[compiler-v2] Disassembler doesn't work with source map constructed by the compiler

[wrwg]

The source map generation introduced in #12534 works fine with the prover but crashes with the disassembler. The problem can be reproduced by searching for the bug number in model.rs, disassemble function.

[wrwg]

Does this bug appear in aptos move disassemble? If so it is urgent.

[brmataptos]

I'm not sure if this appears in aptos move disassemble but if we try to run move-cli tests with MOVE_COMPILER_V2=true we hit this failure. So it's likely we'll see the same problem in Aptos. This at least needs more investigation.

[brmataptos]

@wrwg Do you recall where/how you saw this bug?

[brmataptos]

I'm not sure if the problem I just found and fixed is the one which Wolfgang mentioned originally, as I see an error with disassembly which occurs either with the "dummy" source map

        let smap = SourceMap::dummy_from_view(&view, self.env.to_ir_loc(&self.get_loc()))
            .expect("source map");

or the local source map:

        let smap = self.data.source_map.as_ref().expect("source map").clone();

When I run UB=1 MOVE_COMPILER_V2=true cargo test -p move-cli I see problem with disassemblies in various tests, for example third_party/move/tools/move-cli/tests/build_tests/build_with_warnings/args.exp which yields an error instead of a disassembly with compiler-v2:

Error: Unable to get type for local at index 0

The problem is that FunctionGenerator::temp_to_local (in move-compiler-v2/src/file_format_generator/function_generator.rs) is used to allocate both parameters and locals, but as a side-effect it calls self.gen.source_map.add_local_mapping(...), which should only be called for locals, as parameters are recorded elsewhere. Adding a flag to temp_to_local() to change the behavior in the case where it is used to allocate parameters fixes the problem.

I have a fix I will can clean up and send out after I finish some other issues. Or you can try cherry-picking it from my work-in-progress branch https://github.com/aptos-labs/aptos-core/tree/brm-issue-11146-step3. (Note that branch has some other problems with publishing deps in tests, which I'm still working out.)

[brmataptos]

Btw, if we could publish some more precise instructions for replicating the original issue that would be good.

[brmataptos]

@wrwg please check whether this fixes the original problem you saw.

[wrwg]

We need to test whether aptos move disassemble works, if so, we are good.

[brmataptos]

It looks like aptos move disassemble works on hundreds of .mv files (built e2e-move-tests with MOVE_COMPILER_V2=true).

So I think this is fixed.

[brmataptos]

Fixed.