Re: Position of "use"

[Tero Koskinen <tero.koskinen@iki.fi>]

10.7.2014 18:57, Adam Beneschan wrote:
> GNAT lets things slip through that should be errors.  

To GNAT's defence, this kind of bugs are pretty hard to find,
especially afterwards if enough care has not been taken
during the initial implementation.

Many times you only notice "accepts invalid code" bug
if another compiler rejects the code.[1]

Some of my recent examples:
1)
GNAT compiles Strings Edit 2.9 cleanly, while ICCAda rejects
the code:
>
> "strings_edit-integers-subscript.ads", line 31: Error: Undefined >
identifier:
>
>           Number. [RM 4.1(11), 8.3, 8.6(28)]
>

(See http://build.ada-language.com/job/Strings_Edit_ICCAda/3/console for
full log.)


2)
ICCAda and GNAT both compile Debug package from Adalog
( http://www.adalog.fr/compo2.htm#Debug ) while Janus/Ada rejects
the code:
> In File C:\work\adalog-debug\DEBUG.ADS at line 206
> --------------
>   205:     type Debug_String is new String;
>   206:     procedure Free is new Ada.Unchecked_Deallocation
> (Debug_String, Debug_String_Access);
>-------------------------------------------------------------------------^
> *ERROR* Formal and actual must both be constrained or unconstrained
> (6.4.10) [RM 12.5.4(3)]


Of course, in these cases, one could probably argue which compilers
are correct and which are not; luckily I a user, who does not need
to worry about the compiler implementation. :)

Yours,
 Tero

[1] Yes, there is ACATS, but as you can see, it does not cover everything.