Re: Requesting advice on build dependencies
On Sun, Aug 12, 2001 at 07:34:08PM -0500, Branden Robinson wrote:
> On Sat, Aug 11, 2001 at 11:33:20PM -0400, Matt Zimmerman wrote:
> > I think what would elegantly solve this problem would be a notion of an
> > optional build dependency, which would be installed by the autobuilders if
> > it is available, but if it is not, the build could continue successfully.
>
> I don't like this because then it's not obvious which build-deps were
> actually in place when the package was built.
>
> For instance, let's say vim had a conditional build-dependency on xlibs-dev;
> it would be hit or miss as to whether you had X support on the resulting
> packages, depending on who built it.
That doesn't seem that much worse that the situation we currently have. I
could easily build a package without installing all of its build dependencies,
given that I knew they weren't really essential to the build. I could even
then distribute the resulting binary package, along with the source package
listing the build dependencies I had ignored (muahaha!).
Build-depends should be strictly followed by the autobuilders, but for joe
user, they are sometimes optional.
One problem I could foresee with optional build-depends is where the depended
becomes available for a given architecture after the depending package has
already been built. Then, a recompile would have to be forced (or worse, a new
dummy version uploaded) in order to get the added functionality.
> I think it's much better to explicitly list architectures that have some
> build-depencies missing (or unique). In my opinion the situation on point is
> a corner case.
I agree that listing architectures is generally better, but I think a weaker
build-depends could have its uses.
--
- mdz
Reply to: