Performing a non-square rotate() with Imagick, then cropping, can result in incorrect Image

[kirasong]

Note: This is only for *some* recent versions of ImageMagick. Older ones seem to be a bit more sane, and work fine without any patch whatsoever.

When ImageMagick performs certain operations, it sets its virtual page's x/y origin values to non-zero numbers and leaves them there, which can break some further operations that depend on the virtual page.

In this case, when rotating in non 90 degree increments, Imagick doesn't set the x/y, which means that core doesn't run into this bug, since the built-in image editor only allows for 90 degree rotates in either direction.

However, if you're a user of the API, and perform a "non-square" rotate, then a crop, you could end up with an image result that differs from what you'll get from GD.

There are two ways of fixing this:

• Update the virtual page with each update_size()
• Update the virtual page immediately after a rotate() only

The first is what I think we should land on eventually, because it will prevent plugins' methods from breaking core's accidentally (if they do not reset the page's values). However, that would require an additional error message to be added, which means a new string.

The second would fix the bug for anything core's APIs would do, but would depend on plugins handling resetting the page values appropriately in their own methods. It is, however, a bit safer (and wouldn't require a string change), if this is something that might land for 3.5.

I've attached patches for each method.

[kirasong]

Pre-Patch Example of 30 degree rotate, then crop.

[kirasong]

Pre-Patch Example (ImageMagick 6.7.7-6) of 30 degree rotate, then crop.

[kirasong]

Post-Patch Example (ImageMagick 6.7.7-6) of 30 degree rotate, then crop.

[kirasong]

Attached example rotate, then crop(s).
GD is the same as attached before and after patch.

[nacin]

I'm thinking this should wait until 3.6.

[nacin]

Basically, this only comes into play when a plugin is using the rotation API outside of standard 90-degree increments. Good find, but punting.

[markoheijnen]

Patch 22663.in_rotate.diff​ is the fix for this ticket.

[kirasong]

Would still like to see this go in for 3.6.

Agreed with @markoheijnen on this, 22663.in_rotate.diff​​ is the least heavy-handed solution.

[ryan]

22663.in_rotate.diff is still good to go?

[kirasong]

Yep, double-checked with tests. Good to go.

[kirasong]

Can we land 22663.in_rotate.diff​ for 3.7? It's pretty straight-forward, and would be great for it not to ride another release cycle.

[nacin]

In 25636:

Fix non-square rotations when using the Imagick image editor.

props DH-Shredder.
fixes #22663.