Manual name mangling in C

Question

I would like to implement some kind of name mangling for a C interface. I will have functions and types like

struct Foo;
struct Foo* Foo_create(struct Bar* bar);
void Foo_destroy(struct Foo* obj);
struct Foo_InnerClass;
...

Now, the _ is used extensivliy in all standard c typedefs, so the scheme would not work good if these typedefs are used. GCC and MSVC seems to accept names using $. Would that work? Are there any other acceptet characters that I can use?

EDIT:

I want to do this because

It avoids duplicated names
It emphases what class an identifier belongs to
It makes it possible to automatically generate C++ interface

EDIT 2:

Reserve an ordinary letter will compile everywhere, but it can also be confusing:

struct Foo* FooIcreate(struct Bar* bar);

EDIT 3:

The symbol may differ depending on target platform, but on that platform, it must not collide with C++ name mangling.

EDIT 4:

It is forbidden to begin an identifier with a digit. So what about:

struct Foo* Foo_1_create(struct Bar* bar);

@larsmans No, rather emphase that InnerClass belongs to Foo in a machine parsable way. — user877329, Oct 23 '12 at 11:50
My first question to @user877329 is : do you want to implement same name mangling algorithm that's used by GCC and MSVC? — Aniket Inge, Oct 23 '12 at 12:46
@PrototypeStark No, because the calling convention differs. If there was no such problem, i could use C++ directly, now i cannot. — user877329, Oct 23 '12 at 12:51

score 1 · Answer 1 · answered Oct 23 '12 at 11:50

1

Not sure why you are doing this, or exactly what the goal is, but most C libraries use some more or less mnemonic prefix.

A plain $ would be awful, since it's a) very uncommon, C identifiers typically start with a letter or underscore and contain only letters, digits and underscores and b) doesn't say anything about to what the symbol belongs.

A library like glib has a g_ prefix for all functions, a g for its types, and G_ for macros. GTK+ is similar, it uses gtk_ and so on.

answered Oct 23 '12 at 11:50

unwind

391,730
64
469
606

Very uncommon makes it good because no one would use it in his own name. – user877329 Oct 23 '12 at 11:59
1

@user877329: `$` is not allowed by ISO C. – Fred Foo Oct 23 '12 at 12:47
@larsmans which is the only bad thing about it (as long as the linker and assembler accepts it). – user877329 Oct 23 '12 at 12:52

score 1 · Answer 2 · edited May 23 '17 at 11:59

1

Don't use $. The C standard does not allow them in identifiers, even though C compilers such as GCC accept them for backward compatibility with 1980s C compilers (mostly VAX/VMS C, I believe).

How about using double underscores?

struct Foo__InnerClass;

EDIT: ok, this prevents your program from ever working in a mixed C/C++ program (see C++ underscore rules), so it's a bad idea. How about a trailing underscore:

struct Foo_InnerClass_

edited May 23 '17 at 11:59

Community

1
1

answered Oct 23 '12 at 11:52

Fred Foo

355,277
75
744
836

Isn't that undefined behaviour? – Kerrek SB Oct 23 '12 at 11:56
1

@KerrekSB, no not for C, but for C++. If there is any chance that the interfaces are also used from C++ this isn't a good idea. They need that for ... name mangling. – Jens Gustedt Oct 23 '12 at 12:09
@JensGustedt Neither GCC ver 3 or later or VC seems to use __ . – user877329 Oct 23 '12 at 12:38
@user877329, so what? Just because some compilers don't use it, it doesn't mean that other wouldn't or even that the same compilers wouldn't do it in other versions. – Jens Gustedt Oct 23 '12 at 13:19

score 1 · Answer 3 · answered Oct 23 '12 at 12:09

It's fine, though I like to use struct Foo_t to show it's a type.

My preference is to keep double _ (__) for something special or a "private" indicator and not regular use. This is pure personal preference though any local coding standards may say something on the subject.

The $ is iffy... while some compilers accept it some don't. It is not a historic part of the C language. While common in some operating systems and languages, not so much in others. History: the big three historic specials of IBM mainframes are $, #, and @ where _ was NOT a member of the club and VMS loves $.

The `_t` suffix is reserved by POSIX, AFAIR, and `__` are reserved by C++ to do name mangling. — Jens Gustedt, Oct 23 '12 at 12:11

Eldar Abusalimov · Answer 4 · 2012-10-23T15:17:07.000

0

It would not use dollar sign in identifiers because it is:

Uncommon, I didn't see it anywhere
Because of the first reason it would confuse users of your code, and maybe even some tools, editors, etc.

It is not supported for all targets, for example:

echo 'void foo$bar(void) { }' | arm-elf-gcc -xc -c -

Outputs:

<stdin>:1: error: stray '$' in program
<stdin>:1: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'bar'

Personally I would use some kind of module prefixes, like in your first code snippet. If you afraid of name collisions with other libraries you can add one more global prefix, like mylib_foo_create.

edited Oct 23 '12 at 15:17

answered Oct 23 '12 at 12:00

Eldar Abusalimov

24,387
4
67
71

The problem with _ is that size_t or int32_t then is interpreted as member t in class size and int32 respectivly. – user877329 Oct 23 '12 at 12:05
@user877329 Is it really an issue? Are you trying to parse you code automatically to extract type information or something like that? – Eldar Abusalimov Oct 23 '12 at 12:08
Even if so, `size_t` and `int32_t` identifiers are part of C language standard and you could handle them in a special way. – Eldar Abusalimov Oct 23 '12 at 12:09
@EldarAbsalimov Yes I could, but it is ugly. – user877329 Oct 23 '12 at 12:35

user877329 · Accepted Answer · 2012-11-21T12:44:22.743

0

In lack of a better sulotion, i think i will use

_1

as separator

edited Nov 21 '12 at 12:44

answered Oct 23 '12 at 14:06

user877329

6,717
8
46
88

Manual name mangling in C

5 Answers5