> Most systems have no "possibly different architecture" to share with,
> many will be short of disc; forcing all machines to build for all
> architectures is bad, for those few machines that are large & networked &
> want to export all man architectures, better to add an env var
> BUILD_ALL_MAN_ARCHS or SUPPRESS_BUILD_ALL_MAN_ARCHS.

It is hardly ever appropriate to add such an option.  This is not a
case where it is appropriate.

Given that, the choice is between: always installing them for all
architectures, and never installing them for all architectures
(i.e. letting the admin of a multi-architecture server deal with it).

I've not decided what i think of these options, and am not likely to
think about it again soon, so i probably won't change it in the near
future.

Normally, i'd agree that it's best to just install all of the man
pages (and other share-things) for all architectures.  However, this
breaks down, in that various per-machine binaries have man pages (and
perhaps other things, e.g. data files) that are only installed on
architectures where the binaries are supported.

In other words, to make either solution work correctly will take some
work, and, though i'd like to have all of share installed by default
every time, it may not be practical to implement correctly.


Also, i think the time and space required to build and install
other-architecture man pages is minimal, compared to the time and
space required to build/install a full source tree.  I don't consider
it significant for people who are keeping "current."  For releases,
it obviously makes a difference -- which is why they're seperated out.



cgd