86_64-linux/assert_function.8.zig:// :3:21: note: condition in comptime branch must be comptime-known
test/cases/x86_64-macos/assert_function.8.zig:// :5:21: note: condition in comptime branch must be comptime-known