[llvm-commits] [PATCH] YAML I/O for review

[Nick Kledzik]

On Dec 5, 2012, at 12:16 PM, Sean Silva wrote:

> On Wed, Dec 5, 2012 at 2:58 PM, Michael Spencer <bigcheesegs at gmail.com> wrote:
>> +  struct IntVector : public std::vector<int> {
>> +    // The existence of this member causes YAML I/O to use a flow sequence
>> +    static const bool flow = true;
>> +  };
>> 
>> Deriving types like this breaks the constructors for vector. There
>> needs to be a different way to do this that doesn't require flow to be
>> a member.
> 
> traits++
> 
> You can just have a template (e.g. FlowFormattingTraits<>) which you
> specialize to indicate that a class should be flow formatted. The
> primary template can conveniently use the presence (using SFINAE) and
> value of the class member (`flow` in Nick's original patch, although I
> think a longer and more unique name would be better) to select flow
> formatting if there is no explicit specialization for the type.

It did seem a little odd that you needed a class wrapper to make flow sequence.  
So I do like the FlowFormattingTraits<> idea.  But that brings up the issue
that sequences in general are not trait based like the other parts of YAML I/O
are.  Your existing data has to have begin(), end(), push_back(), iterator,
and type_value to work as a sequence.   It would be better if you could specify
a trait on your native type that tells YAML I/O how to iterate through it and
how to append to it.  Hopefully, I could then have a built in trait specialized on
any std::vector<>.  Thus if your native type is a vector, it would automatically
convert to a YAML  sequence.  But if you have some other form of container
you can write a trait which allows YAML I/O to use your type as a sequence.

-Nick