feat(sema): integer explicit conversions

[mablr]

Closes #611

• Introduced integer explicit conversions
• Added tests

Follows Solidity 0.8.0+ rules regarding type conversions (see https://docs.soliditylang.org/en/latest/080-breaking-changes.html#new-restrictions)

[DaniPopes]

can you also check overflow fails? same line but with 9223372036854775808

[mablr]

That's right, mb. The check on int literal is not sufficient btw. I'll fix it to comply w/:

2. Explicit conversions between literals and an integer type T are only allowed if the literal lies between type(T).min and type(T).max. [...]

[mablr]

I think it's actually blocked for the moment (maybe solved by #566)

In fact we need to check the value and the sign of the literal to make this assertion : type(T).min <= literal_value < type(T).max.

So, mapping from/to Ty objects is not enough, as we need the value from LitKind::Number.

[DaniPopes]

would it be possible with TypeSize storing bits? that way min..=max for intX is X-1 bits and uintX is X bits

[mablr]

That's what i tried initially, but it seems to be to restrictive given the granularity of TypeSize (byte).

IntLiteral(positive) -> Uint: OK

(IntLiteral(false, size_from), Elementary(UInt(size_to))) => { if size_from.bits() <= size_to.bits() { Ok(()) } else { Result::Err(()) }},

IntLiteral(negative) -> Int: I guess OK (but can't test it without #566)

(IntLiteral(true, size_from), Elementary(Int(size_to))) => { if size_from.bits() <= size_to.bits() { Ok(()) } else { Result::Err(()) } },

IntLiteral(positive) -> Int: too restrictive

(IntLiteral(false, size_from), Elementary(Int(size_to))) => { if size_from.bits() < size_to.bits() { Ok(()) } else { Result::Err(()) } },

i.e int8 d = int8(0xF0); is rejected because size_from.bits() == 8 and size_to.bits() == 8 which doesn't reflect the value bounds (−2^(n−1) to 2^(n−1) − 1)

[DaniPopes]

right, what if typesize had bit granularity for literals? so that intX::MIN is typesize(X-1 : u16) instead of typesize(X/8 : u8)

[mablr]

what if typesize had bit granularity for literals?

That would be great!

Correct me if I’m wrong, i think to produce Ty IntLiteral objects, it will be necessary anyway to hold the sign in the ast/hir number (the value as u256 and boolean for the sign).

[dipanshuhappy]

what if typesize had bit granularity for literals?

That would be great!

Correct me if I’m wrong, i think to produce Ty IntLiteral objects, it will be necessary anyway to hold the sign in the ast/hir number (the value as u256 and boolean for the sign).

Yep it seems so, looking at the expression enum here
Lit only stores the numerical value, while sign is held by Unary

All these are translated in the typechecker to Ty::IntLiteral here

[onbjerg]

I'm going to mark this blocked by #650 (see above discussion).