I converted this to a draft. In the new forms, the shapes of the logo have darkened edges. Snippet to zoom in on the top of the ‘h’:

((Form fromBinaryStream: PolymorphSystemSettings pharoLogoContents base64Decoded readStream)
	contentsOfArea: (75@0 extent: 30@30)) magnifyBy: 5 smoothing: 1

Old form:

New form:

I’m not sure how to fix that though.

Having done some reading, I understand the problem is the new forms use premultiplied alpha, while what ImageMorph expects, and what the old form uses, is straight alpha. See the section ‘Straight versus premultiplied’ in the Wikipedia article on ‘Alpha compositing’. I could convert from premultiplied to straight: commit e4f54c1. It would seem better to draw the forms directly using straight alpha, but I’m not sure that’s possible: the documentation for image surfaces in Cairo doesn’t seem to list a counterpart that uses straight alpha for ‘CAIRO_FORMAT_ARGB32’ which uses premultiplied alpha. Unless it’s ‘CAIRO_FORMAT_RGBA128F’ but the description doesn’t say.

@tinchodias: Perhaps you know more about this question of straight versus premultiplied alpha in Cairo and Athens?

Hello @Rinzwind I will read more in detail of this FormSet issue during the week, but for the moment I send you the imformation I have at hand.

I learnt a bit of this topic, for example in: pharo-graphics/Alexandrie#35

The following are parts of emails I sent to bloc mailing list:

The Bloc image background had a Form converted to Cairo image, and the debugging tool converts again to Form to show it in a Morph.

An instance of AeCairoImageSurface can be created from a Form, and it provides asForm to get again a Form, but there was an error in the middle that is shown in this example:

f := Form extent: 1@1 depth: 32.
f fillColor: (Color red alpha: 0.4).
f colorAt: 0@0. 
">>> (Color r: 1.0 g: 0.0 b: 0.0 alpha: 0.4) --> this is right, the color we used to fill"

(AeCairoImageSurface fromForm: f) asForm colorAt: 0@0. 
">>> (Color r: 0.3998044965786901 g: 0.0 b: 0.0 alpha: 0.4) --> wrong, the red component is not 1.0"

Curiosity: Both formats are 32 bits, and ARGB order, but there is calculation to perform on each pixel on conversion.

Cairo format is described as follows:

CAIRO_FORMAT_ARGB32: each pixel is a 32-bit quantity, with alpha in the upper 8 bits, then red, then green, then blue. The 32-bit quantities are stored native-endian. Pre-multiplied alpha is used. (That is, 50% transparent red is 0x80800000, not 0x80ff0000.)
(Source: https://www.cairographics.org/manual/cairo-Image-Surfaces.html#cairo-format-t)

Form, is "straight alpha", that can be explained as:

"In this world, RGB and alpha are independent. You can change one without affecting the other. To make an object fade out, you would gradually reduce its alpha value while leaving RGB unchanged."
(Source: https://learn.microsoft.com/en-us/windows/apps/develop/win2d/premultiplied-alpha#straight-alpha)

We already had the conversion Form -> Cairo, but not the Cairo -> Form, so that's what I pushed to close the issue. Plus added 2 tests.

and then:

I had to push a further fix.

Now the conversion from Cairo (premultiplied alpha) to Form (straight alpha), restores the brightness (or saturation?) of each translucent pixel (non opaque).

As this is proper to Cairo to Form conversion, it's not only Alexandrie, also Athens-Cairo should do it.
I ignore what happens in the SurfacePlugin that Athens-Cairo bindings use, but I guess this is not done.
I remembered that when I worked in Hiedra, there was always an annoying effect in the "antialias" pixels
when the background is white. I did a test in a Hiedra example and it seems to be true.

Left: current, Right: applying the transformation from the premultiplied format of Cairo:

zoom:

Thank you for the discussion. This is great to have you around.

@tinchodias: Thanks for the info!

I have now regenerated #pharoLogoContents1 and #pharoLogoContents2 with the assignment to ‘byteArray’ in the snippet changed to the following to convert from premultiplied to straight alpha:

byteArray := ByteArray streamContents: [ :byteArrayStream |
	| surfaceForm convertedForm |
	surfaceForm := surface asForm.
	convertedForm := Form extent: surfaceForm extent depth: 32.
	0 to: convertedForm width do: [ :x |
		0 to: convertedForm height do: [ :y |
			| color |
			color := surfaceForm colorAt: x@y.
			convertedForm colorAt: x@y put:
				(color alpha > 0 ifTrue: [ color / color alpha ] ifFalse: [ color ]) ] ].
	PNGReadWriter putForm: convertedForm onStream: byteArrayStream ].

I force-pushed the branch to a new commit to avoid keeping the previously generated methods, which are a bit large, in the history.

Is it ready for the merge now?

Yes, please do merge this!