What I do notice reading https://www.sqlite.org/fileformat.html (if I get it
right) is that index lines are in the form (for an index on a,b,c ie):

<a1> <b1> <c1> <pk1>
<a1> <b1> <c1> <pk2>
<a1> <b1> <c1> <pk3>

Whereas they could be represented as:

<a1> <b1> <c1> [ <pk1>, <pk2>, <pk3> ](3)

whith [pk_list] being a subtree; reverse lookup from table record to index
record would be arguably as fast if not faster. So this would have the
advantage of more compact indexes (less data), having an index line count
(no prefix so there is always just 1 update involved), with the downside of
the complexity of an added level of indirection.

Sent from: http://sqlite.1065341.n5.nabble.com/
sqlite-users mailing list

Reply via email to