* contrib/hbnf/fttext.c
* contrib/hbnf/hbnf.hbx
* synced with Viktor's modifications:
- use FILE API instead of FS API
- protection against GPF when FT area has no valid file handle
- alternative function names which exceeds 10 character Cl*pper limit
! fixed handle and memory leak when new file is open in FT area which
already has open file
* include/harbour.hbx
* src/harbour.def
* src/vm/hvm.c
+ added new PRG function: __vmItemRefs( <xVal> ) -> <nRefs>
Please remember that passed <xVal> parameter increases the counter so
if you want to eliminate it then pass it by reference, i,e.:
aVal := {}
? __vmItemRefs( aVal ), __vmItemRefs( @aVal )
* include/hbapi.h
* src/harbour.def
* src/vm/arrays.c
* src/vm/codebloc.c
* src/vm/hashes.c
+ added new C functions:
HB_COUNTER hb_arrayRefs( PHB_ITEM pArray );
HB_COUNTER hb_hashRefs( PHB_ITEM pHash );
HB_COUNTER hb_codeblockRefs( PHB_ITEM pItem );
* src/rtl/itemseri.c
* added new internal function to replace 3 times repeated code for
resizing array of references.
% use new functions to check number of references to hashes, arrays and
objects to ignore in system detecting multiple references the items
which have reference counter smaller then 2. Such items cannot have
multiple references.
In practice this modification resolves the time problem when very big
items are serialized to the level comparable to HB_SERIALIZE_IGNOREREF
so now this flag is useless though I'll keep it because only in such
mode serialization code can be used as filter in streams.
I also created binary tree to store references but after this
modifications it is not necessary so I decided to not commit it yet.
Please only remember that practice limit of such serialization code
is created by physical memory attached by OS to the process. When this
limit is exceed then swap is actively used what completely kill the
performance. The CPU usage is reduced to less then 1% because it waits
for restoring swapped memory pages by extremely heavy used disks. In
such case I cannot help.
2b4c4b3b5f