Re: Re: How is it possible ???

On Sun, 2005-01-09 at 00:34 +0100, Paul de Vrieze wrote:
> On Saturday 08 January 2005 08:22, Alec wrote:
> >
> >     I would note, that not all(most?) devs are willing to work on bugs
> > when you went out of your way to strip out the protections that they
> > left in place.  Some of them are necessary, some are probably not (
> > outdated or somesuch ).  The problem then becomes when you have a bug in
> > the system thats app related, is it due to the app, or due to your
> > compiling flags?  Obviously some bugs are easy to spot where there will
> > be no correlation, but especially relating to app crashes and system
> > stability, it's not easy to figure out.
> 
> Also, (normally it's hard to find out if someone went out of his/her way) even 
> if a dev is willing to help, these kinds of bugs are often extremely hard to 
> diagnose. And that's only the bugs comming from broken compiler behaviour. 
> Besides there are still people that use (normally not filtered) what I call 
> silly flags. These are flags that are not appropriate for a general level. 
> They are clearly stated in the info/man page as being so.
> 
> Examples are:
> *) -fno-rtti, this has no use for, C, but in c++ disables a part of the
>       languagefeature that is extremely usefull, and not uncommon either.
>       Setting this globally will break many c++ programs.
> *) -ffast-math, this is more subtle, but some programs do depend on accurate
>       floating point math according to the IEEE specifications. Fast-math will
>       break this.
> *) -fno-exceptions, a similar C++ feature, that can be used in normal c++
>       control flow, not only in exceptional cases. This also can only be
>       disabled on a program basis when exceptions are not being used.
> *) Many others
> 
> The normal suggestion for these (or very difficult flags) is: "Get sane flags, 
> try again, and see whether it works" If it does, the user is suggested to 
> drop the flags. If there is a known broken flag normally one does just state 
> that.
Sometimes its not that easy.  gnome-control-center's theme mangager
wouldn't work, even after compiled with -O0 for me.  Turns out I needed
to be more sane with gnome-theme-[something].  That took a while to
figure out, but just shows that b0rking one package, may affect others
as well.  

> 
> Paul
> 
> ps. There are also very hard packages like glibc/xorg-x11 that are very 
> fragile, have optimization dependend bugs, and that might trigger compiler 
> bugs by uncommon code. Those are normally heavilly filtered to ensure that 
> these important packages build easilly and (Even more importantly, don't get 
> build wrongly but unnoticed, fucking up your system)

This may not be the place to ask, but does -fweb work alright now?  I
tried it a while ago and a whole mess of things started breaking.
-- 
 .+"+.+"+.+"+.+"+.+"+.+"+.+"+.+"+.+"+.+"+.+"+.+"+.+"+.+"+.+"+.
( AIM:    EvilComputer69                    YIM:  MighMoS     )
 )MSN:    [email protected]        ICQ:  161301521  (
( Jabber: [email protected]                SILC: MighMoS     )
 "+.+"+.+"+.+"+.+"+.+"+.+"+.+"+.+"+.+"+.+"+.+"+.+"+.+"+.+"+.+"

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.0 (GNU/Linux)

iD8DBQBB4IVqN7YJmgYMk4YRAnaWAKCe63Hy+xG+5QFgNP8CLg4S0AhdoQCeP0k0
RKJKrIXm4EEuGvgN+foE0Hc=
=vzsD
-----END PGP SIGNATURE-----