[9fans] stdarg & va_copy

[9fans] stdarg & va_copy

[erik quanstrom]

is there any reason that /$objtype/include/u.h does not
define va_copy?  are there objections to this c99 macro?

- erik

Re: [9fans] stdarg & va_copy

[Russ Cox]

> is there any reason that /$objtype/include/u.h does not
> define va_copy?  are there objections to this c99 macro?

Yes.  The definition and semantics of va_copy are
sufficiently murky that it seemed best to omit it.

If you are porting code that uses va_copy, you can just
    #define va_copy(x, y) (x) = (y)
in your own compatibility headers.

I'm still annoyed that the C99 committee outlawed taking
the address of a va_list.

Russ

Re: [9fans] stdarg & va_copy

[erik quanstrom]

On Thu Jan 24 17:04:00 EST 2008, rsc@swtch.com wrote:
> > is there any reason that /$objtype/include/u.h does not
> > define va_copy?  are there objections to this c99 macro?
> 
> Yes.  The definition and semantics of va_copy are
> sufficiently murky that it seemed best to omit it.

i didn't see anything in the definition that would make va_copy
wrong given the plan 9 definition of va_list.  is there a particular
case on plan 9 that would be a problem?

> If you are porting code that uses va_copy, you can just
>     #define va_copy(x, y) (x) = (y)
> in your own compatibility headers.
> 
> I'm still annoyed that the C99 committee outlawed taking
> the address of a va_list.

understandable.  the whole thing is quite annoying.

but plan 9 does have va_start and va_end.  wouldn't make sense
to have va_copy as well?

- erik

Re: [9fans] stdarg & va_copy

[Russ Cox]

> i didn't see anything in the definition that would make va_copy
> wrong given the plan 9 definition of va_list.  is there a particular
> case on plan 9 that would be a problem?

it might make sense to *have* va_copy, but since plan 9
programs don't use va_copy, there's no need to provide
an implementation.  and honestly, the fewer people who
use va_copy, the better.

> but plan 9 does have va_start and va_end.  wouldn't make sense
> to have va_copy as well?

plan 9 makes no claims of being C99 compliant,
although it happens to have a few of the same extensions.
va_start, va_end, and va_arg are from an earlier standard.
those (in particular va_arg) provide useful functionality.
va_copy does not.

like i said, if you need it, it's easy to put a #define
in your own compatibility headers.  if you don't
need it, don't worry about it.

russ

Re: [9fans] stdarg & va_copy

[Pietro Gagliardi]

One possible use of va_copy is in an implementation of print[f]()  
that supports an option to get a specific argument. For example,

	print("%2#s %1#s", "second", "first");

prints

	first second

For example, one could maintain an array of the first n va_list  
items, save them as they are read, and then use the copies when  
necessary. The other option is
	va_end()
	va_start()
	while (newcount < count)
		va_arg()
which could flood your code.

On Jan 24, 2008, at 5:39 PM, Russ Cox wrote:

>> i didn't see anything in the definition that would make va_copy
>> wrong given the plan 9 definition of va_list.  is there a particular
>> case on plan 9 that would be a problem?
>
> it might make sense to *have* va_copy, but since plan 9
> programs don't use va_copy, there's no need to provide
> an implementation.  and honestly, the fewer people who
> use va_copy, the better.
>
>> but plan 9 does have va_start and va_end.  wouldn't make sense
>> to have va_copy as well?
>
> plan 9 makes no claims of being C99 compliant,
> although it happens to have a few of the same extensions.
> va_start, va_end, and va_arg are from an earlier standard.
> those (in particular va_arg) provide useful functionality.
> va_copy does not.
>
> like i said, if you need it, it's easy to put a #define
> in your own compatibility headers.  if you don't
> need it, don't worry about it.
>
> russ
>

Re: [9fans] stdarg & va_copy

[Charles Forsyth]

> One possible use of va_copy is in an implementation of print[f]()  
> that supports an option to get a specific argument. For example,

we'd just use the copy operation provided by the language:  =

Re: [9fans] stdarg & va_copy

[Douglas A. Gwyn]

erik quanstrom wrote:
> is there any reason that /$objtype/include/u.h does not
> define va_copy?  are there objections to this c99 macro?

Probably it was left out due to not being in the C90 spec.
There shouldn't be any problem with it, but is it needed?

Re: [9fans] stdarg & va_copy

[erik quanstrom]

> erik quanstrom wrote:
> > is there any reason that /$objtype/include/u.h does not
> > define va_copy?  are there objections to this c99 macro?
> 
> Probably it was left out due to not being in the C90 spec.
> There shouldn't be any problem with it, but is it needed?

no.  neither is va_end or the compiler accepting, but ignoring,
the "restrict" qualifier.  but we already do these things.

on the other hand, it does limit needless incompatiblity.

- erik