mixing untyped (negative) floats and integers in arithmetic sometimes drops the sign

[thehowl]

I don't want to investigate precisely what happens right now; in any case, these are some equivalent expressions and their output in gno repl:

println(-3.0 * 1.0 / 2.0) // -1.5
println(-3.0 * 1.0 / 2) // -1.5
println(-3.0 * 1 / 2.0) // -1.5

println(-3 * 1.0 / 2.0) // 1.5
println(-3 * 1.0 / 2) // 1.5
println(-3 * 1 / 2.0) // 1.5

It seems that if a division happens between an untyped bigint and an untyped bigdec, the sign is dropped.

[waymobetta]

Do we mean unsigned here, not untyped?

[thehowl]

Nope, untyped.

In Go syntax, constant expressions (like "hello", or 2 + 4 -- consider them as any arithmetic/basic operation that doesn't reference a variable) are considered untyped.

That is to say that if we had the variables x, y, z := -3.0, 1.0, 2.0 (the .0 serves to make them "floats"), then x * y / z works fine, because at that point the values are typed into floats, and the expression is handled at the machine level, while an expression like -3 * 1 / 2.0 requires already some handling at the preprocessor level.

[waymobetta]

Right, understood. So since there is no declaration, this is where type inference can be incorrect. I was thinking that bigint and bigdec are already typed.