Auditable Macros in C Code | Hacker News Reader

Auditable Macros in C Code

cossacklabs.com

55 pointspraxis2312 comments

TL;DR

Let the compiler preprocess a source file by first removing #include lines which contain definitions of macros we don't want to see expanded (expanding only the definitions from the local file or from chosen headers, achieving less cluttered and more readable output). Say we have a file like:

    #include "foo_defs.h"
    #define BAR 42
    FOO;
    BAR;

We filter out the include line and then preprocess the file, getting the following output:

    FOO;
    42;

TLDR of why less cluttered preprocessor output matters? I’d say cpp output is usually not that interesting on a daily basis.

If I have a file like this:

    #include <stdio.h>
    #include <assert.h>
    #include "mydefs-control.h"
    #include "mydefs-value.h"

    int main(void) {
        assert(1 == 1);
        printf("%d\n", MYDEF_VALUE_MACRO);
        MYDEF_MACRO_WITH_CONTROL_FLOW();
    }

My goal is to produce output which doesn't include thousands of lines from stdio.h and assert.h, doesn't expand assert() or MYDEF_VALUE_MACRO, and only expands MYDEF_MACRO_WITH_CONTROL_FLOW:

    int main(void) {
        assert(1 == 1);
        printf("%d\n", MYDEF_VALUE_MACRO);

        if (1) {
            return 0;
        } else {
            return 1;
        };
    }

According to the author, having only the "interesting" stuff expanded makes it easier to reason about the control flow of the code.

Which just raises the question for me, why have the macros at all? I did C++ development for years (on low level and performance sensitive applications) but 95% of the macros I saw were there to reduce lines of code or "eliminate boilerplate." Almost all of those could be eliminated by refactoring within C++, without resorting to the preprocessor, but developers love using macros...

I guess it standardizes your boilerplate at write-time even if you have to review it expanded.

The article is about C where not everything can be solved with a typedef, enum or templates. Think about feature macros (e.g. _C_POSIX_SOURCE), generics (pre-C11 or what was it), attributes or calling conventions, conditional compilation, function-like macros, compatibility (e.g. errno when it isn't just an extern int), and indeed shorthands.

Their technique is specific for their set of includes. It would be great if there was a general way to do the macro expansion in a need-to-know eay that could then be a basis for interop tools such as SWIG or python cffi

My uses of macros in C++ these days boil down to a few common use cases that are still insanely not handled well by the language:

0. Anything involving file and line number.

1. Generating a string version of something alongside the value of something, without repeating myself.

2. Creating printing macros to keep me from typing out the entire asinine syntax for something like “std::cerr << foo << bar << baz << std::endl” so I only need PRINT(foo << bar << baz). Substituting code fragments is dead simple with a macro and absurdly complex or impossible with other C++ mechanisms. Ironically I’m only doing it because of the poor design of the entire “iostream” stack.

Relatedly, with all the clang-based tooling like clang-tidy or clang-format, it's kind of disappointing that there isn't a tool to expand arbitrary macros in a source.

That would have been extremely useful when I was recently refactoring some old C code. I ended up writing a small script that can expand one macro at a time, but I wish I hadn't had to.

I'm sure that script doesn't work in the general case.

https://github.com/goldsborough/clang-expand (not perfect because of name conflicts that may occur, but usable)

Can be used in atom: https://atom.io/packages/atom-clang-expand

Waw, thanks. I wish I had found this when I googled a couple months ago when I needed it.

I try to use ObjC PPC macros for things that are constant for the run of the program, and can be reasoned about globally.

For example, some good ones I use are isLandscape or isIPad, which are both variable depending on the user, but constant for the run of the app. Perfect for a PCH file I think.

Reply on news.ycombinator.com