C or C++ to write a compiler?

Question

I want to write a compiler for a custom markup language, I want to get optimum performance and I also want to have a good scalable design.

Multi-paradigm programming language (C++) is more suitable to implement modern design patterns, but I think that will degrade performance a little bit (think of RTTI for example) which more or less might make C a better choice.

I wonder what is the best language (C, C++ or even objective C) if someone wants to create a modern compiler (in the sense of complying to modern software engineering principles as a software) that is fast, efficient, and well designed.

Answer 1

Forget what programming language you use & also given that you have huge memory support in these modern computer era you could write good & fast programs using interpreted language and also very bad & slow running programs using C/C++ (compiled languages) & vice versa.

What is important is to use right data structures and algorithms & follow the style/patterns of the programming language you use to implement it. Remember that some one said "OO is not a panacea" & to the other extent some one else also said "show your data structures and I will code up the algorithm for the problem you are trying to solve".

Answer 2

I dont have any links but from what i hear and from experience C/C++ is a poor language to write a compiler with. First of all, do you really honestly need it to be scalable? Or scalable at this stage? Especially for a markup language? your not compiling 60+ mb of source so i dont think you actually need it to be scalable.

Anyways for my programming language i used bison for the parser (reading bison+flex is a must, try to avoid all conflicts my language has none). Then i use both C and C++ for the code. C because bison uses C and i just call a simple C function which creates and fill in a struct to create an abstract syntax tree. Then when its done it calls my C++ code that runs through the AST and generate the binary.

Standard ML is suppose to be really good with creating a language. If you dont use that a functional language is a good choice because it fits with the mindset (parsing may be left to right but your function calls wont be in that order). So i recommend that if you dont use bison (or know how to call it using C/C++ and bison).

Note: I tried writing a compiler twice. The first time in C without bison the 2nd time with bison. Theres no question that it would have taken me exponentially longer due to the fact that bison finds the conflicts for me and i am not doomed in debug land (i would probably in fact try to figure out a way to report conflicts before i write the code which is exactly what bison does)

Answer 3

People who write compilers in C as their basic language usually have the good sense to use tools for certain parts of it.

Specifically, go find out about lex and yacc (in their free implementations, flex and bison).

This advice almost certainly applies to any other language you choose, be it C++, Java or whatever.

Answer 4

The "expensive" features of C++ (e.g., exceptions, virtual functions, RTTI) simply don't exist in C. By the time you simulate them in C, you're likely to end up with something at least as expensive as it is in C++, but less known, less documented, etc. (let's face it: compiler writers aren't stupid -- while it's possible you can implement a feature "better" than them, it's not really particularly likely).

In the other direction, templates (for one example) often make it relatively easy to write code that is considerably faster than is practical in C. Just for one obvious example, C++ code using std::sort will often be two to three times as fast as equivalent C code using qsort.

Bottom line: the only reason for a C++ program to be slower than an equivalent written in C is if you've decided (for whatever reason) to write slower code. Common reasons are simplicity and readability -- and in most cases, those are more important than execution speed. Nonetheless, using C++ doesn't necessarily carry any speed penalty. It's completely up to you to decide whether to do something that might run more slowly.

Answer 5

Have you considered OCaml? Functional languages are well-suited for compiler writing. Pattern matching is an extremely useful construct, and the lack of side effects will make parallelization easy.

OCaml can be compiled to native code, and its performance is comparable to C and C++. Its standard library is somewhat lacking, but you don't really much else to write a compiler.

F# is a very similar language if you prefer a .NET environment.

Answer 6

C++ adheres to a "pay only for what you use" policy. You are not going to see performance hits due to the language choice; the performance of your application will be purely dependent upon your implementation.