Public SourceMage Discussion List

[neuron <vlmarek AT volny.cz>]

[...]
> Support for a given optional dependency is not a function of whether
> or not the spell is installed or not, it is a function of what the
> user wants. Their decision might be effected by the status of the spell,
> but thats an implementation detail.

Ok, that was the main point of misunderstanding for me. I thought that
'dependency' is here mainly for tracking how one spell depends on other
spell being present. I understand the other look too.

[...]
> If I have arts installed and cast xmms, i might not want xmms linking
> against arts at all. Note that optional_depends has additional parameters
> to specify an enabled and disabled option, which in the default case
> are passed to the ./configure script.

Yes, indeed. My point was that there is no way of disabling of some
features using ./configure script. For example, you can't stop dosbox
using sdl_net, if you have sdl_net installed. As it was discussed it
here before, there are two options to fix it.

* Fix the configure itself (or configure.in)
* Modify depends so that they match the reality, which was my ugly
  if spell_ok sdl_net
  ...
  fi


> Also, depends/optional_depends can only be run during DEPENDS, which
> is only run once during cast. It also makes little sense to check if a
> dependent spell is installed or not because the dependent spell might
> be installed by the time the current spell is being built.

Yes, this is a valid point, I haven't thought about that. My hacking of
depends is not 100% solution then. Now that gives me the reason for
fixing configure.


> The real answer to this problem is that spells need to explicitly list
> all of their dependencies, and properly disable linking to disabled
> optional_dependencies. The first problem is something that could be solved
> using the prometheus testing tool and a script written by dufflebunk to
> list missing dependencies. The hacky work-around thats being used is to
> check what spells are linking against the spell's libraries and
> trigger those whether or not a dependency exists. Each of those missing
> dependencies are actually bugs.

And I had the strange feeling, that this idea is so obvious that someone
must have implemented that allready :)

Thank you for the explanation.

--
        neuron

Attachment: pgpxI4bobH1fE.pgp
Description: PGP signature