Re: [boost] New library? CallableTraits

Subject: Re: [boost] New library? CallableTraits
From: Barrett Adair (barrettellisadair_at_[hidden])
Date: 2016-05-16 15:32:24

On Sun, May 15, 2016 at 8:13 AM, Louis Dionne <ldionne.2_at_[hidden]> wrote:
>
> Barrett Adair <barrettellisadair <at> gmail.com> writes:
>
>> (snip)
>
> > 1. Is there any interest in reviewing a future version of this for
> > inclusion in Boost?
>
> I think so. I've also heard a couple of people say that they needed
similar
> functionality this week in Aspen, so I imagine this would be a useful
> addition to Boost. In the long term, I think we need something similar
> (perhaps a subset of CallableTraits) in the standard.
>

This is great to hear.

> > 2. If so, does this belong in its own library, or should it be
incorporated
> > into Boost.TypeTraits or something similar?
>
> No strong opinion here, though it might be a bit large for an addition to
> Boost.TypeTraits? In all cases, if there's a review of the library, you
can
> count on me to submit one.

Also great!

>
> > 3. What you like to see added, removed, renamed, or redesigned?
>
> Here are a few comments on the documentation and the library itself:
>
> - IMO, the overview should include explanations in running text rather
than
>   just comments in the code. Or if you want to keep code only, which
might be
>   a good idea, I would make it a bit shorter. Do not try to show all the
>   features and details of the library; this is just an overview.
>

I've updated the introduction so that it begins with a simple, terse
overview [1]. Is this a step in the right direction?

> - The Java-style interface example is _huge_, and IMO it doesn't fit into
an
>   example at the beginning of the documentation. It might be better to
include
>   it as an appendix.
>

That's a good point. I've moved it to the end of the documentation for now.

> - The library documents many concepts, but it seems to me that they are
not
>   really concepts as we usually mean in C++. To me, most of them are just
>   vocabulary terms (that could e.g. appear in a glossary). When I
introduce
>   concepts in a library, I usually design my concepts as some kind of
>   hierarchy with semantic properties that allow me to describe a few key
>   types of objects. Instead, I see some "concepts" in CallableTraits as
just
>   glossary terms without semantics. For example, instead of the
"FunctionPtr"
>   concept, you could have a "Function pointer" entry in a glossary.
>

Hmm, that's another good point. I will think about this suggestion, and
I'll change the name of the section at the very least.

I came up with the concept "members" idea while writing the reference
documentation - I think it will make the reference section easier to write
and comprehend. However, I may be completely off-base about this.

>
> - Suggestion: use `replace_at` instead of `overwrite_at`?
>
> - A few metafunctions have the `_at` suffix; is this necessary?
>

I renamed `overwrite_at` to `replace_args`. I did the same for the other
`_at` metafunctions.

> - Instead of `can_invoke`, something like `is_invokable` might be better?
>   Querying traits often start with `is_`.
>

It seems that this offers similar functionality to C++17's `is_callable`,
but it might be worth keeping around for those who are stuck with older
compilers/std libraries. I may actually rename this to `is_callable` and
then remove CallableTraits' existing `is_callable` trait.

>
> > I'd say CallableTraits is about 85% finished. Here's what's left:
> > * Add INVOKE semantics to a couple of things for consistency (arity,
etc)
> > * Remove unused/undocumented code
> > * Account for transaction_safe and C++17 noexcept
> > * Finish/test the calling convention features
> > * Finish/update the reference documentation
>
> Finishing the reference documentation is, IMO, the #1 priority if others
> express interest towards the library and you want to make it part of
Boost.
> I don't think a formal review can take place before the documentation is
> done, but you might want to wait to see if there's interest before
investing
> a lot of time in the documentation.
>

Yes, definitely. Part of my reasoning for bringing this to the mailing list
at this stage is that I'll be much more motivated to finish the reference
docs in a timely manner if people are interested in adding this to Boost.

Yesterday I committed the code changes for INVOKE consistency, and removed
a lot of unused code.

Thanks for your detailed response.

Barrett

[1]
http://badair.github.io/callable_traits/doc/html/index.html#callable_traits.introduction