From Richard's mail:


                int rowIndex = getNodeRowIndex(child);
                int columnIndex = getNodeColumnIndex(child);
                int rowEnd = getNodeRowEnd(child);
                int columnEnd = getNodeColumnEnd(child);

Each call to get some layout constraint ultimately involves doing a map lookup on the node (and in some cases, multiple map lookups). For example, getNodeRowEnd calls getNodeRowSpan and getNodeRowIndex. getNodeRowSpan calls getRowSpan, which calls getConstraint, which looks up the value in the map. GetNodeRowIndex calls getRowIndex which calls getConstraint which looks up the value in the map. And we're doing these calls for each and every child.

If we're going to store constraints in a Map, we ought to at least have constraint objects per layout container, and store all the constraints in those objects. These need not be public API (in fact, I do not consider these map index names to be public API, thankfully I did manage to ensure that all of these keys are kept as private static final Strings instead of public). So for example, we could have a GridPaneConstraints object with all of the various constraints one might set. Then for any given managed node added to a GridPane, a single entry is added to the map.

Then, during layout, we only have to look this constraints object up once per child and then we can read all the values from it directly. Since this constraints object is private to the implementation, it can literally be a private static final class with private fields that the GridPane implementation can read / write to directly, thereby skipping any method call overhead entirely.

And then in places like line 1119, we are calling getNodeRowIndex all over again for each child. And again on 1692, and so on. I bet if we counted up the number of times we call HashMap.get it would be enormous.