C project avoiding naming conflicts

Question

I'm struggling to find pragmatic real-world advice on function naming conventions for a medium sized C library project. My library project is separated into a few modules and submodules with their own headers, and loosely follows an OO style (all functions take a certain struct as first argument, no globals etc). It's laid our something like:

MyLib
  - Foo
    - foo.h
    - foo_internal.h
    - some_foo_action.c
    - another_foo_action.c
    - Baz
      - baz.h
      - some_baz_action.c
  - Bar
    - bar.h
    - bar_internal.h
    - some_bar_action.c

Generally the functions are far too big to (for example) stick some_foo_action and another_foo_action in one foo.c implementation file, make most functions static, and call it a day.

I can deal with stripping my internal ("module private") symbols when building the library to avoid conflicts for my users with their client programs, but the question is how to name symbols in my library? So far I've been doing:

struct MyLibFoo;
void MyLibFooSomeAction(MyLibFoo *foo, ...);

struct MyLibBar;
void MyLibBarAnAction(MyLibBar *bar, ...);

// Submodule
struct MyLibFooBaz;
void MyLibFooBazAnotherAction(MyLibFooBaz *baz, ...);

But I'm ending up with crazy long symbol names (much longer than the examples). If I don't prefix the names with a "fake namespace", modules' internal symbol names all clash.

Note: I don't care about camelcase/Pascal case etc, just the names themselves.

Answer 1

Prefixing (well, affixing) is really the only option. Some patterns you'll see are <library>_<name> (e.g., OpenGL, ObjC runtime), <module/class>_<name> (e.g. parts of Linux), <library>_<module/class>_<name> (e.g., GTK+). Your scheme is perfectly reasonable.

Long names aren't necessarily bad if they are predictable. The fact that you are ending up with crazy long names and functions that are too big to stick with related functions in a single source file raises different concerns. Do you have some more concrete examples?

Answer 2

What about having a global structure variable prefilled with function pointers?

lib.h

#pragma once

typedef struct
{
    void (*doFoo)(int x);
    const char *(*doBar)(void *p);
} YourApi;

extern const YourApi yourApi;

lib.c:

#include "lib.h"

#include <stdio.h>

static void doFoo(int x)
{
    printf("Doing foo %d\n", x);
}

static const char *doBar(void *p)
{
    printf("Doing bar: %p\n", p);
    return "Hello";
}

const YourApi yourApi = {
    doFoo,
    doBar};

Harness:

#include "lib.h"

int main()
{
    yourApi.doFoo(42);
    yourApi.doBar("asd");
}

The static keyword limits the scope to the translation unit so it won't collide with others.

The user can then shorten it by using a pointer like YourApi *ya = &yourApi, then using ya->doFoo(...).

It also provides a nice way to mock your library for testing.

Answer 3

The usual convention for C libraries is to use the library name as prefix for externally usable names, e.g.

struct MyLibFoo;
void MyLibAFooAction(...);

For library-internal names that must still be accessible in multiple units of the library, there is no strict convention, but I would use a prefix of the library name and an indication that it is an internal function. For example:

struct MyLibInternalFooBaz;
void MyLibInternalFooBazAction();

I agree that this can lead to names that are quite long and hard to type, but unfortunately that is the price we have to pay for not having a mechanism like C++ namespaces. To reduce the length of the names, at the cost of some clarity, you can opt to use abbreviations in your names, but you have to weigh the advantages and disadvantages carefully.

Answer 4

If you do want to avoid long prefixes, you can abbreviate library names like what Apple does in iOS and OS X:

• NSString is a string from the NextStep roots of the OS
• CALayer is a Core Animation layer