Re: Need build for xulrunner
On Sun, Jun 25, 2006 at 11:04:57AM -0700, Peter Naulls wrote:
> In message <[🔎] 20060625161339.GV15411@country.grep.be>
> Wouter Verhelst <wouter@debian.org> wrote:
> > On Sun, Jun 25, 2006 at 08:51:34AM -0700, Peter Naulls wrote:
> > > for cross compiling. However, for cases like Mozilla and other large
> > > ARM packages, there does rather seem to be a strong motivation to do so.
> >
> > You think I don't know? Hint: I'm involved with the m68k port.
>
> Then you should be even more keen to address this for targetted
> packages.
I'm not.
> If things like xulrunner hold up much of the port, then that should
> also be significant motivation.
* To be able to cross-build packages, a cross-toolchain has to be
installed. There are no prebuilt packages for this, you need to build
them yourselves.
* Debian build-dependencies do not distinguish between stuff that's
needed because you need to run it, and stuff that's needed because you
need the .so or .a file, or some (possibly architecture-specific)
header files.
* Cross-building packages makes it rather troublesome to do test suites,
since you can't run target code on the compile host.
The above two combined mean that maintaining a cross-buildd host takes a
lot more effort than building packages natively. Sure, things like
scratchbox and distcc help, but they don't fix all of the above issues.
Additionally, you risk introducing hidden FTBFS or other bugs into
compiled packages due to the fact that your compiler is different from
the native compiler that is distributed with the target Debian port. And
since a machine that runs 24/7 can compile a surprisingly high amount of
software, even at 50Mhz, I'm not very keen on switching to
cross-compilation.
Of course, that doesn't mean you shouldn't do cross-compiled packages at
all; and if you can somehow guarantee the quality of the compiled code,
then by all means, go for it.
It's just that you seemed to assume that the slower Debian architectures
would probably cross-compile their stuff, and I wanted to point out that
this is, in fact, wrong, for very good reasons. That's all.
--
Fun will now commence
-- Seven Of Nine, "Ashes to Ashes", stardate 53679.4
Reply to: