[filter-effects-1] Clarify color-interpolation-filters behavior on feMerge

[RazrFalcon]

https://drafts.fxtf.org/filter-effects/#propdef-color-interpolation-filters, https://drafts.fxtf.org/filter-effects/#feMergeElement

Let's say I have something like this:

<svg viewBox="0 0 200 200" xmlns="http://www.w3.org/2000/svg">
    <filter id="filter1" filterUnits="userSpaceOnUse" width="200" height="200">
        <feGaussianBlur in="SourceGraphic" stdDeviation="1" color-interpolation-filters="sRGB" result="blur1"/>

        <feGaussianBlur in="SourceGraphic" stdDeviation="1" color-interpolation-filters="linearRGB"/>
        <feOffset dx="40" dy="40" result="blur2"/>

        <feMerge color-interpolation-filters="linearRGB">
            <feMergeNode in="blur1"/>
            <feMergeNode in="blur2"/>
        </feMerge>
    </filter>
    <g filter="url(#filter1)">
        <rect x="40" y="40" width="40" height="40" fill="yellow"/>
        <rect x="60" y="60" width="40" height="40" fill="blue"/>
    </g>

    <rect x="1" y="1" width="198" height="198" fill="none" stroke="black"/>
</svg>

1. What color space feMergeNode should use? The one that set on the parent feMerge or on the result primitive?
2. I have to convert all in images/buffers to the specified color space, than merge them, than convert the result back to sRGB? Or should I leave it in the specified color space?
   This one actually affects all filters. Should the result be kept in the specified color space or should it be converted back to sRGB?

The actual results are mostly ok. The main problem is that Batik breaks colors and my application have a wrong color on the overlapped part.

* the diff is relative to Chrome

[fsoder]

1. feMerge is the filter primitive and feMergeNodes are its inputs - color-interpolation-filters define the colorspace in which the filter primitive operates. So here the feMerge operation should be performed in the linearRGB space. color-interpolation-filters does not apply to feMergeNode (see the property definition you linked).

2. Since - per above - the colorspace to use is linearRGB, all inputs (feMergeNodes) should be converted to the filter primitive "operation" colorspace. I don't think the spec says anything about how to "store" any intermediate results, but since conversions usually are lossy (and not generally free) I would imagine most implementations store the result in the operation colorspace (the one specified by color-interpolation-filters). Also, since the default colorspace is linearRGB, usually the whole filter chain will operate as linearRGB only converting from/to sRGB on input(s) and output. Maybe the spec should say something about intermediates.

[RazrFalcon]

So basically it's implementation defined, as long as buffers are not converted twice to the same color space? So if the input is already Linear, the current filter primitive should not convert in to the Linear again (in case all of the primitives are LinearRGB).

[fsoder]

I'd go as far as say that it (intermediate storage) is undefined - and maybe it shouldn't be (since it makes assumptions about buffer precision). (Having it undefined makes it difficult to test reliably too.)