Problem wit a thread in loop

[TouEnMasm]

Hello,
I have a problem with a thread used in loop.
One used only one time,it work.
One used in loop there is some failed.
What is the best memory function to use in a thread ?.

[Tedd]

HeapAlloc without the HEAP_NO_SERIALIZE flag will avoid problems caused by multiple threads trying to allocate from the same heap at the same time -- if that is your problem.

Use of a mutex/semaphore may be appropriate, depending what you're doing.

[TouEnMasm]

I use Heapalloc like this:

invoke HeapAlloc,dword ptr [esi][MEMHEAP.Hheap],HEAP_ZERO_MEMORY,edx

i run it:

            invoke CreateThread,NULL,NULL,Thread_BoucleTraduction,HwndProc,NULL,addr thread.threadid
            mov thread.Htread,eax

And the thread:

Code Select Expand


Thread_BoucleTraduction PROC Hfenetre:DWORD
Local  retour:DWORD
mov retour,1
        invoke working functions;-------
FindeThread_BoucleTraduction:
invoke CloseHandle,thread.Htread
mov thread.Htread,0
mov thread.threadid,0
invoke PostMessage,HwndProc,WM_COMMAND,IDM_TRADUIRE6,NULL ;next call for thread

mov eax,retour
ret
Thread_BoucleTraduction endp


[TouEnMasm]

It seems I have found the problem (Heap fragmented).

   invoke CloseHandle,thread.Htread
   mov thread.Htread,0
   mov thread.threadid,0
   mov edx,NomdesDword.Hheap               ;get a handle heap in use
   .if edx != 0
      invoke HeapCompact,edx,0             ;-------------------- and all is normal ---------
   .endif

[adeyblue]

I think Tedd was also referring to the race conditions you have. It's entirely possible for the CloseThread in Thread_BoucleTraduction to happen before the main thread assigns the handle to thread.Hthread, which would leak the handle. It'd be even worse if you create lots of threads in a loop.

[hutch--]

Yves,

If you need to repeatedly call a procedure that is pointer to by creating a new thread, you need some method to delay tyhe caller until the identification of any data sent to the new thread handling procedure is completed. I personally use an operating system type of spinlock, a closed loop that exits when a value is changed in the target procedure and the idea is to hold up the creation of a new thread until the last one is up and going.

Have a look at a masm32 example in "example\threads\multidl".

[dedndave]

i doubt that i'd allocate memory in the thread code like that
allocate a single block, and pass a pointer to the indexed buffers

[jj2007]

invoke HeapCompact,edx,0             ;-------------------- and all is normal ---------

... and that is a recipe for disaster. Heap fragmentation is a condition that *may* happen after you shuffled around megabytes of little strings, and even then, it rarely causes problems unless you are tight with RAM. So HeapCompact accidentally "solves" something with a totally different cause.

Depending on what your threads do, the spinlock thing proposed by Hutch could be a solution, but we should see the complete code before.

[TouEnMasm]

I will study the example\threads\multidl
The effect of HeapCompact can be tested with headinc.exe
http://masm32.com/board/index.php?topic=576.msg4665#msg4665