[gdal-dev] GDAL reference counting is thread unsafe

[Tamas Szekeres]

Martin,

Seems reasonable, however I expect we could avoid using critical
sections in the windows code in this particular case using the
Interlocked function would be quite enough.
I'm uncertain how the windows headers are included in your code, but
it looks like simply including <windows.h> in some places may
interfere with other
declaration in GDAL in other places easily.

Best regards,

Tamas


2008/11/2 Martin Chapman <mchapman at fortifieddatacom.com>:
> If anyone is interested the following code has been used safely on Windows
> and Linux for thread safe reference counting.  You can paste it into a
> header file if you want, or use it as a guide.  It uses pthreads on Linux
> and the Win32 thread functions on Windows with InterlockedIncrement() and
> InterlockedDecrement().
>
> Best regards,
> Martin
>
>
> #ifndef Increment_h
> #define Increment_h
>
> #define _WINDOWS_
>
> #ifndef _WINDOWS_
> #include <pthread.h>
> #include <bits/pthreadtypes.h>
> #endif
>
> class CIncrement
> {
> public:
>
>        #ifdef _WINDOWS_
>        CRITICAL_SECTION m_CS;
>        #endif
>
> CIncrement()
> {
>        #ifndef _WINDOWS_
>        pthread_mutex_init(&m_Mutex, NULL);
>        #endif
>
>        #ifdef _WINDOWS_
>        InitializeCriticalSection(&m_CS);
>        #endif
>
>        #ifdef _WINDOWS_
>        EnterCriticalSection(&m_CS);
>        #endif
>
>        m_nVal = 0;
>
>        #ifdef _WINDOWS_
>        LeaveCriticalSection(&m_CS);
>        #endif
> }
>
> ~CIncrement()
> {
>        #ifndef _WINDOWS_
>        pthread_mutex_destroy (&m_Mutex);
>        #endif
>
>        #ifdef _WINDOWS_
>        DeleteCriticalSection(&m_CS);
>        #endif
> }
>
> CIncrement& operator=(const CIncrement& thread)
> {
>    if (this == &thread)
>        return *this;
>
>        #ifndef _WINDOWS_
>    pthread_mutex_lock(&m_Mutex);
>    #endif
>
>        #ifdef _WINDOWS_
>        EnterCriticalSection(&m_CS);
>        #endif
>
>    m_nVal = thread.m_nVal;
>
>        #ifdef _WINDOWS_
>        DeleteCriticalSection(&m_CS);
>        #endif
>
>    #ifndef _WINDOWS_
>    pthread_mutex_unlock(&m_Mutex);
>    #endif
>
>    return *this;
> }
>
> int operator++(int nVal)
> {
>    return Add(1, true);
> }
>
> int operator++()
> {
>    return Add(1, false);
> }
>
> int operator--(int nVal)
> {
>    return Add(-1, true);
> }
>
> int operator--()
> {
>    return Add(-1, false);
> }
>
> bool operator==(const int nVal)
> {
>    #ifndef _WINDOWS_
>    pthread_mutex_lock(&m_Mutex);
>    #endif
>
>        #ifdef _WINDOWS_
>        EnterCriticalSection(&m_CS);
>        #endif
>
>    bool bReturn = (m_nVal == nVal);
>
>        #ifdef _WINDOWS_
>        LeaveCriticalSection(&m_CS);
>        #endif
>
>    #ifndef _WINDOWS_
>    pthread_mutex_unlock(&m_Mutex);
>    #endif
>
>    return bReturn;
> }
>
> int Add(int nNum, bool bReturnNew)
> {
>    int rtn = 0;
>    #ifdef _WINDOWS_
>    return nNum > 0 ? InterlockedIncrement((volatile LONG*) &m_nVal)
>                                        : InterlockedDecrement((volatile
> LONG*) &m_nVal);
>
>    #else
>    pthread_mutex_lock(&m_Mutex);
>    int oVal = m_nVal;
>    m_nVal += nNum;
>    rtn = bReturnNew ? m_nVal : oVal;
>    pthread_mutex_unlock(&m_Mutex);
>    #endif
>
>    return rtn;
> }
>
> protected:
>        int m_nVal;
>
>        #ifdef _WINDOWS_
>        void* m_Mutex;
>        #else
>        pthread_mutex_t m_Mutex;
>        #endif
> };
>
> #endif
>
>
>
>
>
>
>
>
>
>
>
> -----Original Message-----
> From: gdal-dev-bounces at lists.osgeo.org
> [mailto:gdal-dev-bounces at lists.osgeo.org] On Behalf Of Tamas Szekeres
> Sent: Saturday, November 01, 2008 5:40 PM
> To: gdal-dev at lists.osgeo.org
> Subject: [gdal-dev] GDAL reference counting is thread unsafe
>
> Folks,
>
> I've found the following issue with the current refcounting
> implementation in GDAL:
> http://trac.osgeo.org/gdal/ticket/2648
>
> Do you have idea whether we could handle this problem possibly for the
> next release?
>
>
> Best regards,
>
> Tamas
> _______________________________________________
> gdal-dev mailing list
> gdal-dev at lists.osgeo.org
> http://lists.osgeo.org/mailman/listinfo/gdal-dev
>
>
>