Re: [Libvisual-devel] VisError

[Gustavo S. B. <gu...@gs...>]

On Wednesday 13 October 2004 12:20, Dennis Smit wrote:
> On Wed, 2004-10-13 at 11:58 -0300, Gustavo Sverzut Barbieri wrote:
> > On Wednesday 13 October 2004 11:54, Gustavo Sverzut Barbieri wrote:
> > > On Monday 11 October 2004 19:30, Dennis Smit wrote:
> > > > Alright, it took me a while, but I've put some thought into this.
> > > >
> > > > 1. There won't be special reentrant functions.
> > > >
> > > > Why ?, Because when we implement an advanced error report system
> > > > we will do it reentrant by design, everywhere. If we go this road
> > > > I'd like to do it the following:
> > > >
> > > > Last function argument is of type VisError or something.
> > > > NULL can always be passed.
> > > >
> > > > It's no problem if the core breaks, Rather do it good once, than
> > > > work around half solution, or providing multiple broken interfaces.
> > > >
> > > > I personally totally go for the last argument is VisError * solution
> > > >
> > > > What do you guys think ?
> > > >
> > > > I kinda formulated my start sentence as if this is final, it's
> > > > ofcourse not, so if someone wants to discuss this, please go ahead.
> > >
> > > Dennis,
> > >
> > > Talking to a friend he mentioned TLS (Thread Local Storage). We can use
> > > that and the error will be local to the thread it happened... all for
> > > free (glibc does that).
> > >
> > > But someone need to look at it...
> >
> > Ah, the bad point is: it just work on linux and solaris... other's don't.
>
> Well I want portability. What do you think about having an extra
> VisError ** where it's needed, ofcourse far from every function really
> needs VisError. Glib does it the same way with GError.

It's ok with me.

-- 
Gustavo Sverzut Barbieri
---------------------------------------
Engenharia de Computacao 2001 - UNICAMP
GPSL - Grupo Pro Software Livre
Cell..: +55 (19) 9165 8010
Jabber: gsb...@ja...
  ICQ#: 17249123
   GPG: 0xB640E1A2 @ wwwkeys.pgp.net