RE: Using EH from C

RE: Using EH from C

[david daney]

On Sat, 1 May 2004, Richard Henderson wrote:

>> On Sat, May 01, 2004 at 05:09:23PM +0300, Stelios Xanthakis wrote:
>> > But of little use without another extension to create landing pads...:(
>>
>> Why in the world would you need that?

>Because if there's no landing pad, _Unwind_RaiseException
>ends up on abort()?
>
>I may be totally wrong, but what I'm trying to do is implement
>"stack unwind with cost-free cleanups from C".
>
>That's three things
>
>1. register destructors on EH
>  Done: attribute(cleanup(function))
>
>2. call 'throw'
>  Done: libgcc_eh and unwind.h
>
>3. set up a landing pad somewhere (try-catch).
>  ???
>
>Where does _Unwind stop otherwise? main()?
>
>~S


gcc/testsuite/gcc.dg/cleanup9.c kind of does what you are talking about.  I would setjump and set an attfibute(cleanup(function)) at the catch point, and then longjump from the cleanup function.
 
Why can't you use C++ for the try/catch part?  Since all of this is GCC specific, if you have C and excpetions working it doesn't seem like using G++ would be a problem.
 
David Daney.
 

RE: Using EH from C

[Stelios Xanthakis]

On Sat, 1 May 2004, david daney wrote:

> 
> I would setjump and set an attfibute(cleanup(function))
> at the catch point, and then longjump from the cleanup function.
>  

Not a bad idea. Like this?

	if (setjmp (&context)) {
		struct pseudo_object X __attribute__((cleanup(pdtor)));
		X.normal_termination = 0;
		X.ctx = &context;
		/* ... */
		X.normal_termination = 1;
	} else {
	}

And pdtor() will either just return or longjmp to ctx depending
on normal_termination.

There's still a minor cost but I guess it's ok.

Thanks,

Stelios

RE: Using EH from C

[Stelios Xanthakis]

On Sun, 2 May 2004, Stelios Xanthakis wrote:

> On Sat, 1 May 2004, david daney wrote:
> 
> > 
> > I would setjump and set an attfibute(cleanup(function))
> > at the catch point, and then longjump from the cleanup function.
> >  
[snip] 
> And pdtor() will either just return or longjmp to ctx depending
> on normal_termination.
> 

Although the manual sais:
``  Note that the `cleanup' attribute does not allow
      the exception to be caught, only to perform an action.  It is
      undefined what happens if CLEANUP_FUNCTION does not return
      normally. ''

Seems like somebody already thought that somebody else might
think of using 'cleanup' to catch exceptions :)

For the moment it works and I hope we can resolve this in the
next version. If the only problem is the un-freed struct
_Unwind_Exception, we can take care of it.


stelios

Re: Using EH from C

[Richard Henderson]

On Mon, May 03, 2004 at 02:17:56PM +0300, Stelios Xanthakis wrote:
> Seems like somebody already thought that somebody else might
> think of using 'cleanup' to catch exceptions :)

See glibc's pthread unwind implementation.  The correct place to 
put the longjmp is in the force unwind handler.


r~

Re: Using EH from C

[Stelios Xanthakis]

On Sat, 1 May 2004, Richard Henderson wrote:

> On Sat, May 01, 2004 at 05:09:23PM +0300, Stelios Xanthakis wrote:
> > But of little use without another extension to create landing pads...:(
> 
> Why in the world would you need that?

Because if there's no landing pad, _Unwind_RaiseException
ends up on abort()?

I may be totally wrong, but what I'm trying to do is implement
"stack unwind with cost-free cleanups from C".

That's three things

1. register destructors on EH
  Done: attribute(cleanup(function))

2. call 'throw'
  Done: libgcc_eh and unwind.h

3. set up a landing pad somewhere (try-catch).
  ???

Where does _Unwind stop otherwise? main()?

~S

Re: Using EH from C

[Richard Henderson]

On Sat, May 01, 2004 at 05:09:23PM +0300, Stelios Xanthakis wrote:
> But of little use without another extension to create landing pads...:(

Why in the world would you need that?


r~

Re: Using EH from C

[Stelios Xanthakis]

On Fri, 30 Apr 2004, Stelios Xanthakis wrote:

> 
> On Thu, 29 Apr 2004, Richard Henderson wrote:
> 
> > On Thu, Apr 29, 2004 at 02:07:18PM +0300, Stelios Xanthakis wrote:
> > > The main challenge is 'calling destructors on unwind
> > > with no performance cost if  no exception is raised'.
> > 
> >   type var __attribute__((cleanup(function)));
> 
> That's just great.

But of little use without another extension to create landing pads...:(

Or is there?

Re: Using EH from C

[Stelios Xanthakis]

On Thu, 29 Apr 2004, Richard Henderson wrote:

> On Thu, Apr 29, 2004 at 02:07:18PM +0300, Stelios Xanthakis wrote:
> > The main challenge is 'calling destructors on unwind
> > with no performance cost if  no exception is raised'.
> 
>   type var __attribute__((cleanup(function)));

That's just great.

Thanks.

Stelios

Re: Using EH from C

[Mike Stump]

On Thursday, April 29, 2004, at 06:48 PM, Richard Henderson wrote:
> On Fri, Apr 30, 2004 at 02:36:24AM +0200, Giovanni Bajo wrote:
>>>   type var __attribute__((cleanup(function)));
>>
>> It looks like this was never applied though:
>
> That would be incorrect.  Glibc uses it today.

The link was to some doc content.  Lack of that doesn't prevent glibc 
from using it.

Re: Using EH from C

[Richard Henderson]

On Fri, Apr 30, 2004 at 02:36:24AM +0200, Giovanni Bajo wrote:
> >   type var __attribute__((cleanup(function)));
> 
> It looks like this was never applied though:

That would be incorrect.  Glibc uses it today.


r~

Re: Using EH from C

[Giovanni Bajo]

Richard Henderson wrote:

>> The main challenge is 'calling destructors on unwind
>> with no performance cost if  no exception is raised'.
> 
>   type var __attribute__((cleanup(function)));

It looks like this was never applied though:
http://gcc.gnu.org/ml/gcc-patches/2003-06/msg00658.html

Giovanni Bajo

Re: Using EH from C

[Richard Henderson]

On Thu, Apr 29, 2004 at 02:07:18PM +0300, Stelios Xanthakis wrote:
> The main challenge is 'calling destructors on unwind
> with no performance cost if  no exception is raised'.

  type var __attribute__((cleanup(function)));


r~

Using EH from C

[Stelios Xanthakis]

Hi.

I'm working on a 'C++ to C' code generator and I'm trying
to figure out (how|if) it's possible to use unwind info from
C programs. The main challenge is 'calling destructors on unwind
with no performance cost if  no exception is raised'. This probably
means that we will have to construct unwind tables manually
by getting the addresses of labels and putting them in  special
sections. And the -funwind-tables?

If anybody knows, please contact me.

Thanks and sorry for the OT but this seems
to be a quiet period on the list, I couldn't find any
info on google and Aldy said I should ask here.

stelios


- The project is [http://students.ceid.upatras.gr/~sxanth/lwc]