Typesetting features need not force complex for single-character runs

• Component: Text
• Status: RESOLVED FIXED

[Ned Holbrook]

The only typesetting features are kerning and ligatures, neither of which applies for a single character. As such, there's no need to spool up the complex text machinery for a single-character run just because the font has typesetting features. This is particularly interesting when hyphenation is enabled.

[Ned Holbrook]

Created attachment 104401 [details]
Proposed changes.

In most cases this isn't much of a problem (a percent or less) but every little bit helps.

[Ned Holbrook]

Running the full set of tests shows that fast/writing-mode/text-orientation-basic.html is failing with this patch, I'm investigating now.

[Ned Holbrook]

Created attachment 104428 [details]
Diffs from failing test.

[Ned Holbrook]

(In reply to comment #2)
> Running the full set of tests shows that fast/writing-mode/text-orientation-basic.html is failing with this patch, I'm investigating now.

This should be fixed by 66733.

[Ned Holbrook]

As of r94090, I see no fast or svg pixel test failures due to the proposed changes. I also ran the entire suite of tests without pixel tests and saw no failures.

Comment on attachment 104401 [details]
Proposed changes.

View in context: https://bugs.webkit.org/attachment.cgi?id=104401&action=review

> Source/WebCore/ChangeLog:11
> +        No new tests. (OOPS!)

I think this line will stop the commit bot from landing this patch

[Ned Holbrook]

(In reply to comment #6)
> I think this line will stop the commit bot from landing this patch

Oops, indeed! I will change it to note that the changes are for performance and not expected to change layout.

[Ned Holbrook]

Created attachment 105977 [details]
Updated ChangeLog.

[WebKit Review Bot]

Comment on attachment 105977 [details]
Updated ChangeLog.

Clearing flags on attachment: 105977

Committed r94303: <http://trac.webkit.org/changeset/94303>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.

[James Robinson]

On Chromium Mac on Snow Leopard this patch caused the text metrics on fast/text/international/vertical-text-glyph-test.html to change very, very slightly.  Here's the text diff:

--- /b/build/slave/Webkit_Mac10_6__CG__dbg_/build/layout-test-results/fast/text/international/vertical-text-glyph-test-expected.txt 
+++ /b/build/slave/Webkit_Mac10_6__CG__dbg_/build/layout-test-results/fast/text/international/vertical-text-glyph-test-actual.txt 
@@ -1,8 +1,8 @@
-layer at (0,0) size 785x634
+layer at (0,0) size 785x635
   RenderView at (0,0) size 785x600
-layer at (0,0) size 785x634
-  RenderBlock {HTML} at (0,0) size 785x634
-    RenderBody {BODY} at (8,21) size 769x605
+layer at (0,0) size 785x635
+  RenderBlock {HTML} at (0,0) size 785x635
+    RenderBody {BODY} at (8,21) size 769x606
       RenderBlock {P} at (0,0) size 769x32
         RenderText {#text} at (0,5) size 161x21
           text run at (0,5) width 161: "Simple text path"
@@ -15,8 +15,8 @@
       RenderBlock {P} at (0,320) size 769x32
         RenderText {#text} at (0,5) size 181x21
           text run at (0,5) width 181: "Complex text path"
-      RenderBlock (anonymous) at (0,373) size 769x232
-        RenderBlock {SPAN} at (0,0) size 32x223
-          RenderText {#text} at (5,0) size 21x223
-            text run at (5,0) width 223: "string\x{300C}\x{3042}\x{3001}\x{5909}\x{3063}\x{FF01}\x{300D}\x{3002}"
+      RenderBlock (anonymous) at (0,373) size 769x233
+        RenderBlock {SPAN} at (0,0) size 32x224
+          RenderText {#text} at (5,0) size 21x224
+            text run at (5,0) width 224: "string\x{300C}\x{3042}\x{3001}\x{5909}\x{3063}\x{FF01}\x{300D}\x{3002}"
         RenderText {#text} at (0,0) size 0x0

This also changes the pixel output slightly - the text all renders the same, but the scrollbar looks different because the scrollable area is 1px taller.  I'm planning to rebaseline since this looks harmless, but since the comments indicate that this should not have changed any layout posting as an FYI.

[Ned Holbrook]

Interesting. I wonder if there aren't some minute differences in vertical metrics on SL.