The algorithm as written is meant to be two pass:
1. first grid items that are constrained such that only one dimension is auto should be placed
2. after that all grid items having both dimensions as auto can be placed
The spec text makes it sound like steps 1 and 2 are in a single step though.
Also it isn't clear with the current algorithm whether the "base grid" onto which items are placed is updated after step one is performed for the purposes of wrapping grid items to new rows/columns.
Additionally, this algorithm may not be the optimal auto-placement algorithm. Maybe it should be a one pass algorithm that finds the first free spot for an item beginning at 1,1, or maybe this should just be removed altogether.
Refer to this thread for some discussion on this topic:
Reassigning all grid bugs to me.
Here's a new thread relevant to auto placement. If updating the algorithm, test against the use cases presented here:
There used to be an ISSUE note "Review this wrt concerns in bug 16044."
in the spec but it's no longer there in the current version:
Does that mean this bug can be closed? Or was the ISSUE note
in the spec lost accidentally?