ImageSharp.Drawing transformations are unintuitive

[ryanackley]

If you're coming from any other graphics API, ImageSharp's transformations are unintuitive because they behave a little strangely in comparison. Look at this code

transform.AppendTranslation(new SixLabors.Primitives.PointF(0, 2.75f * dpi));
transform.AppendRotationDegrees(slantedAngle);
img.Mutate(ctx =>
{
     ctx.Transform(transform);
     ctx.DrawText(text, font, SixLabors.ImageSharp.Color.Black, new SixLabors.ImageSharp.PointF(0f, 0f));
});

Coming from OpenGL, GDI+, Javascript Canvas, etc., you would expect this code to apply the transformation to DrawText . What it actually does is apply the transformation to what's already been drawn. This makes it impossible to apply discrete transformations on a single draw operation. For example, drawing a single thing rotated at an angle. I'm very new to ImageSharp so maybe I'm missing something.

Pretty much every other drawing library out there applies the transform as a state on the drawing context that applies to subsequent operations. You can typically manage the state using push/pop type functions.

[JimBobSquarePants]

Please do not open non-issues in the issue tracker. There is a template (that you deleted) for a reason.

I will transform this to the correct repository (drawing) and move it to the correct discussions channel (ideas) so that any conversation can continue there.

[ryanackley]

Thanks for being a maintainer on this library. I understand you're a volunteer. You don't owe me anything. Keep in mind, I'm volunteering my time too. By taking the time to create this issue and articulate my feedback and provide the benefit of my experience, I am trying to help this project. I have no other ulterior motive. We've already figured out workarounds and have working code. There is no expectation that this will be fixed to address a particular need.

1. Didn't delete the template. Github didn't present me with one. Probably a bug because I wasn't logged in. It redirected me to authenticate after I clicked the create issue button. After I authenticated, I was presented with a blank create issue form.
2. When you moved the issue, Github didn't link from the old issue. It said the issue was moved to a repo I don't have permission on or was deleted. So I created another one. Wasn't trying to spam you.

[tocsoft]

I was planning on proposing and trying to shift us over to have this sort of API... We already have state in the IImageProcessingContext we don't need to create a new concept for this we just need to first provide overloads that take in a transformation Matrix (pre-req. for all follow on work) then on all operation call site update to load a default transform from the grab bag on IImageProcessingContext (if its set) and pass it down into the processors to intelligently apply as required...

Same can happen for Brushs and Pens too if required, however we would probably have to break the DrawText apis too accommodate that because we currently have different logic based on whether a pen/brush is passed in we would have to change that to separate APIs calls one for outline one for draw as we could have a transient pen around that probably wont want to be used by the text drawing code.

[antonfirsov]

@tocsoft ah I see, so we basically evolved to the Canvas-like design. It was clearly the goal with the property bag, but I somehow missed the point.

[antonfirsov]

I wish we could rename IImageProcessingContext to IImageCanvas now.

[JimBobSquarePants]

I wish we could rename IImageProcessingContext to IImageCanvas now.

Nah @antonfirsov context is fine. Canvas to me suggest something I apply multiple layers to.

@tocsoft Are you thinking adding the property to the bag for the Drawing APIs only? Adding the the main would be confusing for consumers of the Transform extensions.

[JimBobSquarePants]

@tocsoft @antonfirsov I cannot believe it's been 4 years since we looked at this!

I actually want to rejig ImageSharp.Drawing to introduce a ICanvas like type for the next major version.

If we designed it correctly (completely separate from IImageProcessingContext) we could allow for different implementations (GPU) and potentially make the library a candidate for UI generation.

My thought process is that it would work like System.Drawing Graphics or Skia's SKCanvas.

What do you think?