Don't skip SetGradientStops

[kvark]

Fixes #2328
r? @gankro

---

This change is 

[kvark]

Note: WR frame building is already happy skipping this:

            // Do nothing; these are dummy items for the display list parser
            SpecificDisplayItem::SetGradientStops => {}

[pyfisch]

Not sure why but with this PR servo crashes for me because of Option::unwrap. https://gist.github.com/pyfisch/8ff4ff4a476152aeb644a340141dc1ca

Maybe because webrender_api and webrender do not match. Even with matching versions servo crashes. On master this crash does not happen.

[Gankra]

This isn't correct; we clear gradient stops at the start of next. as-is this patch just makes it so gradient-stops never make it to the backend.

[kvark]

@gankro what would be your recommendation then? having that continue gated by a run-time boolean flag that we enable only when debug-serializing?

[Gankra]

I'm not familiar with the debug-serializing feature, but one option would be to split next into two methods: a "raw" one, and a wrapper that consumes the raw one and behaves like the current next. Then anyone who needs to reproduce the input can just use the raw one?

[kvark]

@gankro how about we don't hack thing as low as traversing items and instead put the gradient stop logic in the context flattening? That would make all the low-level things like debug serialization robust. Please take a look at the code in this PR now.

[Gankra]

Presumably the same logic would need to be used in the DL->yaml/json writers? (I don't think we have tests for that)

[kvark]

@gankro good point! I fixed the YAML writer to take those into account and tested it by saving/loading a thing. Appears to work fine.
As to JSON/RON - these folks use debug serialization, and they don't need to be loaded (1) afterwards, so not only they don't require extra fixes, but they are also already improved :)

(1) actually, the capture infrastructure does need RON to be loaded, and with gradient stops we can just have better capturing of gradients on the scene level.

[glennw]

@gankro @kvark Is this one ready to go?

[glennw]

ping @gankro @kvark

[kvark]

Marking as blocked, decided to re-think gradient stops more fundamentally with @gankro .

[bors-servo]

☔ The latest upstream changes (presumably #2315) made this pull request unmergeable. Please resolve the merge conflicts.

[pyfisch]

Are there now concrete plans for improved gradient stops? If not what are the open questions?

[kvark]

@pyfisch last time I heard @gankro was looking into it

[glennw]

@gankro Should this be kept open?

[Gankra]

Let's close this PR in favour of an issue (sorry for letting this slip!)