Re: Teaching C/C++ from Ada perspective?

[Maciej Sobczak <see.my.homepage@gmail.com>]

> "They do, they are just a little bit smarter in how they do it by forcing names to be declared within packages." (Sobczak)
> 
> I very much doubt Ada does mere textual inclusion,

It does not do *mere* inclusion in the sense of blindly concatenating bytes, but it does textual read (and processing) of all involved files, which at the end is no different from the point of view of managing large codebases.

> Ada has *semantical* devices for programming in the large.

Let's try these "semantical" devices in practice:

-- p.ads:
package P is
   pragma Pure (P);
end P;

-- q.ads:
with P;
package Q is
   procedure Hello;
end Q;

-- q.adb:
with Ada.Text_IO;
package body Q is
   procedure Hello is
   begin
      Ada.Text_IO.Put_Line ("Hello");
   end Hello;
end Q;

-- test.adb:
with Q;
procedure Test is
begin
   Q.Hello;
end Test;

In this large-scale project the package P has no semantic meaning whatsoever. It does not declare anything and no other package refers to any declarations defined in there (because there are none anyway). The only dependency on this package is in package spec Q, which withs it, for no reason whatsoever. It might be a left-over from some refactoring, perhaps? We don't know, but clearly, the main procedure Test has no relation to P and in particular, the author of Test has never intended to depend on P for any reason.

Yet, the trivial update of P (like adding whitespace, which again has no semantic meaning whatsoever) forces the recompilation of Test (with GNAT). Actually, it forces the recompilation of everything. Why?

There go your "semantic" devices for programming in the large. Sorry, but even though I do understand the intent, I don't see it in practice. As far as I'm concerned, the Ada handling of text files in response to "with" is exactly as primitive as the C++ handling of #include, except for the lack of explicit concatenation (which, by the way, C++ compilers don't do, either - see "precompiled headers"). But the impact on compilation dependencies is exactly the same. In this context both Ada and C++ are shitty in exactly the same way.

> Other issues provoked by mere textual inclusion include polluting *every* unit with
> 
> #ifndef X
> #define X
> 
> ...
> 
> #endif
> 
> This "feature" alone makes any programmer of good taste refuse the language.

Except that nobody cares. This became so idiomatic that every IDE I know of generates these guards automatically. This allows to focus the good taste on other aspects of software engineering.

-- 
Maciej Sobczak * http://www.inspirel.com