ACE VERSION: 5.6 (and almost certainly 5.6.6 - the code is identical)

     HOST MACHINE and OPERATING SYSTEM:
         AMD Phenom 9950 Quad-Core - Windows XP

     TARGET MACHINE and OPERATING SYSTEM, if different from HOST:
         (same as host)

     THE $ACE_ROOT/ace/config.h FILE :
        config-win32.h

     AREA/CLASS/EXAMPLE AFFECTED:
        ACE_OS::mmap()

     DOES THE PROBLEM AFFECT:
         COMPILATION? No
         LINKING? No
         EXECUTION? Yes

     SYNOPSIS:
        ACE_OS::mmap() fails when called with file_handle=ACE_INVALID_HANDLE
and file_mapping=0.  The intent is to create named shared memory using the
system paging file.

      DESCRIPTION:
        Calling ACE_OS::mmap() with file_handle=ACE_INVALID_HANDLE and
 file_mapping=0 will cause file_mapping to be set internally to the
 address of local_handle whose value is ACE_INVALID_HANDLE.  This results
in a call to ACE_TEXT_CreateFileMapping() with both dwMaximumSizeHigh and
dwMaximumSizeLow set to 0.  This is fine when file_handle is valid - the
maximum size of the file mapping object will be equal to the current size of
the file identified by file_handle.  However it is legal to specify an invalid
file_handle in order to create named shared memory using the system paging file
- and in this case, an error is returned if both maximum size high and low are
set to 0.

     REPEAT BY:
     void *imageBuffer =
         ACE_OS::mmap(0,  // addr
                      imageBufferSize, // len            
                      PAGE_READWRITE,  // prot            
                      MAP_SHARED, // flags           
                      ACE_INVALID_HANDLE, // file_handle
                      0,  // off
                      0, // file_mapping
                      0, // sa 
                      sharedMemory.c_str())); // name of shared memory

     SAMPLE FIX/WORKAROUND:
        Set dwMaximumSize[High|Low] to the desired length of shared memory when
fle_handle == ACE_INVALID_HANDLE. (See lines marked FIX).

   // FIX: Move these up from below
   DWORD low_off  = ACE_LOW_PART (off);
   DWORD high_off = ACE_HIGH_PART (off);

   // Only create a new handle if we didn't have a valid one passed in.
   if (*file_mapping == ACE_INVALID_HANDLE)
     {
       ...

       *file_mapping =
           ACE_TEXT_CreateFileMapping (
                file_handle,
                attr,
                prot,
                (file_handle == ACE_INVALID_HANDLE) ? high_off : 0, // FIX
                (file_handle == ACE_INVALID_HANDLE) ? low_off  : 0, // FIX
                file_mapping_name);
     }

Could you add or extend one of the ACE tests under ACE_wrappers/tests as
automated regression test 

Sorry, I blew it with the bit about moving the offset variable initializations
up:

   // FIX: Move these up from below
   DWORD low_off  = ACE_LOW_PART (off);
   DWORD high_off = ACE_HIGH_PART (off);

Leave them where they are and use len, not off:

       *file_mapping =
           ACE_TEXT_CreateFileMapping (
                file_handle,
                attr,
                prot,
                0
                (file_handle == ACE_INVALID_HANDLE) ? len : 0, // FIX
                file_mapping_name);

So it's really just a one=-line fix.

Fixed, thanks!

Created an attachment (id=1053) [details]
Regression test

I'm not sure if I need to attach the modified tests.mpc and run_test.lst as
well.
I'm about to submit a regression test for Bug# 3541.  I'll attach those files
modified for both bugs to that report.

reopen

Marcel, can you integrate this test into the svn repository

test is in the repository and test stats show that the regression test works
out of the box