[CLUE-Tech] Gentoo build questions

[Nate Duehr]

Collins Richey wrote:

>Not a good idea (tm) at bootstrap time. Complete stage 1 and 2 with
>"x86", then switch over to "~x86". Since you haven't taken that
>approach, you have learned that there are some speed bumps <grin>.
>Using "~x86" at bootstrap time means that you have just declared
>yourself as a test laboratory for an environment that may never have
>been tested by any developer. You're out of bounds, and there are lots
>of "fun" possibilities.
>
>  
>
[snipped the rest about ~x86]

Good info, thanks.  Pretty much what I had read.  Hadn't made the 
connecting in my head between the bootstrap on ~x86 and the problems yet 
though -- that makes perfect sense.

>>I understand it's *supposed* to be somewhat unstable, but the types of
>>things I'm running across are really childish... like the build scripts
>>for stuff in "emerge system" calling autoconf 1.8 directly instead of
>>just "autoconf", so they break unless you emerge autoconf 1.8
>>specifically into the system first.
>>    
>>
>
>I've never encountered this. Typically, all appropriate versions of
>autoconf/automake should be merged. They are "slotted," i.e. multiple
>versions can coexist. I have the following, and none of my ebuilds are
>failing. But then I'm long ago past the bootstrap stage.
>
>autoconf-2.13
>autoconf-2.59-r6
>autoconf-wrapper-1
>automake-1.4_p6
>automake-1.5
>automake-1.6.3
>automake-1.7.9
>automake-1.8.5-r2
>automake-1.9.3
>automake-wrapper-1
>  
>
[snipped more automake/autoconf stuff...]

Ahh, further investigation last night showed what was going on... 
automake-wrapper and autoconf-wrapper (to get the "slots" you mention) 
simply hadn't been emerge'd yet.  They need to move those higher in the 
food chain for dependencies and all will be well.  coreutils and 
libtools don't like not having the version of automake they desire!


>>Just some discussion-fodder.
>>
>>Next, does anyone know the proper -march settings for a Via C3 CPU? 
>>    
>>
>
>Nary a clue.
>  
>
Just to follow up on this:

You can't do it during the bootstrap because the gcc version is still 
too old, but after gcc is rebuilt, '-march=c3' works great.

>  
>
>>Finally -- I've seen a few sites that recommend using "-Os" to optimize
>>for size on this processor, but I really don't understand the purpose.
>>Any compiler gurus want to tackle that one?  I'm just curious is all.
>>    
>>
>
>I would forget about this, I did some experimentation with this and
>got no improvement, not only that, I think it caused some instability.
>Some ebuilds specificly strip out "-Os". I just use "-O2
>-fomit-frame-pointer" for everything.
>  
>
Ahh bummer.  I guess I'll change it.  ;-)

Thanks for the thoughts.  The build is going much better now. 

--
Nate Duehr, nate at natetech.com