* include/hbserial.ch
* src/rtl/itemseri.c
+ added HB_SERIALIZE_IGNOREREF flag.
This flag fully disables logic used to detect multireferences to the
same complex (sub)items like arrays or hashes. It increses the speed
of serialization but serialized data does not contain any information
about refences, i.e. aVal[ 1 ] and aVal[ 2 ] in code below:
aSub := { 1, 2, 3 }
aVal := { aSub, aSub }
are serialized as separated arrays. Additionally items with cyclic
references like:
aSub[ 2 ] := aSub
cannot be serialized at all with HB_SERIALIZE_IGNOREREF flag because
it will create infinite serialization loop and crash with out of
memory message.
* src/rtl/itemseri.c
% rewritten algorithm used to detect cyclic and multi references in
serialized items. The original algorithm has very high overhead when
huge arrays were serialized, i.e. serialization of array with 1'000'000
of subarrays needed about 30 minutes on my i5@3.30GHz. Now it needs
less then a second and this time is only a little bit bigger then
used by serialization with HB_SERIALIZE_IGNOREREF flag.
This modification improve performance also in other code using Harbour
serialization mechanism like I18N files, HBNETIO, GTNET, ... when large
arrays or hashes are saved or transmitted.
* include/hbapi.h
* include/hbapicls.h
* src/vm/arrays.c
* src/vm/classes.c
* src/vm/hashes.c
* src/vm/itemapi.c
* replaced algorithm used to detect cyclic and multi references in
array and hash clone operations with new one similar to current
item serial code. The speed improvement for very large arrays is
the same as in case of serialization code.
* src/rtl/gtsln/mousesln.c
! typo in while loop - synced with Viktor's branch
2e65a28363