* harbour/contrib/hbzlib/zip.c
* harbour/contrib/hbzlib/unzip.c
! fixed possible access to uninitialized memory - thanks to Lorenzo
* harbour/contrib/hbzlib/hbzlib.c
* fixed casting - thanks to Andi
* harbour/contrib/hbzlib/hbmzip.c
* use hb_fsGetOsHandle() instead of LongToHandle() - code which
uses hb_fsGetOsHandle() will work also with our own virtual
handles if we add such extensions in the future
* harbour/source/vm/harbinit.prg
! call __SetHelpK() only when HELP() function/procedure exist
Clipper compatible behavior
* xharbour/contrib/rdd_ads/ace.h
* small modification for OpenWatcom on Linux
* xharbour/source/rdd/dbfntx/dbfntx1.c
* minor fix: use *_GOTO( pArea, 0 ) instead
of *_GOTO( pArea, pArea->ulRecCount+1000 )
+ use hb_fsSeekLarge() - for DBFs longer then 2GB
* xharbour/include/hbrddcdx.h
* xharbour/source/rdd/dbfcdx/dbfcdx1.c
+ use hb_fsSeekLarge() - for DBFs longer then 2GB
* my last xHarbour modifications:
improved indexing speed of DBFs with very big number of
records. In small DBFs less then 100'000 records the new
algorithm can even decrease the speed by ~0.5% due to static
costs but in bigger it becomes faster. For 1'000'000 records
DBFs is about 20% faster, for 10'000'000 records about twice
faster. With bigger DBFs more. Please test it. If you will
have any troubles (I hope not) then please comment out in
dbfcdx.c:
#define HB_CDX_NEW_SORT
I should done it before when I was rewriting the indexing but
sorry I was too lazy. Now I've found a while. Current cost of
key comparisons is always n*log2(n) and cannot be more improved.
Some small optimizations still can be done but the static cost
for small database will be bigger then ~0.5% as in this case
so I do not plan to do it in the nearest future.
* harbour/include/dbinfo.ch
* harbour/source/rdd/dbf1.c
+ use hb_fsSeekLarge() - for DBFs longer then 2GB
* synched with xHarbour (added DBRI_RAW*)
* do not call GOCOLD() inside CHILDSYNC() but do it in FORCEREL()
the modification based on CL5.3 NG.
Below is a note I lef in dbf1.c:
/*
* !!! The side effect of calling GOCOLD() inside CHILDSYNC() is
* evaluation of index expressions (index KEY and FOR condition)
* when the pArea is not the current one - it means that the
* used RDD has to set proper work area before eval, DBFCDX does
* but DBFNTX not yet - it should be changed.
* IMHO GOCOLD() could be safely removed from this place but I'm not
* sure it's Clipper compatible - I will have to check it, Druzus.
*/
/*
* I've checked in CL5.3 Technical Reference Guide that only
* FORCEREL() should ensure that the work area buffer is not hot
* and then call RELEVAL() - I hope it describes the CL5.3 DBF* RDDs
* behavior so I replicate it - the GOCOLD() is moved from CHILDSYNC()
* to FORCEREL(), Druzus.
*/
* harbour/source/rdd/dbf0.prg
* harbour/source/rdd/delim0.prg
* harbour/source/rdd/rddsys.prg
* harbour/source/rdd/sdf0.prg
* harbour/source/rdd/dbfntx/dbfntx0.prg
* formatting (adding EOL at EOF)
* harbour/include/hbapifs.h
* harbour/source/rtl/filesys.c
* added hb_fsLockLarge() and hb_fsSeekLarge() which uses 64bit ofsets
* harbour/source/rdd/workarea.c
* set current work area to the used one in EVALBLOCK RDD method
before code block evaluation
* harbour/source/vm/harbinit.prg
* harbour/source/vm/hvm.c
+ added hb_vmDoInitClip() which calls CLIPINIT function to initialize
ErrorBlock() and __SetHelpK()
Because on some platform the execution order of init functions
is out of Harbour control then CLIPINIT has to be called explicitly
in VM initialization process before hb_vmDoInitFunctions() and do not
depends on INIT clause.
* small modification in VM loop which gives noticeable speed-up
