I'm sorry for so late response but I'm overloaded in work some as at home ... You know the winter is coming

Strange. How strong are you in coding + compiling or in debugging? The easiest way without debugging is to add some more debug logs into the code to see where is the contact dropped. I'm unable to reproduce.

Or are you able to find any difference in this one contact?

Sure. It's normal. PB information is stored and read from evolution-data-server but call history is stored and read from rtcom-eventlogger. So the data are harvested from two different storages which has no conection between them. PB just e-d-s, call history just rtcom-eventlogger.

Hmmm, interesting. I think I need to know the wole record. You can get it using sqlite tools or test tools from rtcom-eventlogger-async Oh I didn't create package for it so I attach gzipped armel binaries

test-get-event-by-id [<id>]
if id is ommited 1 is used (id of the event - not the contact)
- returns int columns: id,service-id,event-type-id,storage-time,start-time,end-time,flags,outgoing
- returns string columns: service,event-type,local-uid,local-name,group-uid,remote-ebook-uid,remote-uid,remote-name,channel

test-get-event-list [<query> [<offset> [<limit>]]]
if query is ommited it returns all entries from rtcom-eventlogger
if offset is ommited it returns entries from first occurence
if limit is ommited it returns all entries from the offset position
- returns int columns: id,service-id,event-type-id,start-time,outgoing
- returns string columns: service,event-type,remote-ebook-uid,remote-uid,remote-name

test-get-event-list-count [<query> [<offset> [<limit>]]]
same as test-get-event-list just returns only count of matched entries

Some queries from obexd (for an inspiration):
ich.vcf:
service-id;1;0;local-uid;"ring/tel/ring";0;event-type-id;1;0;outgoing;0;0

och.vcf:
service-id;1;0;local-uid;\"ring/tel/ring\";0;event-type-id;1;0;outgoing;1;0

mch.vcf:
service-id;1;0;local-uid;\"ring/tel/ring\";0;event-type-id;2;0

cch,vcf:
service-id;1;0;local-uid;\"ring/tel/ring\";0

The query is created from triples: <column-name>;<value>;<compare operation>

All this I get from rtcom-eventlogger sources.
- column-names: src/db.c - EventField fields[] - mapping between DB columns and column-name and also the value type (INT, STR, BOOLEAN)
- compare operation:rtcom-eventlogger/eventlogger-types.h - enum RTComElOp - 0 - equal, 1 - not equal; 2 - greater; 3 - greater or equal; 4 - less; 5 - less or equal; 6 - string is in the list of strings; 7 - string ends with

I don't know the BT protocol and even bluez so good. I also have no idea when it appears in dump. What I associate is that I see some patches for Vendor dependent codecs in upstream bluez. But you have to be me more descriptive (+maybe some debug log) so I can understand and take a look on it).

I didn't get useable debug logs and resp. hcidumps from the user who reported this. So feel free to provide such reports ...

It's not so easy. AT commands are recognized and handled by bluez (bluetoothd) while PBAP, IrMC, MAP, FTP, ... profiles are handled byt obexd. That's also the reason why N900 still doesn't return the adressbook using AT commands.