[distcc] Bootable distcc node

[Dag Wieers]

On Tue, 16 Mar 2004, Dan Kegel wrote:

> Dag Wieers wrote:
> 
> > I don't agree that RPM is useless on non-linux boxes. RPM runs on all Unix 
> > boxes and even in cygwin. Even when you're not using RPM as a package 
> > manager, you can still use RPM SPEC files (or source packages) to build 
> > binaries for a specific platform.
> 
> Beware.  I did package up these toolchains as RPM files, and
> the resulting packages were so large they took 15 minutes to
> install.  Unpacking a .tgz, by comparison, took about 5 minutes.
> Maybe rpm is more optimized by now.  If not, this is probably
> an easy performance bug to fix.

Hmmm, tgz takes 5 minutes ? What size are we talking about ? I'd guessed 
it was 1 cross-compiler per architecture. About 3Mb in size each.

I'd go for 1 package for each compiler (as most people want to select 
their own set of compilers depending on the hardware they have).


> It's even worse when you try to use alien to install these
> large RPMs on Debian.  It's way slow, and I've seen people
> run out of disk space with the extra temp copy.

Strange, as RPM was always considered much faster and less memory 
hungry as dpkg in handling packages. I remember the Debian project 
considering of moving to a database backend too, so with alien I guess you 
have to add that up.


> > I agree with you that it probably isn't the best solution for Windows 
> > though. My interest is more in the general case. (and Red Hat 
> > specifically)
> 
> Unless the performance issues are addressed, you may find RPMs
> aren't the best way to handle these large packages.
> 
> By the way, since there are so many combinations of processor, gcc version,
> and glibc version, it would be hard to build all the RPMs for those
> combinations.

True. But there's no need to compile all combinations and RPM may help in 
building for different architectures too. I haven't looked at your script, 
but I'd guess RPM does a lot of things already that you have in your 
script. (build and target specification, compiler-flags, ...)


> If you're supporting a stable of engineers working on the same
> project, it's worth building the combinations you need once
> (possibly as an RPM, if that's how your site likes things).
>
> But Fedora probably wouldn't want to include all the possible
> RPMs my script can spit out.
> This is a case where package managers have a bit of
> a disadvantage compared to an engineer downloading the build script
> from http://kegel.com/crosstool and just building the damn thing themselves.

I respectfully disagree. I see what you mean, but even Source 
RPM packages are very handy to build for different targets with different 
options from a single source. So even when I wouldn't have the combination 
some engineer need, I guess if crafted good, a single SRPM would suffice.

So when I find the time I'm going to look into your tool and see how I can 
take advantage of it as I've been always interested in providing this.

Thanks Dan,
--   dag wieers,  dag at wieers.com,  http://dag.wieers.com/   --
[Any errors in spelling, tact or fact are transmission errors]