Re: [Anima] GRASP API in C?

[Brian E Carpenter <[e-mail]>]

On 25/03/2017 07:46, Michael Richardson wrote:
> 
> Brian E Carpenter <[e-mail]> wrote:
>     >> Brian E Carpenter <[e-mail]> wrote: > Hi,
>     >>
>     >> > We will discuss the GRASP API >
>     >> (https://tools.ietf.org/html/draft-liu-anima-grasp-api-03) in Chicago.
>     >>
>     >> > One related topic is how to map this API into C, which is perhaps
>     >> the > most fundamental mapping.
>     >>
>     >> As someone who writes lots of C, I am not convinced this is a useful
>     >> exercise.
> 
>     > Can you explain? If you're arguing that no ASAs will be written in C,
>     > that's defensible, but maybe that isn't your argument?
> 
> I think that ASAs that are written in C will be written in that language for
> reasons of extreme constraints. As such, they won't bother with a
> library/API.  It will all be statically initialized globals and other stuff.
> (Go read the Contiki code for an example of what I mean...)
> 
> Everyone else will write their ASAs in a higher level language: python, ruby,
> perl, java, scala, *go*, maybe rust.   The really hard thing about the C api
> is that doesn't deal well with indefinites like intents...

It doesn't deal well with flexible types either, a dangerous luxury in Python
that I've got very fond of. But it seems to me that if a GRASP core implementation
is written in C for efficiency, it will *need* to offer an API in C that higher
level languages can build on.

And if we are happy with a core written in a high level language, we'll
need to call it from other languages. That can be complicated too.

BTW, my not-production-quality Python version of the GRASP core is now 
about 1800 lines of code, but if we take out the stuff for diagnostics and
debugging there's maybe 1000 lines. I hope we'll find out in the hackathon
how big the BUPT code is; so far they don't support an API, so they are
on your model.

    Brian