Description François POYER 2012-02-09 10:10:43 EST

In FlowFigure:
<code>
	public void add(IFigure child, Object constraint, int index) {
		super.add(child, constraint, index);
		// If this layout manager is a FlowContext, then the child *must* be a
		// FlowFigure
		if (getLayoutManager() instanceof FlowContext)
			((FlowFigure) child)
					.setFlowContext((FlowContext) getLayoutManager());
		revalidateBidi(this);
	}
</code>

Problem is on first line: "super.add(child, constraint, index);" causes (in Figure#add) a call to #revalidate(), which signals the figure as invalid to the UpdateManager, while the child Figure doesn't have its FlowContext set yet (and so will cause a NPE if UpdateManager tries to validate it right away).

Setting the FlowContext of child figure before calling super.add does fix the problem.

Note: For some reasons specific to my project, I do not use the "standard" DeferredUpdateManager that queues validation work until a PAINT event occurs but a more "synchronous" one that performs validation as soon as invalid figures are signaled.

<code>
public synchronized void addInvalidFigure(final IFigure f) {
    // Note the invalid figure...
    this.invalidFigures.add(f);
    // ...and perform update and validation right away!
    Display display = Display.getCurrent();
    if (display == null) {
        throw new SWTException(SWT.ERROR_THREAD_INVALID_ACCESS);
    }
    display.syncExec(new UpdateRequest());
}
</code>

This is why the problem occurs every time for me.
But the problem might as well happen with the regular DeferredUpdateManager, depending on multithread weaving I guess. 
More importantly: FlowFigure implementation should probably not depend on some assertion about the UpdateManager's implementation, should it?