[oe] "as" has gone missing.

[Richard Purdie]

On Wed, 2008-01-02 at 10:12 -0600, Mike (mwester) wrote:
> "pHilipp Zabel" <philipp.zabel at gmail.com> writes:
> ...
> > > I am need of a clue, if anyone can offer some direction.  Might this be
> as
> > > simple as a configuration option missing when gcc is built (and "as"
> should
> > > be part of the "gcc" package), or is this a case where I'll have to add
> > > support for a separate "as" package to the recipe?
> > >
> > > Thanks,
> > > Mike (~mwester)
> >
> > binutils, binutils-symlinks
> 
> Excellent, that (mostly) works.  So what's missing is the dependency in the
> gcc package.  Does anyone have any particular heartburn if I add a RDEPENDS
> = binutils to the gcc recipe?

Its not strictly an RDEPENDS. You would for example use odcctools+gcc if
you were on arm-darwin (this isn't a hypothetical example). This comes
back to the virtual and RPROVIDER mess I've commented on in the past. 

The thread was "[RFC] Handling of virtual/* namespace in packages" and I
proposed a VIRTUAL-RUNTIME namespace. In this case it would be a:

VIRTUAL_RUNTIME_binutils = "binutils"

and hence:

RDEPENDS = "${VIRTUAL_RUNTIME_binutils}

> (As for "mostly", it works, but there's a mismatch that might be important:
>     $ gcc hello.c
>     /usr/lib/gcc/armeb-linux/4.1.2/../../../../armeb-linux/bin/ld: Warning:
> /tmp/cceIGUyh.o does not support interworking,
>     whereas a.out does

Probably your libc was compiled with thumb interworking and your hello.c
was not. Using the correct gcc flags should remove the warning.

Cheers,

Richard