[Nix-dev] Yet another configure proposal (final ?)

[Michael Raskin]

Marc Weber wrote:
>> So, you do not think that 'just supply and  auto-detect' optional
>> dependencies are critical for fltk, though looks like nixzlib flag is
>> really redundant (auto-detect should work). 
> You are totally right.
> Optional features could be implemented easily the way I've proposed by
> adding the passed optional attributes from args to mandatory .. a hack but
> will work fine
Why don't you want to merge my way? You have "optional" list and just
filter it for being supplied..

>> template-for-unstable.nix.
> Sure.
> Either you'll have to submit it or niksnut has to give me svn access..
And have you asked yet? I think he will give commit rights.

>> I will merge default value support and name suffixes into it. 
> What do you mean by default value support and name suffixes?
> Using default values means you may have
> default yes and default no..
> which also means you need two kinds of flags:
> remove flag / add flag (each time overriding the default)
> This adds complexity.. So my default values would be the flag list in
> all-packags.nix itself.. Or do I missunderstand you?
For example, SANE asks for firmware. Sometimes the easiest way to obtain
it is from driver CD. So you can have default set - list of all easily
downloadable firmware files, or override it with your own. You can also
think about kernel version: there can be a naming scheme and a
possibility to override it. Think of a package that has some hard
performance trade-off which requires integer argument...

>> guess just listToAttrs+attrNames will work better.
> uniqList could be implemented as
> uniqList l = __attrNames ( map ( item : listToAttrs { attr = item; value = "dummy"; } ) l)
> but I haven't thought of wether it should be allowed to use non strings
> as attribute names .. :-)
Take strings and allow converting them into paths.

> We have to specifiy as well wether the attr name "a-b" should be allowed
> when using listToSet.
> Right now:
> builtins.listToSet [ { attr="a-b"; value ="a-b";} ]
> => Attrs([Bind("a-b",Str("a-b",[]),NoPos)])
I think it should be allowed. Look Common Lisp - it has a way to make a
symbol with any name.