Unfortunate interactions between inexact numbers and the image library

[blerner]

From the original emails:

I have a computed ::Number value which is a float between 0 and 1. How do I get an integer to dynamically construct an RGB value?

import image as I
import image-structs as C

I.circle(20, "solid",C.color(240,1,240,1))  # ==>magenta circle

I.circle(20, "solid",C.color(240.2,1,240,1))  # ==>black circle

And

The code below will work, using num-exact to force the number to be exact. But I'm not happy with two things here that will get fixed:

• truncate/floor/ceiling should produce exact results
• one of color or circle should have given you an error message, not produced bogus answers

I.circle(20, "solid",C.color(240,1,num-exact(num-truncate(240.1)),1))  # ==>magenta circle

[shriram]

Indeed, both should happen, right? Exact results and errors on nonsensical inputs?

[blerner]

Related problems: place-image-align(IMAGE1, 2.5, 10, "left", "middle", IMAGE2) results in an internal error right now. Any non-integer value dies, there...

[blerner]

(This seems specifically Scene related: SceneImage.prototype.add is using x and y without going through the jsnums library, so it's just dying horribly...

[blerner]

@jpolitz I've got this almost finished now: I've just gone through and fixed up image.js, image-lib.js and world.js to use jsnums.toFixnum properly everywhere, so we don't have internal errors any more. (I'll want a code review, shortly -- it's fiddly and I might've typoed or missed one, and we don't have a comprehensive test suite for this, right?)

I've also modified data Color in image-structs to annotate its arguments to be NumIntegers. This catches 240.1 as a color component, and also catches ~240. It gives a mediocre error message -- but it's way better than the Internal error we currently have! -- but I'm not 100% certain we want this to be the solution. We also need to revisit #862, since that impacts the ergonomics of this solution as well.

[blerner]

(Also, incidentally, the semantics of colors evidently has changed at some point, so that alpha now ranges over integers from 0 to 255, rather than floats from 0 to 1. Not sure when that happened, but it impacts the original example slightly.)

[sorawee]

@blerner regarding your last comment, see also #475

[schanzer]

IIRC, I implemented it this way to preserve compatibility with DrRacket. CPO is under no such restrictions.

[blerner]

I don't care either way, but it should be documented properly and enforced properly. Even the NumPositive constraint I have right now isn't rigid enough for the rgb components. So this has to get revised a bit anyway...

Which range do people prefer? Both scales are arbitrary, to some degree :-)

[schanzer]

What's most common in the languages we expect new Pyreteers to be most familiar with?

[shriram]

For my incoming cs019 applicants, Java and Python.

[blerner]

Perfect. Java accepts (0-255) for all arguments, or (0.0--1.0) for all arguments. Python supports (0-255), (0.0--1.0), three-char-hex, six-char-hex, eight-char-hex, and explicit percentages.

So...that tells us nothing whatsoever ;-)

[schanzer]

CSS uses 0-1, IIRC. Lots of people come in with minimal web knowledge, and "it works like the internet" is always a convincing response in PD. ;)

(Or maybe let Dorai fix it in the numbers library... ducks and runs)