I'm migrating a file translator/mapper program from D7 to XE5, was using 
RXMemoryData but due to lots of unicode compatibility issues, plugged in 
JVMemoryDataset inplace as direct part replacement.

My program deals with processing text files, translation and mapping.
It is ok when running for smaller files (100k to 400k lines).
However when running for larger files(up to 1mil+ lines), the memory usage 
went over the roof and caused application termination due to OOM issue.

I've enabled madExcept's buffer overrun protection and it keep crashing at 
the line when a string dataset field being asssigned with value.

The exception thrown as below:

Access Violation at address 50059F27 in module 'rtl190.bpl'. Read of address 
04C30000

The address where the exception was thrown from is always fixed at 50059F27 
but the memory address being read is different all the time.
With that being said, the file line number being processed at the time the 
AV occurs are also varies and without consistent pattern.
Hence, eventhough the exact line in the code where it fails is 
consistent(when a string dataset field being asssigned with value), the 
iteration it fails are random and difficult to put a breakpoint just before 
it happen to inspect the memory.

Any ideas?

Cheers,
Arif