Rectangle madness

[vicuna]

Original bug ID: 7498
Reporter: iesvs
Assigned to: @diml
Status: assigned (set by @diml on 2017-04-21T12:43:38Z)
Resolution: open
Priority: normal
Severity: minor
Version: 4.04.0
Category: standard library

Bug description

Graphics.draw_rect and Graphics.fill_rect have strange behaviors.

I will suppose that they use the same meaning for their arguments.
I see two behaviors

1. how I understand the documentation
   fill_rect x y z h
   means { px, py : x <= px < x + w and y <= py < y + h }

2. what is currently implemented on Linux
   fill_rect x y z h
   means { px, py : x <= px <= x + w and y <= py <= y + h }

In any cases the behavior of windows must be the same. I use the C compiler of VS 2010 to compile ocaml.

For the case 1, Linux must be changed and Windows too
I have the fix fix1.patch

For the case 2, Windows does not do the same thing as Linux
I have the fix fix2.patch

I also add a test file (test.ml).

ocamlopt -o test graphics.cmxa test.ml && ./test 1
will test the case 1
and
ocamlopt -o test graphics.cmxa test.ml && ./test 2
will test the case 2

The documentation of get_image states that the arguments have the same meaning than fill_rect. That's why case 1 is probably the expected behavior.

By the way, there is the commit
9f20c9c8a10307f40406890352ba3fbb2e4bd735
which looks like the opposite of fix1 (for the Linux part). That's strange.

I don't have Mac OS X to test, sorry.

File attachments

• rectanglemadness.tar.xz

[vicuna]

Comment author: @diml

I agree that case 1 should be the expected behavior; if you ask for a rectangle of width 10, you should get a rectangle of width 10, not 11.

I tried fix1.patch and it seems to fix the issue, however there are some problems with edge cases on Windows:

• [draw_rect 10 10 0 0] draws a 2x2 square
• [draw_rect 10 10 1 1] draws nothing

On Linux, [draw_rect 10 10 0 0] ends up passing negative values to XDrawRectangle. This doesn't cause a failure or bad behavior but should probably be avoided.

[vicuna]

Comment author: iesvs

I forgot to test edge cases. So I just dit it with fix1.

On Windows when drawing a rect of size 0, the windows API invert the coordinates to always use nondecreasing coordinates. So a rect of size 2 is drawn.
I didn't observe such behavior with fill rect.

With Linux I noticed no problems with fill_rect, but draw rect draw a very big rect when trying to draw a rect of size 0.

Maybe disable draw_rect and fill_rect when width or height is 0 will prevent strange behaviors.

[vicuna]

Comment author: @diml

Yh I suppose we could have special cases when width is zero and/or height is zero. We could do this in the OCaml code, to avoid having to rely on the behavior of the underlying API for edge cases.

BTW, any chance you could submit a PR on github? This would make the review easier.