TensorFlow & python3 support

[onyx22574]

TensorFlow & python3 support

[brunoalr]

Hi @onyx22574, first you need to implement some interfaces to correctly parse the image class(es) you want to plot, see this post here https://csantosbh.wordpress.com/2016/10/15/configuring-gdb-imagewatch-to-visualize-custom-buffer-types/. Long story short, OID needs to know which members of a class it should look for to get information such as width, height, stride etc.

I don't know how these libraries encapsulate images, but if they use OpenCV or Eigen, then we have out-of-the-box support to them.

After that, in theory you can attach gdb or lldb to the python interpreter itself and put breakpoints in the native part of TensorFlow/PyTorch. I haven't tested that myself, though. @csantosbh have you ever tried it before?

[brunoalr]

You can also check how this is done for opencv types here: https://github.com/OpenImageDebugger/OpenImageDebugger/blob/main/resources/oidscripts/oidtypes/opencv.py

[brunoalr]

tensorflow/tensorflow#50443

[brunoalr]

@onyx22574 just to be clear: if you are asking for pdb support, we currently do not have it. But that's a very nice idea, actually. This is going to be considered for implementation :)