sql - Combining rows without a unique identifier -

i use on developing way combine rows of data when rows don't contain unique identifiers other rows reference. there pri_key, trailing rows never reference key.

disclaimer: have no control on design of table. query i'm running against database of software we've purchased , have in production.

here's sample similar data i'm looking at:

pri_key   |   personid   |   type   |   reference   |   details   |   date_time 101              1                   renewal #001   trans # 001   2014-04-01 09:30:00 102              1            b        modify  #001   expires 30d   2014-04-01 09:30:00 104              1                   renewal #001   trans # 002   2014-04-01 09:30:00 105              1            b        modify  #001   expires 60d   2014-04-01 09:30:00 108              2                   renewal #020   trans # 003   2014-04-21 14:30:00 109              2            b        modify  #020   expires 45d   2014-04-21 14:30:00 111              3                   renewal #140   trans # 004   2014-05-11 12:30:00 112              3            b        modify  #140   expires 90d   2014-05-11 12:30:00 

there hundreds more rows, personid's have 4 rows, others have 2. there's b type row following type row , that's i'm after. can surmise, pri_key increases 1 b type row, there gaps between pri_keys seen in sample. , b have exact same time stamp.

using sample above, here's i'd see:

pri_key   |   personid   |   type   |   reference   |   details   |   date_time         |   modify    |   expires 101              1                   renewal #001   trans # 001   2014-04-01 09:30:00  modify  #001    expires 30d 104              1                   renewal #001   trans # 002   2014-04-01 09:30:00  modify  #001    expires 60d 108              2                   renewal #020   trans # 003   2014-04-21 14:30:00  modify  #020    expires 45d 111              3                   renewal #140   trans # 004   2014-05-11 12:30:00  modify  #140    expires 90d 

essentially, need "expires" information b type's details roll type's row. however, i'm getting caught on peronid's have 2 sets of rows. when run various attempts @ join statements, end doubled expires column:

pri_key   |   personid   |   type   |   reference   |   details   |   date_time         |   modify    |   expires 101              1                   renewal #001   trans # 001   2014-04-01 09:30:00  modify  #001    expires 60d 104              1                   renewal #001   trans # 002   2014-04-01 09:30:00  modify  #001    expires 60d 108              2                   renewal #020   trans # 003   2014-04-21 14:30:00  modify  #020    expires 45d 111              3                   renewal #140   trans # 004   2014-05-11 12:30:00  modify  #140    expires 90d 

i'm thinking need find way use pri_key +1 increments advantage, i'm not sure that's best approach or how handle that. can make suggestions? i'm aware i've written here may make perfect sense me , no 1 else, please let me know if can clarify. thanks!

i've created sql fiddle sample data inserted if it's easier reference that.

the commented "and" clauses shouldn't necessary (if you've described correct) , may make things slower, can include them safe. without commented stuff, may considerably faster comparing based on type/person id, , don't have worry indexing columns:

select a.*, b.reference 'modify', b.details 'expires' my_table join my_table b on a.pri_key = b.pri_key - 1   -- , a.date_time = b.date_time   -- , a.personid = b.personid   -- , b.type = 'b' a.type = 'a' 

do need "modify" column?