[Sugar-devel] Announce: OLPC software strategy.

[C. Scott Ananian]

On Thu, Jul 8, 2010 at 3:49 PM, Chris Ball <cjb at laptop.org> wrote:
>   > What about the compiler? IIUC currently a commercial compiler is
>   > required. If that continues to be the case (as I expect it to),
>   > would it be possible for OLPC to provide the (probably very few)
>   > users interested in hacking on the EC code access to a machine
>   > having this compiler installed? I.e. does the license OLPC has
>   > for this compiler allow more than one user (on the same machine)
>   > to use it (if necessary sequentially, ensured by using a lock
>   > file) and would OLPC be willing to give users access to such a
>   > machine?
>
> Good news here too:  we've moved to the free SDCC compiler, so there
> should be no problem here.  I don't know full details, but there have
> been some incompatibilities seen between SDCC and the EC code in the
> past, so staying with SDCC is going to be conditional on being able
> to find a way around those.  SDCC is the plan, though.
>
> For more questions, I think we should move to the OLPC devel list,
> and I'll let Richard Smith answer because this is his project.  :)

If I understand correctly, the biggest problem is that SDCC lets you
reserve a variable at a specific address, but that *doesn't* prevent
it from using that address for its own automatically-allocated
variables.  This is probably just a linker issue, to detect the
overlapping address assignments -- although if you wanted the compiler
to allocate "with holes", that's a compiler issue.

The SDCC source code is very hackable.  This might be something a
volunteer could address at the compiler level.  The alternative is to
very very very carefully read through the legacy EC code to try to
ferret out (and somehow correct) any such conflicts.
  --scott

-- 
                         ( http://cscott.net/ )