Fix Issue 13230: std.variant.Variant Uses Deprecated .min Property in opArithmetic When T is a Floating Point Type

[MetaLang]

https://issues.dlang.org/show_bug.cgi?id=13230

[mihails-strasuns]

.min_normal is something very different from .min and it does not seem to be applicable here. I think proper fix is to harden static if conditions to avoid matching T : int for floating types - there is a dedicated branch for those after all.

[MetaLang]

The compiler is telling me to use min_normal in place of min. If this isn't the correct action in all cases, then that deprecation message needs to be changed.

[mihails-strasuns]

I believe it is wrong deprecation message. More appropriate one may be .min value is not applicable to floating point values, please refactor your code if you expect it to be similar to int.min or use .min_normal if you actually need minimum positive normalized value.

I am not very educated in this topic though.

[MetaLang]

Hold on a moment, why is typeof(float.max) returning true for implicit conversion to integral types in the first place?

[mihails-strasuns]

That is exactly the strange part. All there print false when I try it locally:

pragma(msg, is(typeof(float.max) : int));
pragma(msg, is(typeof(float.max) : ulong));
pragma(msg, is(typeof(float.max) : long));

[MetaLang]

Is the deprecation message printed when you try to run the following code locally?

import std.variant;

void main()
{
    Variant v = 1.0f;
    //Deprecation: min property is deprecated, use min_normal instead
    v += 1.0f;
}

[yebblies]

.min_normal is something very different from .min and it does not seem to be applicable here.

.min for floating point types is exactly the same thing as .min_normal in the compiler.

[MetaLang]

.min for floating point types is exactly the same thing as .min_normal in the compiler.

Then this pull shouldn't change any behaviour other than silencing that deprecation message.

[mihails-strasuns]

Then this pull shouldn't change any behaviour other than silencing that deprecation message

This deprecation message highlights a bug. It effectively checks if T can represent negative numbers - but for floating point T.min will always be positive value.

btw I can see deprecation printed locally, will check how it gets there a bit later.

[yebblies]

The message is being printed because even though the && is short-circuited, the later expressions still get semantic run on them. The correct solution is probably to use std.traits.isUnsigned (or similar) instead of just comparing .min. Alternatively, you could static if (isFloating...) the whole block.

[MetaLang]

The correct solution is probably to use std.traits.isUnsigned...

Works for me.

[yebblies]

Auto-merge toggled on