Added conversion from RGBa to RGB

[radarhere]

Resolves #6706. Alternative to #6707

The new conversion function is just a copy of

Pillow/src/libImaging/Convert.c

Lines 463 to 480 in a4ec9f3

	static void
	rgba2rgbA(UINT8 *out, const UINT8 *in, int xsize) {
	int x;
	unsigned int alpha;
	for (x = 0; x < xsize; x++, in += 4) {
	alpha = in[3];
	if (alpha == 255 || alpha == 0) {
	*out++ = in[0];
	*out++ = in[1];
	*out++ = in[2];
	} else {
	*out++ = CLIP8((255 * in[0]) / alpha);
	*out++ = CLIP8((255 * in[1]) / alpha);
	*out++ = CLIP8((255 * in[2]) / alpha);
	}
	*out++ = in[3];
	}
	}

except with the last channel set to 255, as happens in other conversions to RGB.

Pillow/src/libImaging/Convert.c

Lines 436 to 446 in a4ec9f3

	static void
	rgba2rgb(UINT8 *out, const UINT8 *in, int xsize) {
	int x;
	for (x = 0; x < xsize; x++) {
	*out++ = *in++;
	*out++ = *in++;
	*out++ = *in++;
	*out++ = 255;
	in++;
	}
	}

[bigcat88]

Yep, you're right, that's my bad. Good Job! :)

[bigcat88]

but I was always thinking that algorithm for premultiplied alpha is
v[0] = v[0] * (v[3] / 255)

[nulano]

but I was always thinking that algorithm for premultiplied alpha is
v[0] = v[0] * (v[3] / 255)

That is correct. If we denote RGBA as the 8-bit non-premultiplied values and R'G'B'A as the 8-bit premultiplied values, then to get the premultiplied values you calculate R' = (R * A) / 255. (Doing the division after to avoid rounding errors.)

If you want to go back to the non-premultiplied values, you rearrange that equation to get R = (R' * 255) / A.

[bigcat88]

Can this be merged before a new version is released?

[radarhere]

Simply waiting for a review from another maintainer.

There is often more activity just before a release, and this is a relatively simple change, so in my experience I would expect this to make it in.