Parsing metadata names is very costly, to give an idea, for a resultset like a query selecting 100 int, here are some benchmarks :

• solution A : parsing metadata string (table name, table alias, ...) : 6,677.3 operation per second
• solution B : coping buffer metadata part: 8,014.3 ops/s
• solution C : keeping buffer : 8,781.6 operation per seconds.

Most of the time, depending on option, either 0 or 1 of these metadata is needed.
Current implementation is using implementation B, since the report of CONJS-223 (metadata sporadic corruption)

Goal of this task is to investigate better the CONJS-233 issue, in order to permit using solution C.