Re: [boost] [vmd] Library Review

Subject: Re: [boost] [vmd] Library Review
From: Edward Diener (eldiener_at_[hidden])
Date: 2014-09-14 20:12:21

On 9/14/2014 6:49 PM, Paul Mensonides wrote:
> On 9/14/2014 3:12 PM, Edward Diener wrote:
>> On 9/14/2014 5:10 PM, pfultz2 wrote:
>
>> I am not sure what you mean by namespaces ? Do you mean the type of the
>> token parsed ?
>
> I believe he means macro prefixes such as BOOST_PP_, BOOST_VMD_, or
> BOOST_VMD_NAMESPACE_.
>
>>> Perhaps Paul can explain it better than me.
>>
>> The BOOST_PP_AUTO_REC macro is not documented so maybe you can
>> understand my being puzzled by it and what it is used for. Other
>> undocumented macros in Boost PP are generally easier for me to follow.
>
> The BOOST_PP_AUTO_REC macro is used to find the first macro of a set of
> macros which has not been painted (i.e. will expand).  It requires a set
> of macros which are a power of 2 which are used in order.

Here is my understanding of what BOOST_PP_AUTO_REC does from your 
example. It takes a function-like macro name and a parameter which is a 
Boost PP number which is a power of 2, from a minimum of 2 to a maximum 
of 256. It then calls the function-like macro name passing a number 
starting with 1 through the number passed to it. If the function-like 
macro returns 1 it stops and returns the current number else it 
continues with the next number. I won't ask what BOOST_PP_AUTO_REC 
returns if its function-like macro never returns 1 before it finishes 
cycling through its numbers <g>.

I assume the "painted" part is because it is usually used with 
BOOST_PP_NULLARY as in your example below.

> For example:
>
> #include <boost/preprocessor.hpp>
> #include <boost/preprocessor/detail/is_nullary.hpp>
>
> #define CAT_1(a, b) PRIMITIVE_CAT_1(a, b)
> #define CAT_2(a, b) PRIMITIVE_CAT_2(a, b)
> #define CAT_3(a, b) PRIMITIVE_CAT_3(a, b)
> #define CAT_4(a, b) PRIMITIVE_CAT_4(a, b)
>
> #define PRIMITIVE_CAT_1(a, b) a ## b
> #define PRIMITIVE_CAT_2(a, b) a ## b
> #define PRIMITIVE_CAT_3(a, b) a ## b
> #define PRIMITIVE_CAT_4(a, b) a ## b
>
> #define CAT \
>      BOOST_PP_CAT( \
>          CAT_, \
>          BOOST_PP_AUTO_REC(PRIMITIVE_CAT_P, 4) \
>      )
>      /**/
> #define PRIMITIVE_CAT \
>      BOOST_PP_CAT( \
>          PRIMITIVE_CAT_, \
>          BOOST_PP_AUTO_REC(PRIMITIVE_CAT_P, 4) \
>      ) \
>      /**/
> #define PRIMITIVE_CAT_P(n) \
>      BOOST_PP_IS_NULLARY(PRIMITIVE_CAT_ ## n(,)()) \
>      /**/
>
> Here the CAT and PRIMITIVE_CAT macros are defined as object-like macros
> which expand to the next available CAT_n and PRIMITIVE_CAT_n macros
> which are found via the the BOOST_PP_AUTO_REC macro--which does a binary
> search from 1 to 4 in this case finding the first index for which the
> passed predicate PRIMITIVE_CAT_P expands to 1.
>
> Essentially, it makes it look like CAT can be invoked recursively, but
> CAT itself is not really even a function-like macro.  E.g.
>
> #define A1 CAT(B, 1) // here CAT expands to CAT_2
> #define B1 CAT(C, 1) // here CAT expands to CAT_3
> #define C1 CAT(D, 1) // here CAT expands to CAT_4
>
> CAT(A, 1) // here CAT expands to CAT_1

Thanks for the explanation.

>
>> You could actually do something similar with VMD now but you would have
>> to try each v-type separately to see if it begins the v-sequence. But
>> after Paul Mensonides provided a better solution to identifying
>> v-identifiers and v-numbers I can improve the parsing of VMD v-sequences
>> greatly.
>>
>> In fact with VMD a v-sequence becomes another preprocessor "data type"
>> similar to the high level Boost PP data types of array, list, seq, and
>> tuple. I should really work to produce whatever operations with a
>> v-sequence I can that would mimic some of those Boost PP data types. Its
>> challenging to think about but it will be fun work except for the usual
>> VC++isms.
>
> The alternate registration mechanism (excluding the comma in the
> registration itself) that you guys realized is good and is obvious is
> retrospect.  That would make the registration much cleaner.

I agree. Your proposal is obvious in retrospect but it is like the 
legendary story of Christopher Columbus and the egg.

I am now very glad I can make a v-sequence much easier with better 
functionality. Why I didn't think of your solution to being with ... <g>.