Improve debuggability of LLVM library

When the LLVM AST we have generated is incorrect, this usually results in either (i) a segmentation fault when writing the bitcode/readable IR to a file or (ii) a bitcode file that cannot be parsed, and therefore cannot be disassembled to a readable ll file. In such cases it could be useful to get some feedback. This is purely meant to make developing the code generator easier: the goal is to get more information why the newly generated IR is incorrect, not to assist the end user in any way.

I stumbled across two bits of functionality that can be helpful:

1. LLVMVerifyModule/LLVMVerifyFunction
2. Converting functions to dot graphs

1 can be used like this (though in this case, LLVMPrintModuleToFile has no issues):

#include <llvm-c/Core.h>
#include <llvm-c/Analysis.h>
#include <llvm-c/BitWriter.h>

#include <stdio.h>

int main(int argc, char const *argv[]) {
    LLVMBuilderRef builder = LLVMCreateBuilder();
    LLVMModuleRef mod = LLVMModuleCreateWithName("my_module");

    LLVMTypeRef ret_type = LLVMFunctionType(LLVMInt32Type(), NULL, 0, 0);
    LLVMValueRef f = LLVMAddFunction(mod, "f", ret_type);
    LLVMBasicBlockRef entry = LLVMAppendBasicBlock(f, "entry");
    LLVMPositionBuilderAtEnd(builder, entry);
    LLVMBuildRet(builder, LLVMConstInt(LLVMInt32Type(), 0, 1));
    LLVMBuildRet(builder, LLVMConstInt(LLVMInt32Type(), 0, 1));

    char *error = NULL;
    LLVMVerifyModule(mod, LLVMPrintMessageAction, &error);
    LLVMDisposeMessage(error);

    if (LLVMWriteBitcodeToFile(mod, "x.bc") != 0)
        fprintf(stderr, "error writing bitcode to file\n");
    if (LLVMPrintModuleToFile(mod, "x.ll", NULL) != 0)
        fprintf(stderr, "error writing module to file\n");

    LLVMDisposeBuilder(builder);
}

2 is available in opt with opt-20 -passes='print<memoryssa>' -dot-cfg-mssa=x.dot x.bc >/dev/null, or opt-20 -passes=dot-cfg x.bc >/dev/null and the various alternatives listed there. These must be available from the C API too, but I haven't looked into it.

Just something to be aware of that it's possible and consider implementing the next time something like this happens.