Files
harbour-core/harbour/include
Przemyslaw Czerpak 8f7e9eda97 2011-12-06 11:42 UTC+0100 Przemyslaw Czerpak (druzus/at/priv.onet.pl)
* harbour/include/hbrdddbf.h
    * modified DB_DBFLOCK_HB64 locking scheme
         - extended RLOCK/FLOCK area from 2^31-2 to 2^32-2
           it effectively changes maximum number of records in this
           locking scheme to 4'294'967'294
         - use COMIX like hyper locking mode
           it should increase performance
      Warning: INCOMPATIBLE!
               all applications using DB_DBFLOCK_HB64 locking mode should
               be updated - it is not safe to concurrently access the
               same files using programs compiled with older Harbour
               versions and current one.
    * renamed the following DB_DBFLOCK_* macros:
         DB_DBFLOCK_CLIP    to DB_DBFLOCK_CLIPPER
         DB_DBFLOCK_CL53    to DB_DBFLOCK_COMIX
         DB_DBFLOCK_CL53EXT to DB_DBFLOCK_HB32
    * added new locking scheme: DB_DBFLOCK_CLIPPER2
      This locking scheme is designed to replicate _real_ locking scheme
      used by Cl*pper applications linked with NTXLOCK2.OBJ
      It's different then the documented one.
    ; Now Harbour supports the following locking schemes:
         DB_DBFLOCK_CLIPPER  1 default Clipper locking scheme
         DB_DBFLOCK_COMIX    2 COMIX and CL53 DBFCDX hyper locking scheme
         DB_DBFLOCK_VFP      3 [V]FP, CL52 DBFCDX, SIx3 SIXCDX, CDXLOCK.OBJ
         DB_DBFLOCK_HB32     4 Harbour hyper locking scheme for 32bit file API
         DB_DBFLOCK_HB64     5 Harbour hyper locking scheme for 64bit file API
         DB_DBFLOCK_CLIPPER2 6 extended Clipper locking scheme NTXLOCK2.OBJ

  * harbour/include/hbrdddbf.h
  * harbour/include/hbrddntx.h
  * harbour/include/hbrddnsx.h
  * harbour/include/hbrddcdx.h
  * harbour/src/rdd/dbf1.c
  * harbour/src/rdd/dbfntx/dbfntx1.c
  * harbour/src/rdd/dbfnsx/dbfnsx1.c
  * harbour/src/rdd/dbfcdx/dbfcdx1.c
    + finished support for COMIX like hyper locking.
      This modification enables periodical switch to write locks
      by readers to eliminate starvation effect on some systems
      where many readers can completely block writer.
      By default readers use exclusive lock once per each 16 operations.
    + added optional support for delayed index write locks (flush locks)
      in hyper locking modes.
      If index RDD can use such locks then write lock can be divided into
      two parts:
         1) block writers and new readers (write lock)
         [prepare index modifications in memory]
         2) wait for old readers still active (flush lock)
         [write modifications to index file]
         [release locks]
      This mode reduces time when index is locked exclusively increasing
      reader performance and also increase writer performance because
      waiting for active readers it can prepare index modifications in
      memory.

  * harbour/src/rdd/dbfcdx/dbfcdx1.c
    + added support for delayed index write locks (flush locks) in DBFCDX

  * harbour/src/rtl/filebuf.c
    * automatically change write file locks to read locks when file is
      open in readonly mode in POSIX systems.

  * harbour/src/vm/classes.c
    * disabled inheriting of class variables values from ancestor classes
      when new class is dynamically created.
      Warning: INCOMPATIBLE!
               This modification is not backward compatible and can
               interact with some existing code which has to be updated
               for new behavior.
2011-12-06 10:42:41 +00:00
..