[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Are we losing users to Gentoo(tangent)?



On Thu, Nov 21, 2002 at 01:06:15AM -0500, Carlos O'Donell wrote:
> > For example, I have the following USE flags set: 
> > cdr dga samba jpeg png 3dnow mmx mpeg quicktime truetype xv x86 -alsa
> 
> That's more than just compiler flags.
>  
> > The above determine how apps configure for compiling. I don't use ALSA, so 
> > nothing on my system compiles with ALSA support.
> 
> Faire enough.

Just a random note. This is not a gentoo invention. FreeBSD is doing this
since years. The ports are (should) be configured to watch the global make
file. And the "make (build)world" system is very easy to recompile your
system. There is something to it.

> Arbitrary specification of CFLAGS is going to break something at some
> point. Debian developers continually work around bugs through the use of
> these option variables.

Of course it would be fine to fix gcc and the apps which break on O3. I am
not completely shure on the state of O3, but on ia32 the O3 option used to
be the same as O2.

I know that some applications do have problems with optimizing in case of
missalligend data structures, because the build in memcpy for exampel was
not always able to deal with that. This is why with some gcc versions
networking tools break with O2, cause the used memcpy does not work (and
bcopy does) on sparc, the program will get SIGBUS.

> apt-get install xxxx
> *under the hood*

...

Full ack. Well, personally Debian source packages are so standadized, we do
not have to hide, we are nearly as good as the ports system or gentoo. We
just need to communicate this: yes you can rebuild your system.

Global variables is about the only thing which is missing, because it is not
100% shure it buys something.

> And we already distribute a 180MB net-install ISO which we can call 
> Debian Stage 3 Build system :}

Indeed, this is a good point, adding to that.

> Note that Danjou worked on a gcc wrapper to do some of this, but I still
> think that we need a framework inplace to do the "item" configuration in
> a generic fashion...

Especially the "use" flags as in "compile with support for", to control
which options are passed to configure scripts are interesting. One can think
about turning on/off features like crypto, ipv6, debugging but also
customizing the feature sets for embedded distributions, profiling and what
ever.

Greetings
Bernd



Reply to: