* harbour/make_gnu.bat
* harbour/make_gnu.cmd
* harbour/make_gnu.sh
* added information about GTWVT to GT description for Windows
* harbour/contrib/rdd_ads/ads1.c
* harbour/contrib/rdd_ads/adsfunc.c
* harbour/source/rdd/dbf1.c
* synced with xHarbour
* harbour/source/rtl/console.c
! fixed PROW() updating to be Clipper compatible when row number
exceed 16bit integer value
* harbour/source/rtl/filesys.c
! fixed possible GPF when hb_fsCurDirBuff()
* harbour/source/rtl/str.c
* harbour/source/rtl/strzero.c
* minor optimizations
* harbour/source/rtl/strpeek.c
! fixed accessing the last character in STRPEEK() and STRPOKE()
* harbour/include/hbgtinfo.ch
* harbour/source/rtl/gtwin/gtwin.c
+ synced with recent xHarbour modifications
* harbour/source/rtl/gtwvt/gtwvt.c
* cleaned BCC warning
* harbour/include/hbapi.h
* harbour/source/vm/codebloc.c
* declare BYTE * passed to hb_codeblockNew() as const
* harbour/include/hbpcode.h
! fixed the PCODE numbers in PCODE description, 122 was repeated for
two different PCODEs: HB_P_ONE and HB_P_MACROLIST
* harbour/source/compiler/genc.c
! fixed hb_p_functionshort pcode size returned by genc function
! fixed double value presented in verbose genc mode for HB_P_PUSHDOUBLE
! fixed number of cases presented in verbose genc mode for HB_P_SWITCH
* harbour/source/compiler/hbdead.c
+ added support for dead code elimination in functions which uses
SWITCH ...; CASE ...; ...; END syntax
* harbour/source/compiler/harbour.c
* harbour/source/compiler/hbfix.c
! fixed optimization for HB_P_PUSHLOCALNEAR and HB_P_POPLOCALNEAR.
It was a long existing bug not reported so far because 0, 1 and 2
PCODEs uses only one BYTE and bad PCODE tracing in hbfix.c didn't
caused bad side effects as long as number of locals in function
was smaller then 768. Now for functions HB_P_PUSHLOCAL and
HB_P_POPLOCAL is used and later changed to NEAR version in hbfix.c
if possible.
* harbour/source/compiler/hbfix.c
* join sequential jumps to optimize constructions like:
while ...
...
if ...
exit
endif
enddo
* harbour/source/compiler/hbdead.c
* remove HB_P_END PCODE also at the and of function if is repeated
just before, f.e. in code like (source/rtl/color53.prg):
Function RADGRDEFCO( cColor )
if isdefcolor()
Return applydefau( cColor, "W/N", "W/N", "W+/N")
Else
Return applydefau( cColor, 3, 1, 4)
endif
Return nil
Now such redundant constructions like above are fully stripped by
compiler anyhow if someone has time and think is important then can
clean also this .prg code.
* harbour/include/Makefile
* harbour/include/hbcomp.h
+ harbour/include/hbxvm.h
* harbour/source/compiler/Makefile
* harbour/source/compiler/cmdcheck.c
* harbour/source/compiler/genc.c
+ harbour/source/compiler/gencc.c
+ harbour/source/compiler/hblbl.c
* harbour/source/compiler/hbpcode.c
* harbour/source/vm/hvm.c
+ added support for generating real C code not PCODE in .c files.
Now with -gc3 Harbour will generate .c files with real .c code
which is later compiled to real machine code.
I left PCODE only for code block definitions though it's also
possible to have even code block compiled to machine code but
we will have to add a flag to codeblock structure which will
be checked at runtime and proper method of execution in EVAL
will be chosen. If you think it's worth of afford then I can
add it too in some spare time.
The final binaries are noticable longer then the one which has
only PCODE but they are faster. The speed improvement depends
on type of operations. The pure Clipper code which does not
execute any external time consuming C functions compiled with
-gc3 is from 10% to 50% faster.
The side effect of generating pure machine code is also harder
recompilation but it does not mean that is impossible. If sth
can be executed by some machines (virtual or hardware) then it
can always be decompiled by definition.
This modification opens door for farther optimization, f.e. when
we will have fully working strong type support then we can add
using native .c types.
In fact this module is PCODE to .c translater not part of compiler
and can be used as separate module to translate generated PCODE,
f.e. .hrb files.
Maybe in the future we will change the compiler to generate meta
code not directly PCODE and it will be possible to make optimizations
on this meta code which will depend on selected output type:
VHM PCODE, C code, .NET code, etc. but at this moment adding support
for .c compilation directly into compiler will make the compiler code
too much complicated for me - IMHO it's better to invest time into
clear separation to FRONT_END->REAL_COMPILER->BACK_END
Please make test with your code. I rebuild whole Harbour code with
-gc3 set in HB_FLAGS in harbour/config/rules.cf and all seems to
work correctly. hbtest returns exactly the same results.
I also made tests with my programs and they work correctly but
I had to define actions for all PCODEs and it's possible that made
some typos in some of them which are very seldom used so your tests
will be very important.
* make_gnu.bat
! Fixed maker to make -r
* Added %HARBOURDIR% and removed trailing slash from HB_BIN_INSTALL, HB_LIB_INSTALL and HB_INC_INSTALL
* contrib/dot/pp.prg
* Minor corrections to stringifys
* utils/hbtest/hbtest.prg
* Commented out RddSetDefault( "DBFCDX" )
