Forked off from Bug 689828, comment 17, and Bug 689828, comment 18: [lars] Instead of introducing denseStart would it not be easier simply to accept that there may be holes, and check for them, and then when denseStart is low enough the C++ code would just make it zero and insert holes at the beginning? Not clear the inline code bloat will be any worse than having to deal with denseStart. [felix] the inline code bloat for allowing holes is probably comparable to that for denseStart-compensentation, and would avoid the memory read imposed by loading denseStart. Checking for a hole does introduce a branch instruction that would not necessarily be there with denseStart. I'll try to evaluate both approaches.