python linking hack

[tridge at samba.org]

Hi Jelmer,

 > I've converted the build system to Python, have added tests for it and
 > have been doing some crazy refactoring. So far I've gotten the size of
 > bin/ down to 440 Mb and the overhead of the makefile generator is now
 > less than a second (down from 3 or 4 here locally) but there is still
 > more to be fixed.

That's a little more of a radical than I expected :-)

In case it helps, I did a little more hackish work in my linking-wip
branch, and got bin/ down to about 240M, most of which is
bin/mergedobj/ and bin/shared/, which don't actually have to be
included in Samba packages.

What basic linking approach are you thinking of adopting? I used a
"one big shared lib" approach for my experiments, which works quite
well I think, but it means the binaries need re-linking on install. Is
that what you are looking at using?

I think we also need to make sure we fix the other symptom of our poor
linking, which is data being duplicated. The most obvious example of
this is things like having DEBUGLEVEL appear more than once in a
binary, so setting the debug level in one bit of C doesn't necessarily
set it in another bit of C. 

Cheers, Tridge