[CSS Grid Layout] Fix auto-placement algorithm

Source/core/rendering/RenderGrid.cpp

Index: Source/core/rendering/RenderGrid.cpp
diff --git a/Source/core/rendering/RenderGrid.cpp b/Source/core/rendering/RenderGrid.cpp
index 163edcffe0754ba440f5c24fba58f1a10b62528a..83e5aa72a2e30748533dc962cb336a848e887b2b 100644
--- a/Source/core/rendering/RenderGrid.cpp
+++ b/Source/core/rendering/RenderGrid.cpp
@@ -932,6 +932,9 @@ void RenderGrid::placeAutoMajorAxisItemOnGrid(RenderBox* gridItem)
                 const size_t endOfMinorAxis = autoPlacementMinorAxisDirection() == ForColumns ? gridColumnCount() : gridRowCount();
                 if (minorAxisFinalPositionIndex.toInt() < endOfMinorAxis)
                     break;
+
+                // Discard empty grid area as it does not fit in the minor axis direction.
+                emptyGridArea = nullptr;

[svillar, 2014/06/04 09:48:22]

Wouldn't it be possible to call here
createEmptyGridAreaAtSpecifiedPositionsOutsideGrid() directly instead of
nullifying the variable?

[Manuel Rego, 2014/06/04 10:02:51]

We're looping over the major axis, as you can read in the previous comment:
            // We need to resolve the position for every different index as the
number of cells might vary depending on it.
            // This will happen when we have "span <custom-indent>", which has a
different resolution based on the position.

So we don't have to call createEmptyGridAreaAtSpecifiedPositionsOutsideGrid() if
the empty grid area doesn't fit. As maybe, we found an empty grid area that fits
in the next row/column or whatever.

We just need to call createEmptyGridAreaAtSpecifiedPositionsOutsideGrid() later
if we haven't been able to find any empty grid area.

[Julien - ping for review, 2014/06/10 00:08:05]

It would be good to add such a test case, that would ensure that we don't
regress it by mistake later.

I think this explanation would complement the comment above too.

[Manuel Rego, 2014/06/10 09:49:02]

That was what I was doing in the modification of the original test and now with
the new test case.
I've modified the comment to add some more context.

             }
         }