Re: x86 release builds are slow

[David Laight <david%l8s.co.uk@localhost>]

On Thu, May 01, 2008 at 03:08:24AM +0000, Steven M. Bellovin wrote:
> > 
> That would be a nice direction.  However, there's another approach that
> may be worth talking about: have a single repository for shared .o
> file, and build+link kernels with a bit of unique source plus the
> shared .o files.  (BSD/OS worked that way, since it was mostly not a
> source distribution.)
> 
> The big problem is compile-time options.  It would be nice to get rid
> of many of them...  Failing that, would it be possible to note the
> options used to build the repository -- a .a? -- and do separate
> compilations for any source files that (a) use a given option and (b)
> don't use the options used for the .a?

I tried doing something to cache the .o files with a hash based on
the kernel options in each compilation.
Unfortunately the only files build with matching kernel options
differ due to -Dfoo lines passed to the compiler!

IMHO far to many of the kernel options are included in header files,
so (potentially) affect every object file.

There are also 1000s of kernel options ...

        David

-- 
David Laight: david%l8s.co.uk@localhost