On 06/01/2010 09:06 PM, Brian Gerkey wrote:
> On Tue, Jun 1, 2010 at 11:24 AM, Bill Morris<morris@ee.ccny.cuny.edu>  wrote:
>    
>> Is there a roadmap for when we can expect better cross-compile support?
>>      
> No, not explicitly.  Cross-compilation is something I'm happy to
> support, but not something that we actually do at Willow.   As I
> understand it, several people have gotten much of ROS cross-compiled
> successfully, with most problems coming from third-party packages
> (e.g., Bullet) and system dependencies (e.g., Boost).
>
>
>    

Hi,

Actually, most problems come from the fact that ROS does not do the 
distinction between the tools that needs to be natively compiled (e.g. 
librospack, librosstack, and the message generation things), and the 
things it needs to compile for the target. For instance, when compiling 
a kernel, a few things are compiled natively ([HOST CC]) and then used 
in the compilation process itself.

Ideally, ROS should make a separate bin/lib directory for the tools it 
needs for compilation, and then use whatever CC to recompile the tools 
for the target system.

Currently, when cross-compiling, one needs to run a first pass for the 
host , save the binaries, then remake for the target, save the new 
binaries, replace them by the host, mark the tools ROS_NOBUILD, and make 
the remaining of the tree. I would not call that a clean 
cross-compilation :)

HTH

-- 
Cedric
- http://www.asl.ethz.ch/people/cedricp
- http://www.skybotix.com