fts-lmdb.org

FTS with LMDB

Chunks

Can be very small – currently org elements or lines

Numbers – lexicographically ordered

7 bits	0 - 127	0xxxxxxx
12 bits	128 - 4095	1000xxxx X
20 bits	4096 - 1048575	1001xxxx X X
28 bits	1048576 - 268435455	1010xxxx X X X
36 bits	268435456 - 68719476735	1011xxxx X X X X
44 bits	68719476736 - 17592186044415	1100xxxx X X X X X
52 bits	17592186044416 - 4503599627370495	1101xxxx X X X X X X
60 bits	4503599627370496 - 1152921504606846975	1110xxxx X X X X X X X
64 bits	1152921504606846976 - 18446744073709551615	1111---- X X X X X X X X

Trees

Grams: GRAM-> BLOCK

GRAM is a 2-byte value

OID LIST

Gram 0 holds the info since 0 is not a legal gram

next unused oid
start of free list
group name …

Chunk: OID -> BLOCK

OIDS are compressed integers

Chunk entries are <first tag Y, data> first tag Y is the start of the threaded tag list data is arbitrary data for the object

GID
data (e.g. line number)
totalGrams

Groups: GID -> BLOCK

Used for bookkeeping, to be able to remove chunks

NAME

Group Names: NAME->GID

OID LISTS

# 1-byte OIDS
# 2-byte OIDS
# 3-byte OIDS
# 4-byte OIDS
# 5-byte OIDS
# 6-byte OIDS
# 7-byte OIDS
# 8-byte OIDS
# 9-byte OIDS
OIDS