Apologies for the delay in responding. On Tue, Oct 31, 2006 at 04:41:09PM +0100, Marcus Better wrote: > Users don't usually build the package themselves, do they? > And if they do, they are supposed to install the required > build environment. In this case the required dependency > (fastjar) is really small too.
It's only the required dependency if the maintainer
artificially makes it so.
> Virtual packages in build-deps are evil. Since the
> different jar implementations are not guaranteed to be
> equivalent, now or in the future, the outcome of the build
> will depend on the build environment. This means that the
> package is likely to FTBFS or produce interesting random
> bugs at some point.
This is true. However, I want to Build-Depends: on
<specific> | <virtual>, so that if the dependency is not
satisfied in the current environment, you will get the "well
tested" approach. This will also ensuire deterministic
buildd behaviour.
If a particular implementation of jar does cause an FTBFS,
I'd rather this was found and fixed than avoided with an
incorrect dependency.
> That is exactly why the Java policy explicitly states that
> a specific JDK must be used during build.
I can't find that in the policy. Can you point me at the
right place?
> By the way, what package are you working on?
WadC, an AWT-environment for a domain-specific functional
language:
<http://strlen.com/wadc/>
<http://bugs.debian.org/395364>
Yours,
--
Jon Dowland
signature.asc
Description: Digital signature
