GD: imagerotate produces inaccurate results

[cjhewett]

Description

The following code:

<?php
$gd = imagecreatetruecolor(5000,5000);
imagefill($gd, 0, 0, imagecolorallocate($gd, 255, 255, 255));

$rot = deg2rad(-1);
$a = cos($rot);
$b = sin($rot);

$gdAffine = imageaffine($gd, [$a ,$b, -$b, $a, 0, 0]);

ob_start();
imagejpeg($gdAffine);
$jpg = ob_get_clean();

echo '<img src="data:image/jpg;base64,'.base64_encode($jpg).'">';

$gdRotate = imagerotate($gd, rad2deg(-$rot), 0);

ob_start();
imagejpeg($gdRotate);
$jpg = ob_get_clean();

echo '<img src="data:image/jpg;base64,'.base64_encode($jpg).'">';

Resulted in this output:

• This rotates a 5000x5000 canvas by 1 degree using both imageaffine and imagerotate
• This should result in a black triangle at the top of the canvas with a height of 5000*sin(1deg) = 87 pixels
• Both methods produce a final canvas of 5087x5087 which is the expected size
• imageaffine produces a triangle of 87px high ✔
• imagerotate produces a triangle of 73px high ❌ (this implies an actual rotation of ~0.83 degrees)
• Looking to the right hand side of the top of imagerotate shows substantial clipping, suggesting the image has been scaled up in some way and the edge pixels lost.

The image:

Top left of imageaffine followed by top left of imagerotate:

Top right of imageaffine showing pixels correctly go into edge

Top right of imagerotate showing substantial cropping of pixels along edge.

GD Version: 2.3.3

PHP Version

PHP 8.3.8

Operating System

No response

[cmb69]

This is almost certainly an upstream issue, at least if you're using a system libgd. What does PHP info report about this?

[cjhewett]

phpinfo():

gd

GD Support enabled
GD headers Version 2.3.3
GD library Version 2.3.3
FreeType Support enabled
FreeType Linkage with freetype
GIF Read Support enabled
GIF Create Support enabled
JPEG Support enabled
PNG Support enabled
WBMP Support enabled
XPM Support enabled
XBM Support enabled
WebP Support enabled
BMP Support enabled
AVIF Support enabled
TGA Read Support enabled

[cmb69]

Thanks for the feedback! So this would indeed be an upstream issue.

Anyhow, while I haven't had a closer look at this particular issue, I presume that it is (at least partially) caused by the fixed integer arithmetic used by gdImageRotate(); see libgd/libgd#951, last two paragraphs of the post.

[cjhewett]

Should I open this as a new issue on https://github.com/libgd/libgd/issues, add this as a comment on the gdFixed issue, or leave it open here?

[cmb69]

In doubt, file a new ticket against libgd. (However, there may not be any feedback for quite a while.)

[cjhewett]

I have posted this upstream to libgd/libgd#955 so I will close off this one.