Since 4.8 release, the C++ compiler GCC (the G++ part of it) is written not in C anymore, but in C++ itself. I have a hypothetical question on this.
I wonder how to compile the C++ code of GCC on a new platform that has no C++ compiler yet. Of course, you could use prebuilt binaries compiled on other machines. Or you could use an older version of GCC that was written in C and compile the current version with it.
However, without prebuilt binaries and just the newest version, you were stuck, right? If not, are there other implications on this situation raised by the switch from C to C++ of the GCC project?
7
This is actually a well-known concept called bootstrapping. Basically, there exists, somewhere, a minimal C codebase to build a version of GCC that’s capable of building the current GCC codebase. Self-hosting languages have been doing things like that for decades.
3
Creating a compiler that is written in the same language that it compiles is called bootstrapping. The wikipedia article describes a number of ways that a compiler can be bootstrapped.
Given your restriction that you only have a post-4.8 G++ source code and no pre-built binaries for your target platform (no existing C++ compiler), then bootstrapping the G++ compiler can be done by means of cross-compilation.
When bootstrapping a compiler using cross-compilation, you build several versions of your compiler
- On your PC, you install a C++ compiler (can be any C++ compiler, doesn’t have to be G++)
- Using that compiler, you create a G++ cross-compiler that can execute on the PC and generates code for the target platform
- Using the G++ cross-compiler you just built, you create a native G++ compiler that can run on the target platform and create code for it.
- You are done. You have created a C++ compiler for the new platform.
If you also don’t have a PC (or similar) to perform the initial steps on, then you are indeed stuck, but the chance of being in that situation and trying to bootstrap a compiler are negligible.
What would happen if some virus managed to delete all existing compiled C++ compilers, including all backups, and all C++ compilers were written in C++? We would still have source code to C++ compilers.
We would have a problem. We couldn’t compile any C++ code whatsoever because we have no compiled C++ compilers. We couldn’t build any of the C++ compilers that we have source code for because they are all written in C++. What would we do?
We would look for the best compiler that we can find. Maybe we find a compiled C or C# compiler. We could use that to compile and run C or C# code. We would then take our C++ source code for a C++ compiler, translate it manually to C or C#, compile it, and use that to compile the C++ source code.
We would have left out any code for optimising C++ code, or for compiling C++ features not used by the C++ compiler in our C or C# version, so the result wouldn’t be a good or complete C++ compiler. But it would be good enough to compile our C++ compiler, so we’d have a slow (unoptimised) compiler for the complete language. We compile our C++ compiler source code again with this version, and now we have our original compiler back.
1