Re: GNAT V2.07 (mips-sgi-irix5.3) Gigi abort, Code=116

["John H. Dibbs" <johnd@faac.com>]

John G. Volan wrote:
> 
> Pascal Obry wrote:
> >
> > "John H. Dibbs" <johnd@faac.com> wrote:
> >
> > >HELP!!!
> >
> > >Porting a fairly big app from VADS ada to gnat.
> > >Get a GNAT BUG DETECTED message, and an advisory to
> > >call 1-800-800-4SGI. Still awaiting them to return my call.
> >
> > >GNAT V2.07 (mips-sgi-irix5.3) Gigi abort, Code=116
> >
> > >Have recently upgraded to iris 6.2
> >
> > First of all you must upgrade to the 3.05 version of GNAT.
> >
> > http://www.gnat.com/
> >
> > Maybe this bug is already fixed.
> >
> > Hope this help,
> > Pascal.
> 
> Pascal, your advice would be sound only if John Dibbs were simply using
> plain, unadorned GNAT.  However, since he quotes SGI's 1-800 number as
> showing up in the Gigi abort message, I assume that he's using SGI's
> port of GNAT, which I believe exploits SGI's proprietary thread library.
> I haven't used this version of GNAT for several months, but the last I
> heard was that SGI hasn't (officially) released anything beyond GNAT
> 2.07 yet, so John may be out of luck.
> 
> On the other hand, I assume that, since John's organization has a copy
> of SGI's port, they must have bought it from SGI, and it would follow
> that they'd have a support contract with SGI (and, by extension, with
> ACT).  John, why not call up the 1-800 number and ask for help right
> from the horses mouth?  :-)
> 

I did talk to SGI, eventually (1 week for response).
By that time I found a work-around.

Problem encountered with a for use at.

Code notionally was:

X : float := 0;
for X use at memory_manager.next_address( X'size, "SUBSYSTEM.X" );

We use memory_manager to map things to ScramNet.
Gnat insisted that the at value had to be a prior define constant.
So changed the code to:

X_ADDRESS : constant system.address :=
    memory_manager.next_address( X'size, "SUBSYSTEM.X" );

X : float := 0;
for X use at X_ADDRESS;

This worked until in one unit I reversed the order of the first two
lines of code.

X : float := 0;
X_ADDRESS : constant system.address :=
    memory_manager.next_address( X'size, "SUBSYSTEM.X" );

for X use at X_ADDRESS;

Found what the problem was when attempted to make this change in a spec
where gnat still insisted that the at value had to be a prior defined
constant.

In a body, got the Gigi abort 116 instead. Unsure whether it was
the fact that it was in the body, or whether the new declaritive
block inside the body had anything to do with the error.

But reversing the first two lines of code got rid of the abort code.

Hope this helps someone else.