Fr / En

Preprocessor madness

As developers, we all have written some code of which we are really proud. Clean, beautiful, intuitive, fast… Some little gems born from our minds, carved out of the most precious ore. We also have, alas, written so much “plumbing” code, barely interesting, sometimes tedious, that we can’t recall half of it.

And then there’s the really ugly code, written at night in dark rooms. The kind that exploits the poor tools that have to process it, the kind that scrambles the brains of the fellow devs who try to understand how the thing can even work, the kind that would summon Cthulhu if read backwards.

Can you guess in which group TOOLS_PP belongs?

Yo, dawg…

It has been once said that all problems in computer science could theoretically be solved by another level of indirection. What kind of problems do we sometimes face when we have to write code? Well, for instance, boilerplate code. How to avoid that? Well, we could add another level of indirection, and create a program that would generate our program; we could write code that would generate all the boilerplate code. This is a known technique; it’s a part of what is called metaprogramming.

While nothing forbids you from rolling your own “C++ Python preprocessor” tool, most languages have a built-in or standard way to meta-program. Languages of the Lisp family, such as Clojure, are famous for their homoiconicity, thanks to which they are their own meta-language. Nimrod and Rust have powerful macro systems, Haskell has the almost-standard Template Haskell, all three of which allow to write meta-code in the same language as the target code.

But while those languages have fancy, checked, proper meta-programming tools, C and C++ users are stuck with a far simpler tool: the dreaded C preprocessor.

A general-purpose text-processing tool?

While macro systems such as Rust’s are hard-wired in the compiler itself, the C preprocessor (hereafter referred to as CPP) is a standalone tool. It has its own “markup” syntax, and while C and C++ compilers need it and use it (those two languages do not have any include, using or import statement: they rely on CPP for that), it could be used to transform any kind of text file, any language (such as Brainfuck)…

Its syntax is rather simple: any line starting with a “#” is to be interpreted. There are only a handful of instructions that might be used: include, if / else / endif, and of course define. That last one is used to define macros, the main feature of CPP. And with it comes the feature that can be abused: multiple substitutions. The idea behind it is that a substitution can output code that is still “substituable”.

// This is harmless.

#define WORLD_WIDTH   42
#define WORLD_HEIGHT  64
#define WORLD_AREA   (WORLD_WIDTH * WORLD_HEIGHT)

const int area = WORLD_AREA; // expanded as (42 * 64)


// This isn't.

#define TYPES_ARRAY (Shape<int>, Shape<float>, Shape<double>)

#define TYPE1(X, Y, Z) X
#define TYPE2(X, Y, Z) Y
#define TYPE3(X, Y, Z) Z
#define APPLY(X, Y) X Y
#define SELECT(X) APPLY(X, TYPES_ARRAY)

void method1(SELECT(TYPE1) const& object); // expanded as Shape<int>

With the opportunities it opens in mind, some clever folks wrote Boost.Preprocessor, a library that provide macro definitions that allow one to create and manipulate arrays, tuples, lists and sequences in the preprocessor language, although it’s not Turing complete due to the lack of recursion.

Introducing TOOLS_PP

Using Boost.Preprocessor, I sometimes needed additional features, missing from it. This is how TOOLS_PP came to life: it’s a collection of stuff built on top of BOOST_PP. I like to think I could submit it, have it reviewed as a part of Boost itself, but I haven’t even took the time to figure what it would require.

The main feature of TOOLS_PP is made of the kind of dark, ugly, wicked code I was talking about in the introduction. I take a certain pride in it. It’s a macro function named TOOLS_PP_ARRAY_SORT. (Read its code on Github.)

#include <nauths/tools_pp/array_sort.hh>

#define TEST (9, (2, 1, 1, 4, 3, 5, 4, 5, 3))

TOOLS_PP_ARRAY_SORT(TEST)   // expands to (9, (1, 1, 2, 3, 3, 4, 4, 5, 5))
TOOLS_PP_ARRAY_SORT_U(TEST) // expands to (5, (1, 2, 3, 4, 5))

It isn’t that difficult, conceptually; it’s CPP’s syntax that makes it this ugly. It sorts the given array by simply folding over it (using BOOST_PP_WHILE), inserting each element in the accumulated array. The correct insertion index is found with TOOLS_PP_LOWER_BOUND, which simply iterates over the accumulated array. Conceptually, it does almost the same thing as the following (and far more readable) Haskell code.

sort [] = []
sort (x:xs) = insert x $ sort xs

insert x [] = [x]
insert x (y:ys)
  | x > y     = y:(insert x ys)
  | otherwise = x:y:ys

The really, really troubling thing in this story is not TOOLS_PP_ARRAY_SORT_U in itself: it’s that I’ve actually USED it in another project…

But that’s another story for another time.

Thank you for reading!