Re: Advice requested: how big can we be?

[Andrew Haley <aph at redhat dot com>]

Karel Gardas writes:
 > On Tue, 16 Sep 2003, Andrew Haley wrote:
 > 
 > > Joe Buck writes:
 > >  > On Tue, Sep 16, 2003 at 05:50:32PM +0200, S. Bosscher wrote:
 > >  > > You are going to make it impossible for many people to contribute
 > >  > > to GCC or you'll have to accept that patches will be posted that
 > >  > > have not been tested with Java enabled, and that even fewer
 > >  > > people will be able to produce test results for GCJ.
 > >  >
 > >  > I already skip Java builds and tests most of the time because it is so
 > >  > expensive, and I'm sure that many others do the same.
 > >
 > > So am I.  But a Java compatible library is big, and there's nothing we
 > > can do about that.
 > 
 > I don't know gcj, but what about to divide one lib into several for
 > example following root packages? i.e. libgcjjavanet, libgcjjavalang,
 > libgcjjavautil etc.

We could do that.  It would be a bit cumbersome, so I'm not keen on
the idea.  I see the advantage, though.

 > > What I am interested to know is the class of boxes on which we can
 > > reasonably expect people to build gcc.
 > 
 > Slow (notebook) PIII-1GHz + 512MB RAM + 512 MB swap. Anyway I always buy
 > rather more memory than usuall since this machine should run for 3 years
 > and serve me well. I plan for next purchase (next year) to require 2GB
 > RAM.

Thanks.

Apart from the problem with the very large source build -- which we're
fixing -- the largest process on my box is 93M while liking libgcj.

We can make the libgcj build faster on some boxes by using a lot of
RAM.  On a 1Gbyte box the build would almost definitely be faster
building everything at once.

Andrew.