[Rcppoctave-user] Apple Blues

[Dominick Samperi]

Apple is changing the plumbing (moving to LLVM/clang from gcc/g++), and
while these compilers are
largely compatible, there can be binary inconsistencies. In general,
different compiler versions (even
of the same compiler) may create binaries that will not work together, esp.
when you consider
exception processing or foreign function interfaces. There is no standard
for how FFI's should work.


On Fri, Nov 1, 2013 at 10:17 AM, Renaud Gaujoux <
renaud at mancala.cbio.uct.ac.za> wrote:

> You were initially suggesting that if R is compiled using a compiler
> different than Octave then one might end up having troubles.
> Would this still be an issue if one follows the procedure you described?
>
> Generally speaking (and this is for my curiosity), does this mean that
> libraries generated by one compiler cannot necessarily be linked with
> another program using another compiler. I had hoped the interface would be
> platform-dependent only, but not compiler dependent, with compilers needing
> to follow some standard specs of a given platform.
>
>
>
> On 1 November 2013 16:07, Dominick Samperi <djsamperi at gmail.com> wrote:
>
>> Well, I convinced myself (I could be wrong) that the problem is not
>> related to R or Rcpp,
>> and I did not test RcppOctave. As you suggest, termination protocol may
>> need to be
>> tested...
>>
>>
>> On Fri, Nov 1, 2013 at 4:11 AM, Renaud Gaujoux <
>> renaud at mancala.cbio.uct.ac.za> wrote:
>>
>>> What type of R installation did you test this procedure with?
>>> Ideally this would work with the plain CRAN binary, modulo adapting
>>> ~/.R/Makevars.
>>>
>>>
>>> As you probably know, the last line of embedded.cc (attached)
>>>> is very important. Without it there is still a memory error!
>>>>
>>>>
>>>>
>>> I checked Octave source and clean_up_and_exit is a simple wrapper to
>>> do_octave_atexit, which I call directly to skip the call to exit, which
>>> would otherwise terminate the R session as well:
>>>
>>> [from
>>> http://hg.savannah.gnu.org/hgweb/octave/file/eff3d78367f0/src/toplev.cc#l667
>>> ]
>>>
>>> void clean_up_and_exit (int retval)
>>> {
>>>   do_octave_atexit ();
>>>   if (octave_exit)
>>>     (*octave_exit) (retval == EOF ? 0 : retval);
>>> }
>>>
>>> Note that I actually don't know how to effectively terminate the Octave
>>> session. Don't know if this even possible since it currently runs in the
>>> same thread as R (that's at least my understanding). So even after cleaning
>>> up with do_octave_atexit(), we can still send commands to Octave which
>>> responds fine, except if errors occurs and then a segfault sometimes come
>>> up suggesting some thing is in a bad state. Anyway, this is not really a
>>> big issue.
>>>
>>> library(RcppOctave)
>>> o_runif(1)
>>> octave_end()
>>> o_runif(1)
>>>
>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/rcppoctave-user/attachments/20131101/3f35d49c/attachment.html>