Re: i686-linux GCC package on x86_64

[Danny Milosavljevic <dannym@scratchpost.org>]

[-- Attachment #1: Type: text/plain, Size: 2288 bytes --]

Hi,

On Fri, 04 Oct 2019 17:46:43 +0200
Jelle Licht <jlicht@fsfe.org> wrote:

> Mathieu Othacehe <m.othacehe@gmail.com> writes:
> 
> >
> > --8<---------------cut here---------------start------------->8---
> >     (native-inputs
> >      `(,@(if (not (string-prefix? "i686" (%current-system)))
> >            `(("cross-gcc" ,(cross-gcc "i686-unknown-linux-gnu"))
> >              ("cross-binutils" ,(cross-binutils "i686-unknown-linux-gnu")))
> >            '())))
> > --8<---------------cut here---------------end--------------->8---
> >
> > that uses the current gcc if you're already building on an i686 system,
> > or define and use a cross-gcc targeting i686 systems otherwise.  
> 
> This snippet might make a lot of sense to seasoned schemers/guixfolk,

Basically just ignore the birdshit characters to understand what it does :)

The first unquote is to evaluate (%current-system) at the toplevel which is
what interprets the package definitions in the first place.

"`(,@" is a no-op.  Not sure why it's written like that.

>            `(("cross-gcc" ,(cross-gcc "i686-unknown-linux-gnu"))
>              ("cross-binutils" ,(cross-binutils "i686-unknown-linux-gnu")))

is like we always write inputs, but it's calling the "cross-gcc" function in the
toplevel in order to get the package to use.

> with the multiple levels of (un)quoting and what not. It does not seem
> like what somebody with little experience in either would think of by
> themselves.
> 
> Would it make sense to have a section/stub in the cookbook about cross
> compilation?

I don't think that cross-gcc is stable API that much--and it's not
common that you need it anyway.  A normal user just cross compiles
by using

   guix build --target=i686-linux pkg

or better yet, uses qemu to natively compile it for a foreign system

   guix build --system=i686-linux pkg

.

The above is only necessary when for some reason your package has
parts which only compile on one system and other parts which only
compile on another system--that's very rare.

Nowadays, packages are supposed to be cross-platform, so using cross-gcc
directly is unnecessary, too.

Right now cross-gcc is not documented, so I guess that counts as
"not a public interface".


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]