Long-term mass bug filing for crossbuild support
Pierre THIERRY wrote:
> Scribit Neil Williams dies 13/11/2007 hora 17:02:
>> If you want to build an ARM toolchain to crossbuild for amd64 I'm not
>> going to stop you but don't expect me to debug it!!
> But do your tools make it already possible for me to just ask for the
> build of toolchains for an arbitrary list of target architectures, from
> an arbitrary host architecture?
No. Building a toolchain relies on various levels of support packages
which may or may not be available for your choice. You would have to
build those first. The tools make no promises about whether the choice
is sensible, reasonable or actually going to build. Quite often, it will
not build - you can't blame the tools if the underlying packages don't
work in some weird combo.
There is a vast gap between allowing the tools to support the theory and
actually supporting real toolchain builds in practice. I don't think
that it is worthwhile trying to attain the kind of arbitrary selections
that you describe. I do want to actually get things done with the
toolchains that we already have, I'm not going to spend the rest of my
life building every permutation of every architecture combination for
every compiler version and the rest.
I'm not going to promise support for all possible permutations because
that implies debugging bizarre versions. The tools don't rule it out,
that's about the best I can offer.
Not all permutations of toolchains will ever be possible to build, the
tools cannot do anything about that.
So, in theory, all architectures supported by dpkg-architecture (which
is already far more than supported by Debian itself) can be used with
the tools. In practice, the only "reasonable" choices are to install
toolchains on i386, amd64 or powerpc. Even then, some toolchain packages
will simply fail to build for certain host selections.
Throw in requests for older compilers and even the "regular" selections
> I can well understand why you don't want the N*N-N packages or sets of
> packages in the official Debian archive, but it would be great if the
> tools make it possible to build them.
Possible - yes. Practical ? maybe not. Supported? Definitely NO.
> There's also the fact that it
> could be useful for ports not in the released ones.
Even porters don't need every possible permutation. Some combinations of
--build and --host are just more practical than others. I see no point
in extending the tools to "support" these in anything more than theory.
> In any case, thanks to all developers from Embdebian! That's a great