FEAT: switch pair! from integer! to float32!

[bitbegin]

No description provided.

[qtxie]

We still need to support float pair! in the compiler lexer.

[hiiamboris]

Just curious here :)
Have you considered using a pair of 32-bit integers instead of float-32?
IMO, range from -2147483.648 to 2147483.647 should cover every possible graphics need, even with all the intermediary computation results.
It's just that using floats you open a pandora's box of NaNs, INFs, rounding, and inexact arithmetic.
Is there a reason why float should be preferred to integer in pair's case?

[qtxie]

Have you considered using a pair of 32-bit integers instead of float-32?

It is 32-bit integers in current implementation.

It's just that using floats you open a pandora's box of NaNs, INFs, rounding, and inexact arithmetic.

You have those problems too using integers. Even worse, using integers you lost precision in many places. e.g DRAW dialect, font size.

[hiiamboris]

@qtxie in the above I'm implying that the last 3 digits of the integer can be used as a fractional part. To put it another way, 1/1000 of a pixel minimum increment size. The same tactic that's used for the money datatype, really.

[qtxie]

@hiiamboris Ah, Ok. You mean a custom decimal format. That needs more work. Also most of the modern OSes use float32! in their Graphic APIs. Using float32! make it easier to integrated with the OS APIs.

[hiiamboris]

I see. Thanks for the info :)

[qtxie]

Some notes about this change:
Pros:

1. More accurate drawing in DRAW.
2. More accurate offset and size. E.g. face/offset, font/size.

Cons:

1. Integer pair works better for images.
2. No bitwise operations. (and, or, xor)

Using float32 pair we can work around those Cons by converting pair/x and pair/y to integer first.
While we cannot have the Pros using integer pair.