Re: Generic-Package Elaboration Question / Possible GNAT Bug.

["Randy Brukardt" <randy@rrsoftware.com>]

"Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de> wrote in message 
news:12hfiflyf7pr5$.l3pkpgoid8xt$.dlg@40tude.net...
 > I my experience at least 40% of my debugging
> efforts in Ada 95 was spent on catching accessibility checks faults. Ada
> 2005 improved that greatly. To me this is a vivid example why dynamic
> checks are evil.

It's not "dynamic checks" that are evil, it is checking the uncheckable that 
is evil. 95% of accessibility check failures (static or dynamic) have no 
correlation to any error, so they are just crying wolf. "Checks" that don't 
prevent errors are evil, period -- it doesn't matter how they are checked.

Note that I no longer even try to deal with accessibility checks; I use 
'Unchecked_Access (only). Indeed, I've only found one single instance in my 
code (since Ada 95, about 15 years now) where I could actually use 'Access 
without error. In all other cases, my code either needs '[Unchecked_]Access 
for some short-lived thing (like a call to a C API) which has long-lived 
types (because it is declared at library-level), or I have to manage the 
lifetime issues myself (using controlled types).

OTOH, dynamic checks like overflow checks and index checks are not evil, or 
maybe it would be better to say that they are the lesser of evils (getting 
the wrong answer silently is a greater evil). I don't see a practical 
alternative (proving a multiply does not overflow is difficult to do in 
general; I gave up trying in the Janus/Ada optimizer and simply try the 
operation to see what happens. It's not ideal, but it is the only practical 
solution).

                                Randy.