Fix issue 9999,10560 - Deprecate the implicit int to bool conversion for integer literals

[LemonBoy]

As stated by @MartinNowak in #1942 the implicit conversion from int to bool should be removed and today, 4 years later.
A single PR to phobos [1] and one to DRuntime [2] are needed.

[1] dlang/phobos#5019
[2] dlang/druntime#1732

[dlang-bot]

Fix	Bugzilla	Description
✓	9999	Integer literal 0 and 1 should prefer integer type in overload resolution

[mathias-lang-sociomantic]

s/Removed/Deprecated/

[LemonBoy]

IMO the deprecation phase is long overdue heh
If you strongly feel there should be a deprecation phase please give me a timeline for the definitive removal so I can add it to the deprecation message

[mathias-lang-sociomantic]

IMO the deprecation phase is long overdue heh

Sadly this approach doesn't scale. The reason we have this policy about deprecation is to keep a sane ecosystem package. More than often, you find a library that will only compile with a specific version of DMD, because someone felt a deprecation was not necessary for this specific case.

The "fix your code" approach does not scale either: often libraries have dependencies, which might not be compatible with the same frontend. You might also be blocked from updating because of a compiler regression. In general, we should never force working code to be changed if there is a simple deprecation path, and there is one here.

If you strongly feel there should be a deprecation phase please give me a timeline for the definitive removal so I can add it to the deprecation message

We don't currently have timeline for removal for any of the other deprecations. But it will be a couple of releases for sure.

[LemonBoy]

Alright, you have a point, the only price to pay is having to show the deprecation warning twice but I don't think that's a dealbreaker here.

[UplinkCoder]

@LemonBoy do you have a compelling argument to deprecate this.
Which errors are avoided ?

[LemonBoy]

The two Bugzilla issues referenced in the title contain enough examples of how this (mis)feature gives you a WTF moment.

[MetaLang]

@UplinkCoder

import std.stdio;
void foo(bool b) { writeln("bool"); }
void foo(long l) { writeln("long"); }
void main()
{
    foo(0); //bool?!
    foo(1); //bool?!
    foo(2); //long, okay
}

[UplinkCoder]

We could fix that without deprecating the implicit cast.

[MartinNowak]

Mmh, definitely yes for making that an ambiguous overload, but not too convinced of deprecating bool(0), bool(1), bool(i32 & 1)...

[LemonBoy]

I see your point but I also think that conflating the integral types and bool is not such a wise idea for the very same reason there's a distinct bool type.
I don't have a strong opinion about this though.

[ntrel]

Changing just the overload rules was the original idea from bugzilla, but the initial pull:
#1942

got closed to "drop implicit int literal to bool conversion instead".

[JinShil]

Revived at #7310.