Bruce Dubbs bruce.dubbs at
Mon Oct 6 17:13:23 PDT 2008

DJ Lucas wrote:
> Jeremy Huntwork wrote:
>> DJ Lucas wrote:
>>> I do not have all the details in front of me, but somebody said that gcc 
>>> failed if gmp was not on the host in pass1.  Greg pointed us to a DIY 
>>> thread that showed how to build with GCC.  There was no _need_ to build 
>>> inline beyond that point, because that fixed the problem of the extra 
>>> host system requirement.  The extra packages are minimal, and required 
>>> one way or the other.  Shared seems better IMO as I use GMP anyway.  It 
>>> might not be a bad idea (less maintenance) to build the same way for 
>>> both early builds and only build gmp and mpfr standalone for the final, 
>>> though I haven't tried that yet.
>> I was going to ask 
>> what you thought about making both passes in chapter 5 the same, but 
>> you've touched on that as well. I don't suppose it makes a big 
>> difference there. The only advantages I can see for building inline for 
>> both passes are perhaps a slightly smaller set of temporary tools, a 
>> marginal amount of time saved by the user (especially if building by 
>> hand) and, again, slightly less chance of error.
> You forgot the editor's responsibility.  If something changes in GMP at 
> a later date, it would only have to be changed in one place instead of 
> two.  It's worth investing the time to make and test that change IMO.  
> I'll try it in my Wednesday build.

Typically, the changes for a package are minimal -- a change in md5sum, etc.  A 
patch would require changes in two places, but for these packages, that is 
pretty unlikely.

   -- Bruce

More information about the lfs-dev mailing list