Re: Autobuilding with cross-compiler to solve m68k woes?
>>>>> "Steve" == Steve Langasek <vorlon@netexpress.net> writes:
Steve> What if we had the means of reliably determining which
Steve> packages could be cross-compiled, and which needed to be
Steve> compiled on a native system? Would that make
Steve> cross-compiling a viable option?
I don't think it is possible to reliably determine in a quick manner
eg. trying the cross compilation, and checking for any errors might be
a reliable way(?) of detecting any problems, but it wouldn't be very
efficient.
Take Heimdal, for instance. It would work fine right up to
compiling programs like compile_et, mk_cmds, and asn1_compile.
However, these programs are also used by for the compilation process,
hence, it wouldn't work. I think fixing this would be difficult using
current versions of automake (you would have to build to copies of the
above executables, one native, and one cross compiled).
On the other hand, other people have mentioned potential problems
related to configure tests. However, I am not convinced this really is
a problem.
Most configure scripts I have seen have tests for:
a) Does the compiler work? What parameters does it take?
Does ln -s work? Does ld work?
b) Does library x exist? Does it contain function y?
c) Does header file a exist?
which should work (assuming all libraries can be found) fine even if
cross compiling.
Last I checked, the autoconf info page strongly discouraged any tests
that prevent the program from being cross compiled. I have seen it
display warnings when it identifies potential problems with compiling
configure from configure.in.
--
Brian May <bam@debian.org>
Reply to: