Yet another leak in the sinking ship of @safe

[Nick Treleaven via Digitalmars-d]

On Tuesday, 23 February 2016 at 11:04:31 UTC, Nick Treleaven 
wrote:
> Assuming we don't want to disallow slice.ptr in @safe code, 
> maybe we could have the compiler insert this code before 
> reading slice.ptr:
>
> version(D_NoBoundsChecks) else
>     if (slice.length == 0) throw new RangeError("Unsafe .ptr on 
> empty array");

Tweaking this a bit, a null .ptr dereference could arguably be 
considered safe, so we could only throw when empty and non-null:

version(D_NoBoundsChecks) {} else
	if (slice.length == 0 && slice.ptr)
		throw new RangeError("Unsafe .ptr on empty array");

This could still sometimes break existing @safe code that only 
wants to see if slice.ptr is null (and not dereference .ptr). If 
the above were implemented, doing that might need a @trusted 
wrapper:

@trusted bool isNull(T)(T[] slice){
	return slice.ptr is null;
}

Instead comparing slice.ptr with null could be recognised and 
safely allowed by the compiler, without the runtime check, so 
long as .ptr doesn't escape.