Function are split into smaller parts

[google]

Function are split into smaller parts

[cblichmann]

Hi there!

Thanks for your report. A similar issue seems to occur in b/151242195 . See there for more details.

TL;DR - this is likely or heuristic for detecting non-returning functions.

The heuristic is implemented around here:

binexport/flow_graph.cc

Line 92 in a156e27

// Find non-returning calls. We simply assume any call followed by an invalid

Another related issue is b/151859042.

As optimizations become more aggressive, we seem to trigger this case more often than I would like.
While this is working as intended, the intention is by now at least outdated ;-) -- marking this as a bug.
If you have good ideas on how to improve the heuristic, I'm all ears :-)

[cblichmann]

Also happens in IDA 7.4 and BinExport 11. Both BinDiff 5 and 6.

Another workaround might be to use the Ghidra exporter (also in this repo or in the release version of BinDiff 6). That one does not have the heuristic and relies exclusively on the disassembler to figure out non-returning functions.

[psylocn]

Thanks, the workaround with -OBinExportX86NoReturnHeuristic:FALSE helped me alot!

Sorry, I do not know how to improve the heuristic. But why not let IDA figure out how to deal with non-returning functions? Similar to Ghidra :)

[cblichmann]

[...]
Sorry, I do not know how to improve the heuristic. But why not let IDA figure out how to deal with non-returning functions? Similar to Ghidra :)

Yeah that will likely be the solution. The heuristic is there in the first place, because BinExport is basically a full disassembler and only uses IDA Pro to provide hints where the initial entry points are (and as instruction decoder).

[cblichmann]

BinDiff 7 sets -OBinExportX86NoReturnHeuristic:FALSE by default. We should still propagate any "No Return" attribute that a disassembler sets.