Re: kernel depends?
Steve Langasek <vorlon@netexpress.net> wrote:
>On Tue, 28 Nov 2000, Colin Watson wrote:
>> Actually, you do need bin86 for i386 kernels; one of the standard
>> questions I see on comp.os.linux.setup is "help, it says it can't find
>> as86 and I don't know why". kernel-source-* packages do suggest bin86 -
>> I imagine that they don't have hard dependencies on them because they're
>> Architecture: all. Making them architecture-specific would take up
>> significantly more space in the archives.
>
>Isn't there a way to specify architecture-specific dependencies on
>architecture-neutral packages? I seem to remember seeing some Depends: lines
>come through here that had different dependencies for hurd than for linux.
You can do things like libc6 | libc0.2 for the Hurd (although that
particular example's a bit silly as most packages depending on libc6 on
Linux and libc0.2 on the Hurd are architecture-specific - and you'd need
libc6.1 for Alpha as well). That is, not all packages specified in a
disjunction of dependencies need to exist.
If you needed a syntax to borrow for the rare case of
architecture-specific dependencies in architecture-agnostic packages,
you'd probably use that of Build-Depends: and Build-Depends-Indep:; so
Depends: binutils, bzip2, fileutils (>= 4.0), bin86 [i386]. Really,
though, the Depends: line of kernel-source-* is morally a Build-Depends:
line - it just isn't packaged that way. Maybe it would be better to
invest effort in putting source dependency checking into the standard
tools.
--
Colin Watson [cjw44@flatline.org.uk]
Reply to: