I have a TensorFlow sample application in Island/Oxygene - and somehow I ended up in an exception which gave me a Window error code indicating heap corruption.
I am scratching my head. Maybe this is a stupid question - will by any chance Island memeory manager interfere with tensorflow’s C memory manager?
No the GC has it’s “own” memory (it uses virtualalloc to allocate a buffer for GC memory).
Given that I know nothing about tensor flow, can you elaborate a bit as to what fails and where? (and if it’s needed for me to run this, what I need to run this?)
The github repository includes a list of Elements/Island projects. If you could clone that repository, and run the project: ./projects/TensorInfo, you would see the exception - it was caught by Element, and the error code indicates it was a Windows Heap Corruption error.
@ck@mh
Thank you - I download the 2465 build. The “heap corruption” issue is solved, but the VS2019 debugger (Windows/Island/x64) seems stopped working. Previously, the debugger at least worked to some extent.
The debugger works as expected for non-tensorflow projects (i.e., when there is no linking to TensorFlow), but
The debugger stops working if the program is linked to TensorFlow. By “stops working”, I mean, it won’t hit ANY breaking points, while just go all the way to run and exit (as if the “Start without Debugging” mode).
Any heads up?
UPDATE: it seems build 2465 has debugger totally broken, not working on VS2019/Water. It has nothing to do with whether linking to TensorFlow or not.