Re: [Boost-users] Max Boost Fusion tuple parameters

Subject: Re: [Boost-users] Max Boost Fusion tuple parameters
From: Larry Evans (cppljevans_at_[hidden])
Date: 2013-06-11 14:52:37

On 06/11/13 10:11, Larry Evans wrote:
> On 06/11/13 09:31, Igor R wrote:
>>>> What is the maximum Boost Fusion tuple param count?
>>>>
>>>> I was thinking of using it for about 200-300 elements.
>>>>
>>>> Anyone done this?
>>>>
>>>> I've heard of defines for this, but never actually *knew* of anyone
>>>> actually doing it.
>>>>
>>>> Thanks,
>>>> Rodrigo Madera
>>>
>>> I found some relevant information on this by:
>>>
>>>   1) going to:
>>> http://www.boost.org/doc/libs/1_53_0/libs/fusion/doc/html/index.html
>>>   2) then following the _Class template tuple_ link to:
>>> http://www.boost.org/doc/libs/1_53_0/libs/fusion/doc/html/fusion/tuple/class_template_tuple.html
>>>   3) then following the _vector_ link to:
>>> http://www.boost.org/doc/libs/1_53_0/libs/fusion/doc/html/fusion/container/vector.html
>>>
>>>   4) which said:
>>> You may define the preprocessor constant FUSION_MAX_VECTOR_SIZE before
>>> including any Fusion header to change the default. Example:
>>>
>>> #define FUSION_MAX_VECTOR_SIZE 20
>>
>>
>> Template argument number in a template declaration is limited by
>> compilers. For example, in MSVC10/11 it's limited to 64. So I doubt
>> setting FUSION_MAX_VECTOR_SIZE to 200 would work.
>>
> 
> Thanks Igor.  I had no idea about that :(
> 
> In this case, I'd guess the only solution for Rodrigo is to use
> Christopher Schmidt's variadic tuples:
> 
> http://article.gmane.org/gmane.comp.lib.boost.devel/194322
> 
> I remember (a year or more ago) looking at the Christopher's
> implementation and, IIRC, it used a combination of the boost
> preprocessor library and variadic templates to speed the compilation
> times.  Again, IIRC, Christoper built up a N*CHUNK size tuple of T
> element by creating N structs each containing CHUNK elements of type
> T.  I think the N structs were recursively inherited; so that struct
> tuple_chunked<N> inherited from tuple_chunked<N-1>.  Thus, if CHUNK
> were 64 (in the case of MSVC`10/11, then creating a tuple of 200 would
> just require about 200/64 template instantiations.
> 
> The last I heard about Christopher's effort was this post:
> 
> http://article.gmane.org/gmane.comp.lib.boost.devel/234501
> 
> The code location (I don't know if it's still valid) was give in post:
> 
> http://article.gmane.org/gmane.comp.lib.boost.devel/228820
> 

I probably jumped to the conclusion that "template argument number"
meant "number of template instantiations".  Sorry for that mistake.
I guess "template argument number" means the maximum arity of any
template.  IOW, if "template argument number" was 5, then any
template taking over 5 arguments would cause the MSVC10/11
compiler to diagnose an error.  IOW:

  tuple<T1,T2,T3,T4,T5>

would be OK, but:

  tuple<T1,T2,T3,T4,T5,T6>

would cause MSV10/11 to issue a compiler error.
Is that right?

Also, I also read somewhere that MSVC had not yet implemented
variadic templates:

http://stackoverflow.com/questions/9589192/how-do-i-change-the-number-of-template-arguments-supported-by-msvcs-stdtupl

Sorry for noise.

-regards,
Larry