problem about ground truth collection

[0xling]

Nice work.
However, can the grounp truth collection script be open source?

[0xling]

And I use deepbindiff to run CGC dataset, the command is like this:

python3 src/deepbindiff.py --input1 all_unpached/CROMU_00002 --input2 all_patched/CROMU_00002_patched --outputDir output

but the result matched pairs is empty in some binaries, for example CROMU_00002/CROMU_00004/CROMU_00005/CROMU_00006.

Is it normal? Is there any method to improve the result of deepbindiff?

the CGC dataset can be get from https://github.com/zardus/cgc-bins.

[yueduan]

Hi, unfortunately, I accidentally deleted my code for ground truth collection. But it is fairly easy to implement. I used myers algorithm for source code mapping, and leveraged objdump to dump all the debugging info. Then you need to write a very simple parser to parse the debugging info and find the address to line number mapping. Eventually, you can combine the source code mapping and the address to line number mapping to create ground truth.

[yueduan]

Let me take a look at the binaries in CGC and get back to you.

[AdityaTakkar]

We have some question regarding this code:

1. Do we have to use angr or objdump for disassembly? Also angr is producing very short codes and also similar codes for different exe as inputs.
2. Please provide a workflow, or the idea around the implementation to find ground truth
3. Explain how the debugging information is useful and how do we use it in the code?