On 5 May 2015 at 16:29, Antti Kantee <[email protected]> wrote:
> On 05/05/15 15:13, Martin Lucina wrote:
>>
>> On Tuesday, 05.05.2015 at 11:24, Antti Kantee wrote:
>>>
>>> I think I suggested not introducing defines unless they are
>>> absolutely necessary, or any other similar identifiers for that
>>> matter.  Which are, in fact, the rumprun<platform> falls under.  I
>>> have a vague memory that Martin justified <platform> in the
>>> toolchain name, but I'm not 100% if my memory serves me right.
>>
>>
>> The <platform> needs to be in the toolchain name so that the *user* can
>> distinguish the different toolchains, and also so that they can be
>> installed side by side in the same directory in $PATH.
>
>
> Is that a useful requirement?  I'm questioning its validity especially since
> ...
>
>>>> eg there is no such thing as an arm baremetal platform,
>>>> there is one for pretty much every SoC thats ported to.
>>>
>>>
>>> Good point.
>>
>>
>> I'm not sure what to do about that :-(
>>
>> An idea: we invent some explicit flag that would be passed to our -gcc,
>> -g++ or -ld wrapper and mean "build for this particular board".
>
>
> ... the proposal doesn't satisfy it.
>
> I'm a fan of the ". xcompile-platform.sh" approach, which sets the right
> paths, CC, PS1, etc.
>
> In that namespace we could choose $platform freely without breaking build
> suites.

Its the standard setup eg if you want your stuff to be installed in
packages in a Linux distro. But I am not entirely convinced that we
want to optimise for that use case right now.

Justin

Reply via email to