Board index » cppbuilder » Re: all BCB6 apps: memory corruption in hyperthreading / dualCPU machines

2005-10-26 01:11:40 AM
cppbuilder43
Zach Saw wrote:
with that, I create 8 threads, and run DataSet->Open() and
DataSet->Close() over and over again.
Could you clarify that? It's still not clear whether you are doing
something which one might reasonably expect requires synchronization,
such as using a copies of an AnsiString in multiple threads (with the
associated copy on write problems), etc.
Tom
 

Jonathan Benedicto wrote:
copies of the same string in multiple threads, you might have a problem.
Calling Unique() after making the copy might solve the problem.
Tom
 

"Zach Saw" < XXXX@XXXXX.COM >wrote in message
internally alter the contents of the AnsiString without you knowing it.
In any case, when sharing a global across threads, especially a data type
that is not a POD type (integers, booleans, etc) then you should ALWAYS
serialize access to it.
Gambit
 

{smallsort}

Tom Widmer wrote:
among several threads, just to make sure.
Jonathan
 

well, each thread has its own DataSet -- so I'm only opening and closing the
dataset in their own threads... so there're no sharing of AnsiString across
threads, unless the code in VCL tries to be funny... :)
"Tom Widmer" < XXXX@XXXXX.COM >wrote in message

the global variable is of type integer...
"Remy Lebeau (TeamB)" < XXXX@XXXXX.COM >wrote in message

"Zach Saw" < XXXX@XXXXX.COM >wrote in message
are accessing it at the same time.
Gambit
 

yeah, but there's nothing writing to the global var...
it's there just for testing memory corruption -- to see if the value it is
initialized with is still the same at the end of the test program :)
"Remy Lebeau (TeamB)" < XXXX@XXXXX.COM >wrote in message

Tom Widmer < XXXX@XXXXX.COM >writes:
both call Unique at the same time, and both modify the same original
string, possibly messing up its refcount.
--
Chris (TeamB);