[Rcppoctave-user] Apple Blues

[Renaud Gaujoux]

Nice. Many thanks!

... and along the way we, or at least I, learnt that the hooks
.onUnload and .onDetach are actually not called when R quits and that
required cleanup procedures must be registered via reg.finalizer to be
sure they are called.

What is left:
  * check if a Mac binary package works fine on machine different than
the one it was built on.
  * get CRAN building machines geared with Octave so that binary
packages are available from CRAN. I will follow up with the CRAN
maintainer on this.

Bests,
Renaud



On 8 November 2013 16:23, Dominick Samperi <djsamperi at gmail.com> wrote:
> The biocLite windows binary install works (provided the user has Octave in
> PATH).
>
> The github (ref='develop') install works under Linux and Mac OS X
> (Mavericks).
>
> And by increasing the "biodiversity" (number of OS's supported) a hidden bug
> was
> uncovered and fixed (after 100 posts!).
>
> Cheers,
> Dominick
>
>
> On Fri, Nov 8, 2013 at 4:42 AM, Renaud Gaujoux
> <renaud at mancala.cbio.uct.ac.za> wrote:
>>
>> True :D
>> I thought of trying to get a neat sequence of flags without quote but
>> clearly did not work.
>> I updated the gh and myCRAN versions: they now contain your solution
>> for -Wl and also use mkoctfile rather than octave-config to retrieve
>> the include and linker flags.
>>
>> If possible, can anybody please try the windows binary from myCRAN:
>>
>> source('http://www.bioconductor.org/biocLite.R')
>> biocLite('RcppOctave', siteRepos =
>> 'http://web.cbio.uct.ac.za/~renaud/CRAN')
>>
>> Thank you.
>>
>> Bests,
>> Renaud
>>
>> On 7 November 2013 17:13, Dominick Samperi <djsamperi at gmail.com> wrote:
>> > This solution yields:
>> >
>> > -Wl,-F/Library/Frameworks/R.framework/..,-framework R ...
>> >
>> > and the linker doesn't know that the last "R" belongs to the
>> > -framework option.
>> >
>> > To fix, simply add the 'g' qualifier to the sed command in
>> > configure.ac:
>> >
>> > R_LDFLAGS=`echo "-Wl,$R_LDFLAGS" | sed 's/ /,/g'`
>> >
>> >
>> > My original solution works because the -Wl,"whatever"
>> > option causes mkoctfile to insert the string "whatever"
>> > into the linker command line without parsing. If instead
>> > mkoctfile used a form of exec(3) that expects tokenized
>> > options, the sed solution would be necessary.
>> >
>> >
>> >
>> > On Thu, Nov 7, 2013 at 7:20 AM, Renaud Gaujoux
>> > <renaud at mancala.cbio.uct.ac.za> wrote:
>> >>
>> >> I changed configure.ac into:
>> >> R_LDFLAGS=`echo "-Wl,$R_LDFLAGS" | sed 's/ /,/'`
>> >>
>> >> as I read/saw that multiple options must be comma separated.
>> >> This will break if the path to the R framework contains spaces though.
>> >> Can you please try again:
>> >>
>> >> install_github('RcppOctave', 'renozao', ref = 'develop')
>> >>
>> >> Thanks.
>> >>
>> >> Bests,
>> >> Renaud
>> >>
>> >> On 6 November 2013 22:37, Renaud Gaujoux
>> >> <renaud at mancala.cbio.uct.ac.za>
>> >> wrote:
>> >> > Nice. It is indeed a better solution than changing the flags and
>> >> > their
>> >> > values. The sed command did not work probably because I forgot to
>> >> > double
>> >> > the
>> >> > square brackets, which is necessary to correctly pass configure's
>> >> > first
>> >> > string evaluation.
>> >> > I will incorporate the change tomorrow.
>> >> > Thanks!
>> >> >
>> >> > Bests,
>> >> > Renaud
>> >> >
>> >> > On Wednesday, November 6, 2013, Dominick Samperi wrote:
>> >> >>
>> >> >> The reason we have been hacking away at '-F' vs '-L' instead of
>> >> >> letting 'R CMD config' figure things out is that mkoctfile does not
>> >> >> know about R frameworks ('-F' is invalid). But mkoctfile does have
>> >> >> the -Wl,whatever option, which simply passes whatever down to
>> >> >> the linker.
>> >> >>
>> >> >> I have attached a modified configure.ac that uses this option
>> >> >> instead of using sed to hack the output of R CMD config --ldflags.
>> >> >>
>> >> >>
>> >> >>
>> >> >> On Wed, Nov 6, 2013 at 10:25 AM, Dominick Samperi
>> >> >> <djsamperi at gmail.com>
>> >> >> wrote:
