Adding line numbers – Advanced IR Generation

A debugger allows a programmer to step through an application line by line. For this, the debugger needs to know which machine instructions belong to which line in the source. LLVM allows adding a source location to each instruction. In the previous section, we created location information of the llvm::DILocation type. A debug location provides more information than just the line, column, and scope. If needed, the scope into which this line is inlined can be specified. It is also possible to indicate that this debug location belongs to implicit code – that is, code that the frontend has generated but is not in the source.

Before this information can be attached to an instruction, we must wrap the debug location in a llvm::DebugLoc object. To do so, you must simply pass the location information obtained from the llvm::DILocation class to the llvm::DebugLoc constructor. With this wrapping, LLVM can track the location information. While the location in the source does not change, the generated machine code for a source-level statement or expression can be dropped during optimization. This encapsulation helps deal with these possible changes.

Adding line number information mostly boils down to retrieving the line number information from the AST and adding it to the generated instructions. The llvm::Instruction class has the setDebugLoc() method, which attaches the location information to the instruction.

In the next section, we’ll learn how to generate debug information and add it to our tinylang compiler.