Avoid creating an empty if block

[rfalke]

Version: web version at https://retdec.com/service/api
File: https://github.com/rfalke/decompiler-subjects/tree/master/from_holdec/condition_structure/ia32_elf
Output:

// Address range: 0x8048437 - 0x8048465
int32_t test_1_blocks_1_targets_direct_variant_0(void) {
    // 0x8048437
    if (g2 == 0) {
        // 0x8048437
        // branch -> 0x8048439
    } else {
        // 0x8048457
        puts("then/else block 3");
        // branch -> 0x8048439
    }
    // 0x8048439
    puts("return block");
    return 0;
}

Expected:

The condition should be inverted and there should only be the then-block without else.

[s3rvac]

Thank you for the report. I confirm that the empty body of the if statement is also generated by the current master (8c4b23d).

[PeterMatula]

Output by the current master, after the new LLVM IR to BIR convertor became the default convertor in llvmir2hll:

// Address range: 0x8048437 - 0x8048466
int32_t test_1_blocks_1_targets_direct_variant_0(void) {
    // 0x8048437
    if (g4 != 0) {
        // 0x8048457
        puts("then/else block 3");
    }
    // 0x8048439
    puts("return block");
    return 0;
}

I can't say empty if bodies are never going to be generated, but the current output (all the functions) look quite ok at the moment - nice test binary by the way. If you come across another issue, open an issue :-)
I added regression test (avast/retdec-regression-tests@82b3735) for the reported problem, but unfortunately I did not came up with a way to test all the control-flow structures in all the functions. The test itself is quite ugly - hard-coded string to check for. I don't think we have a better mechanism to check for things like this, and it would be a lot of work anyway, since there is a huge number of test cases in the subject.exe sample.