Add logic to decipher floating point implicit and explicit casts.

[maxhaton]

Logic is consistent with dmd with the exception that dmd seems to allow some casts to void which are not yet accounted for (if ever)

Unit tests have been added.

[deadalnix]

This is unfortunately doign 3 things at once, which is really not idea:

• float to float conversions.
• int to float.
• float to int.

This could go faster if the 2 were separated, as any problem with one wouldn't block the other two.

This also doesn't have any integration tests, which is a bit unfortunate.

[deadalnix]

Not really a problem with this patch, but that function could do with a better name. isFloatingPointType or something like that.

[deadalnix]

This bake int he assumption that the only thing that can convert to integral that isn't an integral is a char. This is the kind of assumption that tend to blow up on one's face later (for instance, we can immagine bool to be convertible to integral, but not being a char).

[deadalnix]

This remains incorrect, as isSigned is not defined for non integral types.

See implementation:

sdc/src/d/common/builtintype.d

Lines 114 to 117 in aa64387

	bool isSigned(BuiltinType t)
	in(isIntegral(t), "isSigned only applys to integral types") {
	return (t & 0x01) == 0;
	}

[maxhaton]

The D spec does consider bool to be an integral

[maxhaton]

Not entirely sure how to proceed

[deadalnix]

Not a huge deal, but I don't think a ternary is the best approach here. The code has a "if this do that, if this do that, if this do that, if nothing matches then return invalid" kind of structure.

This means you can add/change/remove the previous branches independently without the overall structure of the code changing, which is what you get with the if and then return invalid. With a ternary, the "default" option is commingled with the last pattern you match. Not a blocker, but this is the kind of details that makes for elegant subsequent diffs when things change.