Exception propagation with optimization flags

Exception propagation with optimization flags

[ldb]

Can someone explain why this particular combinations of build flags
causes exceptions to stop propagating. I'm kind of at a loss for this.
This is a nested function, which raises, catches, prints some info and
reraises an exception, and a main program which is trying to catch this
exception, to print some further information. I've created a "minimal"
program to show the problem. This is gnatmake/gcc 3.4.5.

And, if anyone is an expert, can you explain to me the connection (if
there is one) with "libunwind-exceptions" in my gcc ./configure.

When I build with -O2 and -fomit-frame-pointer I get this:
EXCEPTION

raised SUPER.AWESOME.AWESOME_EXCEPTION : super.adb:12

If I remove -fomit-frame-pointer _or_ go down to -O1 (both o2 and fomit
are required to get the behavior), I get this:
EXCEPTION
HOLY MOLY


---------------------
with Ada.Text_Io; use Ada.Text_Io;

procedure Super is

   procedure Awesome is
      Awesome_Exception : exception;
   begin
      raise Awesome_Exception;
   exception
      when others =>
         Put_Line("EXCEPTION");
         raise Awesome_Exception;
   end Awesome;


begin
   Awesome;
exception
   when others=>
      Put_Line("HOLY MOLY");
end Super;
---------------------

Re: Exception propagation with optimization flags

[Stephen Leake]

"ldb" <ldb_nospam@hotmail.com> writes:

> Can someone explain why this particular combinations of build flags
> causes exceptions to stop propagating. 

You need to say what compiler and OS you are using, and give the
complete command lines.

-- 
-- Stephe

Re: Exception propagation with optimization flags

[ldb]

Stephen Leake wrote:
> "ldb" <ldb_nospam@hotmail.com> writes:
>
> > Can someone explain why this particular combinations of build flags
> > causes exceptions to stop propagating.
>
> You need to say what compiler and OS you are using, and give the
> complete command lines.
>
> --
> -- Stephe

I did, I thought.

gnatmake 3.4.5 (Red Hat 3.4.5-2)
Linux (Red Hat)

gnatmake -fomit-frame-pointer -O2 super.adb (doesn't work)
gnatmake -fomit-frame-pointer -O1 super.adb (works)
gnatmake -O2 super.adb (works)

Re: Exception propagation with optimization flags

[Georg Bauhaus]

On Wed, 2006-10-04 at 06:31 -0700, ldb wrote:
> Stephen Leake wrote:
> > "ldb" <ldb_nospam@hotmail.com> writes:
> >
> > > Can someone explain why this particular combinations of build flags
> > > causes exceptions to stop propagating.

> gnatmake -fomit-frame-pointer -O2 super.adb (doesn't work)
> gnatmake -fomit-frame-pointer -O1 super.adb (works)
> gnatmake -O2 super.adb (works)
> 

And, earlier,
> And, if anyone is an expert, can you explain to me the connection (if
> there is one) with "libunwind-exceptions" in my gcc ./configure.

A good place to ask is probably the GCC mailing list. Specific
GCC configurations and their effects on front-ends are probably
best known there. A vanilla GCC 42 with Ada enabled compiles
your programs such that it prints/handles exceptions as
expected.


-- Georg 

Re: Exception propagation with optimization flags

[Stefan Bellon]

ldb wrote:

> gnatmake -fomit-frame-pointer -O2 super.adb (doesn't work)
> gnatmake -fomit-frame-pointer -O1 super.adb (works)
> gnatmake -O2 super.adb (works)

I recently had a similar problem on Solaris 8 using the GNU ld/as
instead of the Sun ones. In that case, static linking didn't catch
exceptions. However when I added "-bargs -shared" to the gnatmake call,
the exceptions were caught again. Using the Sun ld/as instead of the
GNU ld/as however catches exceptions regardless of whether linking
-static or -shared.

-- 
Stefan Bellon