1 |
On 02/11/15 23:38, William Hubbs wrote: |
2 |
> On Mon, Nov 02, 2015 at 09:12:30PM +0100, Justin Lecher (jlec) wrote: |
3 |
>> -----BEGIN PGP SIGNED MESSAGE----- |
4 |
>> Hash: SHA512 |
5 |
>> |
6 |
>> How about a virtual here? |
7 |
> |
8 |
> I don't see that working out to well because the compilers are |
9 |
> completely different from each other. As I said, the reference |
10 |
> implementation of the go language is dev-lang/go and the other one, that |
11 |
> is part of gcc, as I understand it, is not quite as fully developed as |
12 |
> dev-lang/go. |
13 |
> |
14 |
> Thanks, |
15 |
> |
16 |
> William |
17 |
> |
18 |
|
19 |
Hi William, |
20 |
|
21 |
but instead of adding |
22 |
|
23 |
DEPEND="||·( |
24 |
>=dev-lang/go-1.4 |
25 |
>=sys-devel/gcc-5.1.0:=[go] |
26 |
|
27 |
to ebuilds, you could create a virtual/go with that dependency and make all |
28 |
packages depend on that instead of this || (). Maybe in future clang gets also |
29 |
support for go, then you would need to edit again all ebuilds. Having a virtual |
30 |
for that case is much simpler. This is how we are doing it for fortran. |
31 |
|
32 |
Another problem which you have with these split deps is the following. Imagine |
33 |
the user has gcc-5[go] installed, which fullfills the dependency requirement. |
34 |
But she selected gcc-4.9 as the active compiler, which doesn't have go support. |
35 |
The build will fail. For fortran we test in pkg_setup if we have a working |
36 |
fortran compiler present. You would need to do the same for go compiler. |
37 |
|
38 |
Justin |