Re: Ada's ranking of popularity at IEEE Spectrum

[sbelmont700@gmail.com]

On Saturday, July 26, 2014 10:01:17 PM UTC-4, David Thompson wrote:
> 
> Huh? Except for application data, which a general-purpose transport
> 
> can't know about, OpenSSL goes to quite a bit of effort to make its
> 
> API about as strongly typed as can be accomplished in C, with mostly
> 
> opaque structs, access routines, shims for ASN.1 and PEM, etc. It's
> 
> not perfect, and certainly not as easy as in Ada, but it's far from
> 
> the all-too-common-in-C void* madness you suggest.


It all comes down to just how "Ada-like" you want it to be.  OpenSSL doesn't throw exceptions, so does the benefits of using them in a binding outweigh the slight deviation from the spec?  The write function takes a length, which in Ada would be inherent in the array, so do you deviate from the spec again?  It doesn't have reference counting (IIRC), so do you use controlled types?  Or do you even go 'full Ada' and make the client instantiate Integer_SSL or Float_SSL functions instead of sending everything as a string?  Or maybe even model it using Ada.Streams?

My point is that even 'good' APIs (like OpenSSL or OpenGL) inherently stink of untyped C, and it becomes a delicate balancing act of figuring out at what point you've failed to implement the spec everyone knows and instead succeeded at inventing a new (better) API that nobody wants.  Victor Porton's copious posts seem to suggest similar problems.

-sb