Crash on accessing symbols by address in Mach-O binaries

[angr]

Crash on accessing symbols by address in Mach-O binaries

[rhelmot]

My input is that a symbol without an address is a concept that CLE handles using the idea of an "import symbol", which comes from ELF binaries. These symbols are given an address of zero by the binary format itself, and cannot be used for anything other than marking a yet-unresolved dependency. This seems like the exact use case you're working with currently, and my recommendation would be to refactor the symbol binding process to use import and export symbols like the rest of CLE, though I don't know anything about mach-o binaries and don't know how well this would work. PE files (windows) don't have an explicit concept of import/export symbols, but we added them in our implementation of the PE loader to simplify the data model.

That actually makes perfect sense, thank you. I'm gonna try to implement that change as soon as I figure out how to properly map Mach-O's symbol concept onto import/export symbols.

Any chance we can get symbols_by_addr and related fields in Backend documented in a bit more detail so it is clear which symbols should be there and which should not? I.e. make the semantics clear which CLE associates with each field.

[rhelmot]

Documentation is an ongoing effort. I sunk a huge amount of time over the summer into rewriting the docs for the angr/simuvex merge, and since I am now back in class I can't really dedicate much time to formal documentation effort.

You shouldn't need to explicitly insert symbols into that dict specifically, since they are automatically added by the Symbol constructor, given that it actually has a nonzero address.

Well that is understandable of course. No worries, I'll make do with reading the source.

Once I can propose a fix I'll let you know.

Finally, PR is up. It passed my internal tests so I hope it won't give you any trouble merging.