Mailing List Archive: one cross-toolchain, many filesystem trees.

one cross-toolchain, many filesystem trees.

Mar 23, 2011, 5:50 AM

Post #1 of 3 (1277 views)

Hi. I'm planning to build 3 ARM filesystem trees. Since they differs
only in softfloat and mach flags in their /etc/make.conf, I don't want
to maintain 3 equal cross-build environments. Thats why I'm thinking
about building only one and than making 2 additional emerge wrappers,
which will set right SYSROOT and call original emerge wrapper. Is it
generally sufficient to make things work?
Thanks in advance.
Sergey.

Re: one cross-toolchain, many filesystem trees. [ In reply to ]

vapier at gentoo

Mar 23, 2011, 7:50 AM

Post #2 of 3 (1243 views)

Permalink

On Wed, Mar 23, 2011 at 8:50 AM, Sergey Mironov wrote:
> Hi. I'm planning to build 3 ARM filesystem trees. Since they differs
> only in softfloat and mach flags in their /etc/make.conf, I don't want
> to maintain 3 equal cross-build environments. Thats why I'm thinking
> about building only one and than making 2 additional emerge wrappers,
> which will set right SYSROOT and call original emerge wrapper. Is it
> generally sufficient to make things work?

"only" !? the two pieces you speak of are core to the code generated.
you'll need sep trees for each FPU target, and for each CPU class.
you could pick the lowest common denominator across the targets march
wise and build for that, but that's no fun.
-mike

Re: one cross-toolchain, many filesystem trees. [ In reply to ]

ierton at gmail

Mar 27, 2011, 1:47 AM

Post #3 of 3 (1237 views)

Permalink

2011/3/23 Mike Frysinger <vapier@gentoo.org>:
> On Wed, Mar 23, 2011 at 8:50 AM, Sergey Mironov wrote:
>> Hi. I'm planning to build 3 ARM filesystem trees. Since they differs
>> only in softfloat and mach flags in their /etc/make.conf, I don't want
>> to maintain 3 equal cross-build environments. Thats why I'm thinking
>> about building only one and than making 2 additional emerge wrappers,
>> which will set right SYSROOT and call original emerge wrapper. Is it
>> generally sufficient to make things work?
>
> "only" !? the two pieces you speak of are core to the code generated.
> you'll need sep trees for each FPU target, and for each CPU class.
> you could pick the lowest common denominator across the targets march
> wise and build for that, but that's no fun.
> -mike
>

I'm sorry, looks like I mentioned wrong 2 pieces. By softfloat I mean gcc flag
-msoftfp and there should be -mcpu, not -mach. Basically, my 3 systems are based
on
1) arm1136
2) arm1176jzf with CFLAGS+=-softfp
3) arm1176jzf without using FPU
I think they could be built with same cross-toolchain.

Sergey