Invariant violation when dumping C code with anonymous structs or unions

[mstachyraa]

CBMC version: 5.89.0
Operating system: ubuntu-20.04 x86_64 GNU/Linux
Exact command line resulting in the issue: goto-cc test9.c -o test9; goto-instrument --dump-c test9
What behaviour did you expect: Program should generate C code from goto binary
What happened instead: Invariant check failed

Program fails to dump C code from binary with this pattern:

struct struct2 {
    unsigned int data;
};

struct struct1 {
    union {
        struct struct2 type[5];
        struct {
            struct struct2 *pointer; // <- "struct struct2 pointer;" wouldn't fail
        };
    };
};

int main() {
struct struct1 strct;
}

Gdb lookup:

b dump_c.cpp:623
run --dump-c test9
info locals
print comp_type.id()

s = "union { struct struct2 type[5l]; struct anonymous $anon0; unsigned __CPROVER_bitvector[192] $pad; } $anon0"
$2 = (const irep_idt &) @0x55555649efb4: {no = 498}

[mstachyraa]

Update:
Seems like the problem concerns any anonymous structs and unions.

struct struct1 {
    struct { // <- "struct tmp" would have worked
        char a;
        void *addr;
    };
};
int main () { struct struct1 b; }

[thomasspriggs]

I have taken a quick look at this to ensure that it can be reproduced. As the INVARIANT violation is missing I have included it below, for reference of anyone undertaking further work to debug this issue -

--- begin invariant violation report ---
Invariant check failed
File: /home/tspriggs/Development/cbmc/src/goto-instrument/dump_c.cpp:623 function: convert_compound
Condition: false
Reason: Unreachable
Backtrace:

This condition is found here -

cbmc/src/goto-instrument/dump_c.cpp

Line 623 in 8a8ecc8

UNREACHABLE;

I am not entirely sure what the code block beginning on line 549 is attempting to achieve or what the fix should be. I think it is something like an initial attempt is made to serialise the type as string s. If the string contains CPROVER_bitvector then enter special case to print the the type based on the C integer type instead of the internal type. However the CPROVER_bitvector type is the type of a nested field and the special cases matching signed/unsigned bit vectors are mismatched as the type node in the AST in this case is a struct/union.

I am not planning to look at this further for the moment. But I thought my initial findings may be useful when someone continues the debugging effort.

[tautschnig]

@thomasspriggs Thank you for the analysis! Minor correction: the string s should not contain __CPROVER_bitvector to enter that branch. With the anonymous members we might end up with padding in the declaration, and that padding might be of said type. Assigning to myself as I am at fault for most of this code.