[Qemu-devel] gccism's are ok?

[Qemu-devel] gccism's are ok?

[Gerd Hoffmann]

  Hi,

Quick question:  Is it ok to use gcc-specific features in qemu?  Or will
that break builds on some platform?

Linux obviously is gcc, MacOS X too as far I know.  MinGW on Windows is
gcc too.  Anything else we care about?

I'm thinking especially about the constructor attribute.  That would
allow to make drivers self-contained, by having a register() call in a
small constructor function ...

cheers,
  Gerd

-- 
http://kraxel.fedorapeople.org/xenner/

Re: [Qemu-devel] gccism's are ok?

[Glauber Costa]

On Mon, Aug 11, 2008 at 10:24 AM, Gerd Hoffmann <kraxel@redhat.com> wrote:
>  Hi,
>
> Quick question:  Is it ok to use gcc-specific features in qemu?  Or will
> that break builds on some platform?
>
> Linux obviously is gcc, MacOS X too as far I know.  MinGW on Windows is
> gcc too.  Anything else we care about?
>
> I'm thinking especially about the constructor attribute.  That would
> allow to make drivers self-contained, by having a register() call in a
> small constructor function ...

I'm doing that in QEMUAccel, and although I'm also not sure how well
it'll go, nobody complained
so far.

>
> cheers,
>  Gerd
>
> --
> http://kraxel.fedorapeople.org/xenner/
>
>
>



-- 
Glauber Costa.
"Free as in Freedom"
http://glommer.net

"The less confident you are, the more serious you have to act."

Re: [Qemu-devel] gccism's are ok?

[Anthony Liguori]

Gerd Hoffmann wrote:
>   Hi,
>
> Quick question:  Is it ok to use gcc-specific features in qemu?  Or will
> that break builds on some platform?
>
> Linux obviously is gcc, MacOS X too as far I know.  MinGW on Windows is
> gcc too.  Anything else we care about?
>
> I'm thinking especially about the constructor attribute.  That would
> allow to make drivers self-contained, by having a register() call in a
> small constructor function ...
>   

You'll have to post a patch.  I had discussed using GCC-isms to do the 
same thing with Paul Brook and at the time, he was not a big fan of it IIRC.

Regards,

Anthony Liguori

> cheers,
>   Gerd
>
>   

Re: [Qemu-devel] gccism's are ok?

[Johannes Schindelin]

Hi,

On Mon, 11 Aug 2008, Anthony Liguori wrote:

> Gerd Hoffmann wrote:
>
> > Quick question:  Is it ok to use gcc-specific features in qemu?  Or 
> > will that break builds on some platform?
> >
> > Linux obviously is gcc, MacOS X too as far I know.  MinGW on Windows 
> > is gcc too.  Anything else we care about?
> >
> > I'm thinking especially about the constructor attribute.  That would 
> > allow to make drivers self-contained, by having a register() call in a 
> > small constructor function ...
> >   
> 
> You'll have to post a patch.  I had discussed using GCC-isms to do the 
> same thing with Paul Brook and at the time, he was not a big fan of it 
> IIRC.

Concretely for the suggestion to replace the nice, deterministic and 
straight-forward registering of drivers with a gcc dependent, unintuitive 
and hard to debug solution, I maintain that it is not a good idea, 
and this has nothing to do with the usage of a gccism.

Ciao,
Dscho

Re: [Qemu-devel] gccism's are ok?

[Thiemo Seufer]

Gerd Hoffmann wrote:
>   Hi,
> 
> Quick question:  Is it ok to use gcc-specific features in qemu?  Or will
> that break builds on some platform?
> 
> Linux obviously is gcc, MacOS X too as far I know.  MinGW on Windows is
> gcc too.  Anything else we care about?

The dyngen targets still require gcc3 to build, OTOH making qemu more
portable is valuable. I believe the policy should be "avoid compiler
extensions unless it is hard to do without".

> I'm thinking especially about the constructor attribute.  That would
> allow to make drivers self-contained, by having a register() call in a
> small constructor function ...

This idea is IMHO of dubious value. It obfuscates the control flow, and
it wouldn't work for a plugin system (where self-containedness would be
the biggest win).


Thiemo