On Thu, 18 Jun 2009, J. Lefebvre wrote:
Hi,
> 2009-06-18 21:18 UTC+0200 Jean lefebvre (jfl at mafact dot com)
> * vm/runner.c
> + Added options to HB_HRBLOAD() :
> HB_HRB_DEFAULT 0 // do not overwrite any functions, ignore
> // public HRB functions if functions with
> // the same names already exist in HVM
> HB_HRB_KEEP_LOCAL 1 // do not overwrite any functions
> // but keep local references, so
> // if module has public function FOO and
> // this function exists also in HVM
> // then the function in HRB is converted
> // to STATIC one
> HB_HRB_KEEP_GLOBAL 2 // overload all existing public functions
> * vm/hvm.c
> + added function hb_vmSetFunction( PHB_SYMB pOldSym, PHB_SYMB pNewSym )
> (Thanks to Przemek)
> * include/hvm.h
> + added function hb_vmSetFunction( PHB_SYMB pOldSym, PHB_SYMB pNewSym )
> (Thanks to Przemek)
Few notes about above commit.
Please try to keep code formatting rules we are using.
No tabs, 3-space indenting, etc.
I haven't expected that you commit this modification to official SVN
repository and I intentionally haven't send hb_vmSetFunction() to this
list but to your private address. I cannot promise that such functionality
will be available in the future and even now it should be updated to work
well with unloaded modules. You made this function public and exported.
The version you committed is not fully functional solution. It simply
tries to block HRB unloading when public function is overloaded (BTW
only tries because it does not work correctly). Such unfinished code
introduce problems for farther modifications when other developers
have to keep it alive and update. So until it will not be fully
functional I would like to keep this modification as undocumented
feature which can be removed by any one working on dynamic code
support in HVM if farther support for it causes any problems.
Otherwise it maybe blocker for farther extensions.
Some small things should be cleaned, f.e. if now we have
HB_HRBLOAD( [ <nOptions>, ] <cHrb> [, <xparams,...> ] )
then in HB_HRBLOAD function instead of:
HB_FUNC( HB_HRBLOAD )
{
ULONG ulLen = 0;
USHORT usMode = HB_HRB_DEFAULT;
USHORT nParam = 1;
char * fileOrBody;
if (hb_pcount()>1)
{
if (HB_ISCHAR(1))
{
ulLen = hb_parclen( 1 );
fileOrBody = hb_parc( 1 );
}
else
{
usMode = (USHORT) hb_parni(1);
ulLen = hb_parclen( 2 );
fileOrBody = hb_parc( 2 );
nParam = 2;
}
}
else
{
ulLen = hb_parclen( 1 );
fileOrBody = hb_parc( 1 );
}
if( ulLen > 0 )
{
PHRB_BODY pHrbBody;
...
we should have:
HB_FUNC( HB_HRBLOAD )
{
USHORT usMode = HB_HRB_DEFAULT;
USHORT nParam = 1;
ULONG ulLen;
if( HB_ISNUM( 1 ) )
{
usMode = ( USHORT ) hb_parni( 1 );
nParam++;
}
ulLen = hb_parclen( nParam );
if( ulLen > 0 )
{
char * fileOrBody = hb_parc( nParam );
PHRB_BODY pHrbBody;
...
This code strictly respects declaration ( does not accept things
like HB_HRBLOAD( date(), cFile ) ) and is simpler and faster.
HB_HRBRUN() parameters should also be updated to respect [<nOptions>, ]
though without clean unload it may introduce problems.
Now inside hb_hrbLoad() we have potential memory leak caused by not freed
pSymFuncExe on some errors. In general this variable can be eliminated
and the code improved. Now it makes unnecessary hb_dynsymFind() calls
for each symbol in HB_HRB_DEFAULT mode. Such improvement should also
fix problem with MT mode introduced by this modification because it
should move symbol updating inside part of code covered by
hb_vmLockModuleSymbols()/hb_vmUnlockModuleSymbols() calls.
best regards,
Przemek
_______________________________________________
Harbour mailing list
Harbour@harbour-project.org
http://lists.harbour-project.org/mailman/listinfo/harbour
