Thanks for the reply. I was recently involved in
the arrangement of an company internal course
and provided the exercises. When I finally came
around to making up sample solutions I came around
this compiler bug. I have not programmed actively
in Ada95 and was hence a bit uncertain about
the validity of the example, but couldn't find what
was wrong. Thanks for re-erecting my "view-of-the-world"!

Daniel

PS! I'm running gnat-3.11p in WinNT 4.0, servicepack 3 DS.


-----Ursprungligt meddelande-----
Fran:   Corey Minyard [SMTP:[log in to unmask]]
Skickat:        Tuesday, May 11, 1999 6:30 AM
Till:   [log in to unmask]
Kopia:  Daniel Wengelin
Amne:   bug in assignment of controlled type by ancestor

This seems to be a rather serious bug in at least gnat-3.11p,
gnat-3.11b, and gnat-3.10p.  This is rather strange that it
hasn't been caught yet, I'm not really sure why it hasn't
been seen.  Daniel reported it to Team-Ada, I've taken the
liberty to report it to ACT.

Anyway, the following program raises a Program Error when
run.  It should produce no output.  I've debugged it some
under gnat-3.11p.  It seems that in e1__vehicle___assign,
the assigned variable is copied to the assignee, including
the value of My_Address in type Record_Controller in
System.Finalization_Implementation.  Then the value of
My_Address is then used in the Ptr_Adjust routine of the
Adjust routine of the same package, but it is obviously
bogus.

Daniel, the program is correct.
[Daniel Wengelin]  SNIP