Different CFG determination

[wideglide]

Any ideas on why I would be getting different generation for the control flow graph of these two functions? These are from two versions of busybox, exported with IDA 8.3 and the f9409a7 commit. I'm not sure where to begin to look for the apparent non-determinism? I haven't reviewed IDA's representation, but I thought BinDiff computed its own control flow.

Here's a second discrepancy from the same diff as well.

[cblichmann]

That indeed looks somewhat surprising. However, BinExport still relies on IDA's representation. In AnalyzeFlow(), we simply follow its internal representation and do some fix-ups. So if IDA has spurious basic blocks, BinExport will have them as well.