[gentoo-dev] binary packaging

[Philippe Trottier]

Hello,

       I am trying to install a lot of systems using gentoo, so I did 4 
machines with distcc to compile things, and the rest of the machines 
will be using precompiled (emerge -k) I ran into some wierd suprises. 
Then I had a chat on #gentoo-dev.

This is a resume...

It is somehow difficult to know and sort all the possible binary 
packages. If the USE flags are different it does give some headache. One 
suggestion is to had a hex bit mask for each USE flags. for example 
package.w.x.y-rz-0A.tgz2 would mean USE flags  are set as follow 
00001010. At least that way someone can guess what package to fetch 
without having 1000 char long file names.

But then repoman should make sure that if USE flags are changed then the 
-r must also change. This won't tell what gcc has been used and may 
other dependecies would cause troubles.

Now if someone uses xar, he could add all or the relevent parts of 
emerge info in the archive. portage could make a sanity check and 
recompile anyway if he thinks the package is too far. Other way to still 
speed up the business would be to inject the precompiled data in ccache 
directly. If the preprocessed code is the same and the compiler is the 
same identical result are to be expected.

xar alone might prove difficult to handle, big file containing all, 
loads of BW ... bad !

Subdirectories should be made under packages, at least separate them by 
arch, I separate them by cpu-gentoo release , like athlon-mp-2004.2, 
7400-2004.3, pentium4-1.4 etc etc works for me...

emerge -k ... seems not to care about PDEPEND for some reason, but I 
might be doing something wierd.

If you have better ideas, please share them. Ill manage myself with the 
current system anyway.

Yours,
Philippe

--
gentoo-dev@g.o mailing list