Re: Adacore joins with Ferrous Systems to support Rust

["Alejandro R. Mosteo" <alejandro@mosteo.com>]

I'm sorry if this pops up twice, something strange happened with my 
first attempt.

On Fri, Feb 11, 2022 at 8:24 PM 'Luke A. Guest' wrote:
 > So, you'd prefer, if Ada was designed now, it didn't do runtime check 
(on pointers) and have compile-time checks?

I'd prefer that, as much as feasible, checks were moved (not removed!) 
to compile-time, yes. I know there are efforts in this direction at
AdaCore to simplify the accessibility checks model.

 > I'm more ashamed now of the whole anonymous pointers and 
accessibility surprises in Ada.
I'm not sure what you mean here.

My problem with runtime checks (which are undoubtedly better than no 
checks, sure), and in particular with accessibility checks, is that 
sometimes you get a failure much later during testing. By that time, 
understanding the problem may be 1) hard and 2) require painful 
redesign. At compile-time you get to deal with the problem immediately.

This is something in which Rust and Ada share the sentiment: "if it 
compiles, it works". So having something in another language found at 
compile-time makes me want to have it also in Ada at compile-time. It 
really spoils you against runtime checks. Much like I prefer the static 
elaboration model in GNAT instead of the dynamic one.

Also there are times in Ada where static checks are false positives that 
require some 'Unchecked_Access, and other times there is no failure yet 
you're doing something wrong. I find these from time to time in pretty 
obscure combinations not easy to provide a reproducer and frankly, I 
hate it. I'm never sure if I'm at fault, the compiler is at fault, or 
I've hit a corner case in the "heart of darkness". Nowadays I won't use 
a pointer even if it means obscene underperformance, until the thing is 
unavoidable.

There are also situations in which marking a parameter as aliased, even 
if you know it is already by reference (a limited/tagged type), will 
alter the things you can do with 'Access/'Unchecked_Access. There have 
been a couple of recent posts about that. Even if it's my fault, I find 
too hard to repeatably remember the finer details.