It's arguably a toolchain deficiency on all arches that the ELF headers
are incomplete when linking with nostdlib, what makes this special on
armhf is that we later rely on those headers being in place. Given that
this has bitten us exactly nowhere in the entire archive except for ICU,
however, I can't see that this is a particularly common use-case.
RedHat/Fedora and Ubuntu[1] all just link ICU with libc6 to avoid
annoyance from ELF-lint checkers, prelinking, and in the Ubuntu case,
armhf. There's no real argument against it, especially in the Debian
package case, since libicu depends on libc6 anyway, and the only way
that people load libicudata is via other libicu* consumers, which all
link with libc6, so libc6 is already both on-disk and in-memory anyway.
I understand what upstream was trying to do here ("hey, it's a data
blob, not an executable, so let's not link it to anything"), but I don't
see any practical purpose in it, except as a clever trick.
At any rate. It would be lovely if you consume the Ubuntu patch
more-or-less as-is, but if you'd like to dig deeper into the madness of
why the toolchain gives incomplete headers with -nostdlib
-nodefaultlibs, be my guest. Either way, enough things depend on ICU
that it should be fixed soon. Pretty please? ;)
I'll happily NMU so you can wash you hand of the whole thing, if you
prefer. :P
... Adam
[1] http://patches.ubuntu.com/i/icu/icu_4.8.1.1-1ubuntu1.patch
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
