Re: [boost] [type_erasure]suggest tuple doc contain requires clause specifying T... are unique

Subject: Re: [boost] [type_erasure]suggest tuple doc contain requires clause specifying T... are unique
From: Larry Evans (cppljevans_at_[hidden])
Date: 2012-07-31 05:34:15

On 07/30/12 16:58, Steven Watanabe wrote:
> AMDG
> 
> On 07/30/2012 11:06 AM, Larry Evans wrote:
>> http://steven_watanabe.users.sourceforge.net/type_erasure/libs/type_erasure/doc/html/boost/type_erasure/tuple.html
>>
>> Specifies that the T... are placeholders; however, it doesn't
>> specify that they be unique; hence,
>>
>>   tuple<concept,_a,_a> x(int(0),std::vector<int>(0));
>>
>> would satisfy the current description but wouldn't be
>> right.  Of course no one would intentionally do that; however,
>> if the Tags are calculated in some meta-program, it might
>> happened accidentally.  Hence the requires class should also
>> say, "If T... are not unique, no diagnostic occurs", or something
>> similar.
>>
>> The reason I think "no diagnostic occurs" is that I've looked
>> briefly in tuple.hpp, at the:
>>
>>      ::boost::type_erasure::cons<Concept, T...> impl;
>>
>> in the class tuple definition and the cons definition, and I don't
>> see any check for duplicates.
>>
> 
> The placeholders don't have to be unique, but
> if a placeholder is used more than once, the
> corresponding constructor arguments all need to
> be the same type.  This is another instance of
> the same bug that you reported earlier (with the
> any constructor.).  A fix to that, will automatically
> fix this as well.

Confirmed, as show by running the attached with
the fix to any.hpp suggested here:

http://article.gmane.org/gmane.comp.lib.boost.devel/233104

The output of running the attached is:

./tuple_dup_tags.exe
v<3>()
v<2>()
v<1>()
v<1>( v<1> const&)
v<2>( v<2> const&)
v<2>( v<3> const&)
&t=0x7fffcd6f8e60

Showing that the first arg to the any CTOR corresponding
to a _b is the one that determines the type of _b.
As shown by the output, even though the 2nd _b is created
with a v<3> argument, a v<2> is created.

However, I think the docs should point out that this
is what happens because some user might expect that
a _b in the above test would map to a v<3> instead
of a v<2>.

-regards,
Larry