__VA_COUNT__: Count number of variadic arguments in a variadic macro

[alejandro-colomar]

Hi!

While you can do this already with macro trickery, it requires some very unreadable code. I'd like to be able to extract that information directly in a way similar to how __VA_OPT__() works.

#define foo(a, b, ...)  __VA_COUNT__

foo(a, b, c, d)

The code above would expand to

2

#define bar(...)  __VA_COUNT__

bar()  // 0
bar(x)  // 1

Could this be added as a Clang extension?

Cc: @AaronBallman

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: Alejandro Colomar (alejandro-colomar)

Hi!

While you can do this already with macro trickery, it requires some very unreadable code. I'd like to be able to extract that information directly in a way similar to how __VA_OPT__() works.

#define foo(a, b, ...)  __VA_COUNT__

foo(a, b, c, d)

The code above would expand to

2

#define bar(...)  __VA_COUNT__

bar()  // 0
bar(x)  // 1

Could this be added as a Clang extension?

Cc: @AaronBallman

[erichkeane]

I think I'd be OK with this, I don't see a good reason why not. IMO we'd want to make sure that we communicated with GCC to see if we could make them do it at the same time so that it is useful (I don't really like messing with the preprocesor at all without them doing the same thing!).

I'd probably want to bikeshed the name a bit better (perhaps __VA_OPT_COUNT__ or something, so it is clear they 'go together'), but I don't see any problems with this.

[alejandro-colomar]

I think I'd be OK with this, I don't see a good reason why not.

Thanks!

IMO we'd want to make sure that we communicated with GCC to see if we could make them do it at the same time so that it is useful (I don't really like messing with the preprocesor at all without them doing the same thing!).

Sure; here you go: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121665

I'd probably want to bikeshed the name a bit better (perhaps __VA_OPT_COUNT__ or something, so it is clear they 'go together'), but I don't see any problems with this.

In __VA_OPT__, OPT refers to the argument being optionally pasted.

In this case, it's not optional. The thing that makes them go together is the __VA_ prefix, which means "this is about variadic macros".

It also goes together with __VA_ARGS__, which of course also has the __VA_ prefix, meaning it's about variadic macros.

[pinskia]

A name suggestion from 2008 was __VA_ARGC__ (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=33877).

[erichkeane]

I think I'd be OK with this, I don't see a good reason why not.

Thanks!

IMO we'd want to make sure that we communicated with GCC to see if we could make them do it at the same time so that it is useful (I don't really like messing with the preprocesor at all without them doing the same thing!).

Sure; here you go: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121665

I'd probably want to bikeshed the name a bit better (perhaps __VA_OPT_COUNT__ or something, so it is clear they 'go together'), but I don't see any problems with this.

In __VA_OPT__, OPT refers to the argument being optionally pasted.

Ah, right, I meant __VA_ARG_COUNT__, kind of a recency-bias /brain fart :)

In this case, it's not optional. The thing that makes them go together is the __VA_ prefix, which means "this is about variadic macros".

It also goes together with __VA_ARGS__, which of course also has the __VA_ prefix, meaning it's about variadic macros.

Yeah, see above.

A name suggestion from 2008 was __VA_ARGC__ (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=33877).

I see that one... that name is awful :)

[alejandro-colomar]

@pinskia Ahhh, I like __VA_ARGC__! It's a traditionalist name in C. :)

As for __VA_ARG_COUNT__... Meh, argc is the traditional abbreviation for that. It's too long.