Col width is not honored when dynamically updated and it would not make table narrower

Source/core/rendering/RenderTableCol.cpp

Index: Source/core/rendering/RenderTableCol.cpp
diff --git a/Source/core/rendering/RenderTableCol.cpp b/Source/core/rendering/RenderTableCol.cpp
index 26c5935f97f1d3bd065cfbc31b1acc90e6fc6c8a..00b1e8c4e1978b8658d7bca2e1ec72ff357b1059 100644
--- a/Source/core/rendering/RenderTableCol.cpp
+++ b/Source/core/rendering/RenderTableCol.cpp
@@ -51,8 +51,28 @@ void RenderTableCol::styleDidChange(StyleDifference diff, const RenderStyle* old
     // If border was changed, notify table.
     if (parent()) {
         RenderTable* table = this->table();
-        if (table && !table->selfNeedsLayout() && !table->normalChildNeedsLayout() && oldStyle && oldStyle->border() != style()->border())
+        if (table && !table->selfNeedsLayout() && !table->normalChildNeedsLayout() && oldStyle && oldStyle->border() != style()->border()) {
             table->invalidateCollapsedBorders();
+        } else if (oldStyle && oldStyle->logicalWidth() != style()->logicalWidth()) {
+            unsigned nEffCols = table->numEffCols();
+            unsigned currentColumn = 0;
+            for (RenderTableCol* column = table->firstColumn()->nextColumn(); column; column = column->nextColumn()) {
+                if (this == column)
+                    break;
+                currentColumn += column->span();
+            }

[Julien - ping for review, 2014/03/06 00:54:47]

This span computation code actually doesn't work, totally missed that in
previous reviews. It only works because both test cases have the exact same
layout:
<colgroup>
   <col></col>
</colgroup>

The following are allowed too and will be mishandled by your code:

<colgroup span="5"/>
<col/>

or

<colgroup>
   <col span="3"/>
   <col/>
</colgroup>

The rule is that <colgroup> is only allowed to have a |span| attribute if it
doesn't have <col> children (see HTML5
http://www.whatwg.org/specs/web-apps/current-work/multipage/tabular-data.html...).

I don't think we have some code to do this computation and I fear that we don't
want to spread such tricky code around (RenderTable already has some column
information but we can't reuse it). My take is to revert back to the original
approach to invalidate all the cells but put a FIXME that explains that we could
do better.

The previous cases should be tested as they would have been broken by this
change.

+            for (RenderTableSection* section = table->topSection(); section; section = table->sectionBelow(section)) {
+                unsigned currentColSpan = currentColumn + span();
+                for (unsigned j = currentColumn; j < currentColSpan && currentColSpan <= nEffCols; j++) {
+                    for (unsigned i = 0; i < section->numRows(); i++) {
+                        RenderTableCell* cell = section->primaryCellAt(i, j);
+                        if (!cell)
+                            continue;
+                        cell->setPreferredLogicalWidthsDirty();
+                    }
+                }
+            }
+        }
     }
 }