Joshua Nichols wrote:
> I've been doing a little work on axxo's overlay recently, particularly
> with choosing a sane compiler, ie it supports the desired -source and
> -target. It seems there's a number of packages that jikes doesn't like
> (surprise). I was talking with axxo about how to handle this, and
> wanted some input from you.
> The current way of handling jikes is to not use eant/ejavac when the
> package doesn't work with all the compilers. I'm not particularly fond
> of this, because it doesn't give any indication that the package
> doesn't work with a particular compiler.
> Instead, I think an ebuild should know that it doesn't work. As its
> implemented, you can specify JAVA_PKG_FILTER_COMPILER to specify a
> list of compilers not to use. I think that this should be used in
> ebuilds that don't work with jikes (or alternatively, have a function
> which handles this modifying this variable).
> Axxo had some reservations about this, because he thought it wouldn't
> be much better than the old way of doing it with use flags. I don't
> think this is quite the case though, you don't get 'use flag
> pollution' with multiple compilers, and that most packages work with
> all the compilers.
> Please let me know what you think.
Well, we need to come with *some* solution. As I see it, there are
essentially two choices:
1a) For each package, we explictly list the compilers we know this
package works with. This can be done in src_compile, via an
eclass function; java-pkg_declare-compilers "javac ecj jikes".
1b) Same as 1a, but by default, we resort to a JAVA_SUPPORTED_COMPILERS
variable defined in the java-pkg.eclass as the default setting.
2a) We do your filtering stuff.
2b) We define a fancy-schmancy filter function in the spirit of
flag-o-matic, java-pkg_filter-compilers "jikes", and invoke this
in src_compile before eant/ejavac/ewhatever
For now, I don't see any other sane alternatives.
My personal favourite is 2b, since there is already a long standing
precedence with flag-o-matic on this.
-- Karl T
email@example.com mailing list