Bug 99874

Summary:

[Chromium] SVG repaint issues

Product:

WebKit

Reporter:

Florin Malita <fmalita>

Component:

SVG

Assignee:

Florin Malita <fmalita>

Status:

RESOLVED FIXED

Severity:

Normal

CC:

d-r, eric, krit, leviw, pdr, schenney, stephen.bannasch, webkit.review.bot, zimmermann

Priority:

Version:

528+ (Nightly build)

Hardware:

Unspecified

OS:

Unspecified

Attachments:

Description	Flags
Leaves unpainted artifacts when shrinking	none
Updated test.	none
Screenshot of observed ghosting.	none
Patch	none

Florin Malita

Reported 2012-10-19 14:32:29 PDT

Created attachment 169697 [details] Leaves unpainted artifacts when shrinking When scaling (shrinking) SVG shapes dynamically, the old area is not always fully repainted.

Attachments
Leaves unpainted artifacts when shrinking (4.79 KB, text/html) 2012-10-19 14:32 PDT, Florin Malita	no flags	Details
Updated test. (865 bytes, text/html) 2012-10-22 08:00 PDT, Florin Malita	no flags	Details
Screenshot of observed ghosting. (2.87 KB, image/png) 2012-10-22 08:20 PDT, Florin Malita	no flags	Details
Patch (105.68 KB, patch) 2012-10-24 10:23 PDT, Florin Malita	no flags	Details Formatted Diff Diff
View All Add attachment proposed patch, testcase, etc.

Florin Malita

Comment 1 2012-10-22 07:00:16 PDT

Not limited to scaling, see http://code.google.com/p/chromium/issues/detail?id=156597

Florin Malita

Comment 2 2012-10-22 08:00:47 PDT

Created attachment 169909 [details] Updated test. The initial test only exhibits the issue for certain browser window sizes. Updated.

Florin Malita

Comment 3 2012-10-22 08:08:56 PDT

Also, this appears to be Chromium-only.

Florin Malita

Comment 4 2012-10-22 08:20:34 PDT

Created attachment 169917 [details] Screenshot of observed ghosting.

Florin Malita

Comment 5 2012-10-24 07:33:16 PDT

I tracked this down to accumulated rounding errors in SVGRenderRoot::paintReplaced() when using subpixel layout: IntPoint adjustedPaintOffset = roundedIntPoint(paintOffset); childPaintInfo.applyTransform(AffineTransform::translation(adjustedPaintOffset.x() - x(), adjustedPaintOffset.y() - y()) * localToParentTransform()); The problem is we're rounding layout units both for adjustedPaintOffset and in localToParentTransform() and effectively adding the results => we're getting 2x the rounding delta. If the delta happens to be 1/2 pixles then we end up being off by one pixel, which skews the painting/repainting rects. We can either adjust by one in the special case of rounding delta == 0.5, or use localToBorderBoxTransform() to avoid getting the delta injected twice. I'll put a patch up shortly.

Florin Malita

Comment 6 2012-10-24 10:23:03 PDT

Created attachment 170422 [details] Patch

Florin Malita

Comment 7 2012-10-24 10:25:20 PDT

Not sure whether subpixel layout is used on other platforms besides Chromium - if it is we'll have to add new expectations to the patch.

Levi Weintraub

Comment 8 2012-10-24 10:28:11 PDT

Comment on attachment 170422 [details] Patch LGTM. Thank you.

Levi Weintraub

Comment 9 2012-10-24 10:28:40 PDT

(In reply to comment #7) > Not sure whether subpixel layout is used on other platforms besides Chromium - if it is we'll have to add new expectations to the patch. Not yet, no.

WebKit Review Bot

Comment 10 2012-10-24 11:14:12 PDT

Comment on attachment 170422 [details] Patch Clearing flags on attachment: 170422 Committed r132377: <http://trac.webkit.org/changeset/132377>

WebKit Review Bot

Comment 11 2012-10-24 11:14:16 PDT

All reviewed patches have been landed. Closing bug.

Florin Malita

Comment 12 2012-10-25 06:05:38 PDT

Committed r132477: <http://trac.webkit.org/changeset/132477>

Note You need to log in before you can comment on or make changes to this bug.