697057 - Scaled patterns are blurred

Status: RESOLVED FIXED

Description

[Robert Longson]

bug 691646 fixed one way that patterns become blurred but broke another - bug 535185.

In bug 535185 the pattern is small and is scaled when it's used. In bug 691646 the pattern has a patternTransform which scales it. We need to account for both scalings when determining the size of the pattern to create.

Comment 1

[Robert Longson]

Created attachment 569330 [details] [diff] [review]
patch

I'll see if I can do a reftest for this.

Comment 2

[Robert Longson]

Created attachment 569335 [details] [diff] [review]
patch with reftest

Comment 3

[Daniel Holbert [:dholbert]]

Looks like pattern-scale-01a.svg and pattern-scale-01-ref.svg are missing from the patch.

Comment 4

[Daniel Holbert [:dholbert]]

oh, sorry -- the ref already exists, and "a" is renamed from an existing file.  Cool.  Darn bugzilla's diff viewer.

Comment 5

[Daniel Holbert [:dholbert]]

Comment on attachment 569335 [details] [diff] [review]
patch with reftest

>@@ -254,21 +254,23 @@ nsSVGPatternFrame::PaintPattern(gfxASurf
>   // routine.
>   *patternMatrix = GetPatternMatrix(bbox, callerBBox, callerCTM);
[...]
>+  float scale = nsSVGUtils::MaxExpansion(callerCTM);
>+
>   bool resultOverflows;
>   gfxIntSize surfaceSize =
>     nsSVGUtils::ConvertToSurfaceSize(
>-      gfxSize(patternWidth * fabs(patternMatrix->xx),
>-              patternHeight * fabs(patternMatrix->yy)),
>+      gfxSize(patternWidth * fabs(patternMatrix->xx) * scale,
>+              patternHeight * fabs(patternMatrix->yy) * scale),
>       &resultOverflows);


Why do we need this extra "scale" factor?  Shouldn't we already be getting all the scaling information that we need from |callerCTM| when we create patternMatrix, in the "GetPatternMatrix(..., callerCTM)" call?

See also bug 691646 comment 13... Perhaps that helps with this?

Comment 6

[Daniel Holbert [:dholbert]]

Actually I'm less sure about bug 691646 comment 13 now, as noted there, so disregard that for now.

However, I'm still a little iffy on the patch here...  Note that GetPatternMatrix() already involves a scale by 1/MaxExpansion(callerCTM), so your attached patch here is basically reversing that scale, for the purposes of the surface-size computation.  That seems like it could be extra work.

Perhaps we really want to remove that scale from GetPatternMatrix completely? (possibly lifting it up to this method, and applying it to patternMatrix after we've computed the surfaceSize?)

I'm not sure -- I don't totally grok what's going on here.  I'd like to get a better idea of why the patch here (or its successor) is correct, before granting r+.

Comment 7

[Robert Longson]

We only want to apply the patternTransform scaling (the scaling in the pattern) without the callerCTM (i.e. the scaling on to the object). You are right in that we're multiplying and dividing in by MaxExpansion order to get the number we'd get from GetPatternTransform directly.

Comment 8

[Robert Longson]

Created attachment 569636 [details] [diff] [review]
updated patch to address efficiency concerns

Comment 9

[Robert Longson]

Created attachment 569638 [details] [diff] [review]
right patch

Comment 10

[Daniel Holbert [:dholbert]]

Comment on attachment 569638 [details] [diff] [review]
right patch

Could you put a short chunk of documentation over nsSVGPatternFrame::GetPatternMatrix() (either the decl or the impl), to reflect its new usage?

r=me with that.

Comment 11

[Robert Longson]

pushed https://hg.mozilla.org/integration/mozilla-inbound/rev/ff296aef84fe

Comment 12

[Ed Morley [:emorley]]

Do you think this could have caused the following regressions?

Talos Regression :( V8 increase 2.78% on MacOSX 10.6.2 (rev3) Mozilla-Inbound
Talos Regression :( Dromaeo (String/Array/Eval/Regex) decrease 4.17% on MacOSX 10.6.2 (rev3) Mozilla-Inbound

https://groups.google.com/forum/#!topic/mozilla.dev.tree-management/w2WZEx3ugn0

Comment 13

[Daniel Holbert [:dholbert]]

(In reply to Ed Morley [:edmorley] from comment #12)
> Do you think this could have caused the following regressions?

No. The patch *only* touched code for the <pattern> element in SVG, so it has no effect on non-SVG content. (I'd be very surprised if either of those test suites included SVG at all, let alone SVG <pattern> elements.)

I'm guessing that the treewatcher bot might have mis-attributed blame here, and the real guilty party might be immediately before/after (or maybe an infra change?)

Comment 14

[Ed Morley [:emorley]]

(In reply to Daniel Holbert [:dholbert] from comment #13)
> No. The patch *only* touched code for the <pattern> element in SVG

Ah yes. Sorry was in a rush so forgot to actually check the diff :$

Comment 15

[Daniel Holbert [:dholbert]]

No problem; thanks for your vigilance!  This does look like a real regression (with unknown cause so far) -- let's follow up in m.d.tree-management.

Comment 16

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/ff296aef84fe