926258 - Compute path lengths with Moz2D

Status: RESOLVED WONTFIX

(Core :: Graphics, defect)

Description

[Matt Woodrow (:mattwoodrow)]

SVG curently uses gfxFlattened path to compute lengths of paths, and coordinates along the path.

We need a way to implement this with Moz2D.

Comment 1

[Matt Woodrow (:mattwoodrow)]

Attachment: Part 1: Get rid of gfxFlattenedPath

This merges the gfxPath and gfxFlattenedPath API's.

I don't think we want to have a separate 'flattened path' object before we can do measurements, since D2D at least provides an API for doing directly on any path.

This is going to be a bit slower than what we currently have (a few extra path copies), but we should be able to improve it again as we switch to moz2d. I also doubt it's called enough to matter.

Comment 2

[Matt Woodrow (:mattwoodrow)]

Attachment: Part 2: Implement a Moz2D version of gfxPath

Comment 3

[Matt Woodrow (:mattwoodrow)]

Comment on attachment 816407 [details] [diff] [review]
Part 2: Implement a Moz2D version of gfxPath

Review of attachment 816407 [details] [diff] [review]:
-----------------------------------------------------------------

::: gfx/2d/2D.h
@@ +451,5 @@
> +
> +  virtual double ComputeLength() { return 0; }
> +
> +  virtual Point ComputePointAtLength(double aLength,
> +                                     Point* aTangent) { return Point(); }

These should become pure virtual of course, once we have actual implementations.

Comment 4

[Matt Woodrow (:mattwoodrow)]

Attachment: Part 3a: D2D implementation of path measurement

Comment 5

[Matt Woodrow (:mattwoodrow)]

Attachment: Part 4: Start using Moz2D gfxContext's when measuring paths

Comment 6

[Matt Woodrow (:mattwoodrow)]

Refactoring (not actually using moz2d for anything): https://tbpl.mozilla.org/?tree=Try&rev=aaadf3b32b9e

Enabled Moz2D path measurement for d2d (some bugs left): https://tbpl.mozilla.org/?tree=Try&rev=2de0895f0884

Comment 7

[Matt Woodrow (:mattwoodrow)]

Bas had ideas for doing path measurements of curved path segments without flattening.

I have some in-progress patches for refactoring the iteration of cairo paths and moving that into PathHelpers.h so that we can do an implementation for Moz2d/cairo.

I'll let him weight in here before continuing on that too much to avoid potentially wasting work.

Comment 8

[Bas Schouten (:bas.schouten)]

Comment on attachment 816407 [details] [diff] [review]
Part 2: Implement a Moz2D version of gfxPath

Review of attachment 816407 [details] [diff] [review]:
-----------------------------------------------------------------

::: gfx/2d/2D.h
@@ +450,5 @@
>    virtual FillRule GetFillRule() const = 0;
> +
> +  virtual double ComputeLength() { return 0; }
> +
> +  virtual Point ComputePointAtLength(double aLength,

We probably want this to be 'Float' as that's what we generally use in Moz2D.

::: gfx/thebes/gfxPath.cpp
@@ +113,5 @@
> +        // The y value of aOffset is the offset along the normal vector to apply
> +        Point normal(-tangent.y, tangent.x);
> +        result += normal * aOffset.y;
> +
> +        if (aAngle)

nit: { }

Comment 9

[Bas Schouten (:bas.schouten)]

Comment on attachment 816409 [details] [diff] [review]
Part 3a: D2D implementation of path measurement

Review of attachment 816409 [details] [diff] [review]:
-----------------------------------------------------------------

::: gfx/2d/PathD2D.h
@@ +87,5 @@
>  
>    virtual FillRule GetFillRule() const { return mFillRule; }
>  
>    ID2D1Geometry *GetGeometry() { return mGeometry; }
> +  

nit: whitespace

Comment 10

[Matt Woodrow (:mattwoodrow)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/b37762151fbd
https://hg.mozilla.org/integration/mozilla-inbound/rev/21793156b6ea

Comment 11

[Matt Woodrow (:mattwoodrow)]

Assigning to Bas since he's looking at the other implementations.

Comment 12

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/b37762151fbd
https://hg.mozilla.org/mozilla-central/rev/21793156b6ea

Comment 13

[Jonathan Watt [:jwatt]]

This blocks landing of the patch in bug 930468.

Comment 14

[Jonathan Watt [:jwatt]]

Make that bug 930577.

Comment 15

[Jonathan Watt [:jwatt]]

A suggestion from Andreas on IRC:

gal	I would add a helper for that using skia
gal	instead of trying to add that to every backend
gal	at least as a fallback
jwatt	gal: how does skia fit in here - does that mean calling skia API from gecko code, or calling a Moz2D helper that uses skia regardless of platform?
jwatt	the Moz2D helper taking a mozilla::Path
gal	one way of doing it is adding a Factory::CreateDefaultSoftwareDrawTarget()
gal	and then have skia be that
gal	and then your helper can go there, get a DT, create a path, and do the math
gal	you implement it for the actual skia backend, the rest fall into this path

Comment 16

[Matt Woodrow (:mattwoodrow)]

I don't think the skia path API implements what we need here. SkPathHelpers has a getLength function, but I don't see any ComputePointAtLength equivalent.

I was thinking we'd probably have to do the same sort of thing with a cairo path. Unsupported backends replay their Path object into a cairo path object, and then do the computations using that.

Comment 17

[Jonathan Watt [:jwatt]]

It would certainly be helpful if the same DT backend can be used to get PathBuilders to create paths for painting, measuring, etc. That way we can cache the one path and reuse it for everything, rather than having to cache more than one path, or regularly recreate paths.

What's the current status of the various backends in regards to this bug? Which support what we need, and which do not?

Comment 18

[Bas Schouten (:bas.schouten)]

I will do the required work here on the very short term, sorry I haven't gotten to it yet.

Comment 19

[Andreas Gal :gal]

Using Cairo here seems counter-intuitive. We are trying to kill cairo.

Comment 20

[Jonathan Watt [:jwatt]]

Comment on attachment 816410 [details] [diff] [review]
Part 4: Start using Moz2D gfxContext's when measuring paths

This has been obsoleted.