I'm working on writing Birch for trunk and I noticed the following:


Prior to 3.0 the offset was the literal offset into the data file, yet now we 
seem to be doing the position encoded with the key (for all rows regardless of 
if they're > 64kb and thus have an index component) plus the serialized offset. 
I also see there is now a a "header" offset.

In RowIndexEntry there is:

 * @return the offset to the start of the header information for this row.
 * For some formats this may not be the start of the row.
public long headerOffset()
    return 0;

 * The length of the row header (partition key, partition deletion and static 
 * This value is only provided for indexed entries and this method will throw
 * {@code UnsupportedOperationException} if {@code !isIndexed()}.
public long headerLength()
    throw new UnsupportedOperationException();

In 2.1 we stored the partition key, deletion, but not static row -- but we 
didn't need or use this so I'm guessing this is actually just to support static 
rows? Is there any further documentation around the header in other classes 
that I just haven't come across yet? Any thoughts on position + offset and why 
this behavior changed? Thanks


Reply via email to