>> >> >>
>> >> >> Hello Renaud,
>> >> >>
>> >> >> So, this R technical issue (reg.finalizer) was the crux of the
>> >> >> problem
>> >> >> all
>> >> >> along. Good work!
>> >> >>
>> >> >> The '..' is still there and the build still fails without my hacked
>> >> >> Makefile.in, but
>> >> >> with this change everything works (no more install memory errors).
>> >> >>
>> >> >> The '..' in the output of R CMD config --ldflags is there for a
>> >> >> purpose,
>> >> >> because
>> >> >> the R framework is under /Library/Frameworks, and not under
>> >> >> /Library/Frameworks/R.framework. This is analogous to
>> >> >> -L/usr/lib/mylibs -lfoo, where /usr/lib/mylibs may contain a
>> >> >> collection
>> >> >> of libraries (instead of frameworks), not just libfoo.so.
>> >> >> (A Mac OS framework is analogous to an R package.)
>> >> >>
>> >> >> Make sure '..' is not in the path supplied to '-L' and everything
>> >> >> should
>> >> >> work.
>> >> >>
>> >> >> Cheers,
>> >> >> Dominick
>> >> >>
>> >> >>
>> >> >> On Wed, Nov 6, 2013 at 8:19 AM, Renaud Gaujoux
>> >> >> <renaud at mancala.cbio.uct.ac.za> wrote:
>> >> >>
>> >> >> Hi,
>> >> >>
>> >> >> I made some changes available in the develop version, which should
>> >> >> hopefully solve both issues:
>> >> >>   * octave_end is now called on exit (via reg.finalizer: see my post
>> >> >> on r-devel
>> >> >> https://stat.ethz.ch/pipermail/r-devel/2013-November/067912.html)
>> >> >>   * the extra dots are removed by the sed in configure.ac, but I
>> >> >> think
>> >> >> we still need to figure out why these dots are there...
>> >> >>
>> >> >> To test:
>> >> >>
>> >> >> devtools::install_github('RcppOctave', 'renozao', ref = 'develop')
>> >> >>
>> >> >> Thanks.
>> >> >>
>> >> >> Bests,
>> >> >> Renaud
>> >> >>
>> >> >> On 6 November 2013 11:15, Renaud Gaujoux
>> >> >> <renaud at mancala.cbio.uct.ac.za>
>> >> >> wrote:
>> >> >> > Couple of questions to understand what is going on:
>> >> >> >
>> >> >> >   * why does R CMD config --ldflags returns these extra '..' if
>> >> >> > they
>> >> >> > are
>> >> >> > incorrect?
>> >> >> >   * why does the compilation work if the path is incorrect to R
>> >> >> > library
>> >> >> > is
>> >> >> > incorrect?
>> >> >> >   * why removing the dots should solve the malloc error?
>> >> >> >
>> >> >> >
>> >> >> >
>> >> >> > On 5 November 2013 16:38, Dominick Samperi <djsamperi at gmail.com>
>> >> >> > wrote:
>> >> >> >>
>> >> >> >> R CMD config --ldflags
>> >> >> >> yields
>> >> >> >> -F/Library/Frameworks/R.framework/.. -framework R
>> >> >> >>
>> >> >> >> This needs to be hacked for mkoctfile to:
>> >> >> >> -L/Library/Frameworks/R.framework/Libraries -lR
>> >> >> >> (no '..')
>> >> >> >>
>> >> >> >> Applying this to the github source fixes it for
>> >> >> >> Mac OS X Mavericks provided the --no-test-load
>> >> >> >> and octave_end tricks are used.
>> >> >> >>
>> >> >> >>
>> >> >> >> On Tue, Nov 5, 2013 at 9:04 AM, Renaud Gaujoux
>> >> >> >> <renaud at mancala.cbio.uct.ac.za> wrote:
>> >> >> >>>
>> >> >> >>> Not sure where the extra '..' would come from.
>> >> >> >>> Are you installing the latest version from github?
>> >> >> >>>
>> >> >> >>> devtools::install_github('RcppOctave', 'renozao')
>> >> >> >>>
>> >> >> >>>
>> >> >> >>>
>> >> >> >>> On 2 November 2013 07:00, Dominick Samperi <djsamperi at gmail.com>
>> >> >> >>> wrote:
>> >> >> >>>>
>> >> >> >>>> It appears that octave_end is not called at all under Mac OS X.
>> >> >> >>>> But
>> >> >> >>>> it
>> >> >> >>>> is also not called under Linux! I inserted Rprintf statements
>> >> >> >>>> to
>> >> >> >>>> check.
>> >> >> >>>>
>> >> >> >>>> Perhaps the explanation is that failing to call octave_end
>> >> >> >>>> under
>> >> >> >>>> Linux
>> >> >> >>>> (or Windows?) is non-fatal, but this is fatal under Mac OS X?
>> >> >> >>>> If
>> >> >> >>>> this
>> >> >> >>>> is so the Mac OS X behavior is less dangerous...
>> >> >> >>>>
>> >> >> >>>> If I do not use --no-test-load, I get the result in
>> >> >> >>>> RcppOctave.log
>> >> >> >>>> (attached)
>> >
>> >
>
>