[Phylobase-devl] Compilation woes
pdc at berkeley.edu
Mon Aug 4 08:07:33 CEST 2008
I can't say that I have any special insight into the current
compilation issues. Last time we went through this Stephan eventually
emailed me the R code that runs the package building. So I do know
that a fresh copy of the SVN directory is made into a temporary
directory. So I would be very surprised if the issue was related to
build products remaining.
There seem to be a lot of configuration files under version control in
the top NCL folder. A quick look at the SourceForge version shows
that they have fewer files. I do think that at least some of these
configure files are auto created. Perhaps we should pare down to only
the necessary files.
As for the immediate future of NCL, I think it's going to be a killer
feature of the package. However, so far it has consumed a lot of
developer time just trying to get it to build properly. At least on
my part that is due to a lack of expertise with makefiles and C++ in
generally, perhaps this could be solved by contacting the NCL
developers? I have also had frustrating experiences using the NCL
commands. I sent an email regarding this experience previously to the
list. I suspect that the performance issues will easily overcome. My
vote would be, if NCL is preventing successful building and thus
adoption of the package we should consider separating it, perhaps into
a branch. I think there is a way for us to build multiple version of
the package see the "distr - S4 classes for distributions" package.
My 2 cents
On Sun, Aug 3, 2008 at 9:49 PM, Steve Kembel <skembel at berkeley.edu> wrote:
> Hi Brian,
>> Thanks for looking into this, Steve and Ben. If it is just a matter
>> of old files not being rewritten properly, this should be able to be
>> taken care of in the Makefile (effectively, make clean first). I'll
>> check it out some more.
> I'm not 100% sure about this. The install errors I get in Linux are
> not very reproducible, but that's my guess. For example I got a very
> similar error to the R-Forge linux build errors the first time I tried
> to build, but then slightly different errors on subsequent attempts,
> and no errors with a fresh copy of the code.
> A few more things about this:
> -make clean doesn't actually seem to work properly, there isn't a
> clean action defined in the ncl makefile, maybe this is part of the
> -there is a script pkg/cleanup that gets rid of a lot of the
> accumulated files. I don't know if this could be incorporated into the
> make clean action. I missed this until now but I think this is
> something that Thibaut wrote to get phylobase to compile on linux_x64
> a long while back. The message in which he describes it was pre-R-
> Forge so I'm including a copy of the text below. Sounds like part of
> the problem is indeed files not being deleted and messing up
> subsequent builds.
> -if i build the package from a clean copy of the code, the following
> files are left behind.
> steve at pundit:~/code/phylobase$ svn status
> ? pkg.Rcheck
> ? phylobase.Rcheck
> ? phylobase_0.3-1.tar.gz
> ? pkg/config.log
> ? pkg/config.status
> ? pkg/src/phylobase.so
> ? pkg/src/Makevars
> ? pkg/src/RcppSrc/libRcpp.a
> ? pkg/src/ncl/config.log
> ? pkg/src/ncl/config.status
> ? pkg/src/ncl/stamp-h1
> ? pkg/src/ncl/libtool
> ? pkg/src/ncl/configure.lineno
> ? pkg/src/ncl/Makefile
> ? pkg/src/ncl/ncl/.deps
> ? pkg/src/ncl/ncl/Makefile
> ? pkg/src/ncl/ncl/.libs
> -if I then run the cleanup script, the following files are still left:
> steve at pundit:~/code/phylobase/pkg$ svn status
> ? src/ncl/config.log
> ? src/ncl/config.status
> ? src/ncl/stamp-h1
> ? src/ncl/libtool
> ? src/ncl/configure.lineno
> ? src/ncl/Makefile
> ? src/ncl/ncl/.deps
> ? src/ncl/ncl/Makefile
> Hope this helps. I definitely agree about the value of having NCL in
> phylobase, let's see if we can sort out these issues.
> Begin forwarded message:
>> From: Thibaut Jombart
>> Date: February 13, 2008 9:16:53 AM PST
>> To: phylobase-devel at nescent.org
>> Subject: Re: [phylobase-devel] 64bits compilation troubles: IT WORKS
>> Ben Bolker wrote:
>>> OK, but ...
>>> I thought your last e-mail said that even hacking -fPIC into
>>> the Makefile didn't work. I think "we" should first figure out
>>> if we can get it to build with a hacked Makefile, then figure
>>> out how to make the Makefile get constructed properly.
>>> (At worst, that might allow us to make a 64-bit binary
>>> to post on the web page, although fixing this properly
>>> would certainly be better.)
>>> I was guessing that ./configure --with-pics would
>>> work, but you tried that ...
>>> I think we might also be able to shove it into the
>>> automake flags somewhere ...
>> Hi people,
>> So I now have phylobase installed on my 64bits laptop. Yet, it does
>> not answer how we will generally fix the problem.
>> In fact, hacking the src/ncl/configure file indeed works, but didn't
>> for the following reason. It seems that .o objects are compiled only
>> when they do not already exist. If then do, then all what is done is
>> linking, and so the hack was not effective. So, this works for me:
>> 1) first run the cleanup script to override stuff like .o files.
>> 2) modify the configure script like follows:
>> (starting from line 20560)
>> CXXFLAGS_OPTIM_SPEED="-O3 -ffast-math"
>> CXXFLAGS_OPTIM_SPEED="-O3 -ffast-math"
>> CXXFLAGS_OPTIM_SIZE="-Os -ffast-math"
>> # modif by Thibaut
>> case "$build_os" in
>> darwin*) CXXFLAGS_WARNINGS="$CXXFLAGS_WARNINGS -Wno-long-
>> if test "x$CXXFLAGS_OPTIM_SIZE" = "x"; then
>> CXXFLAGS="$CXXFLAGS $CXXFLAGS_OPTIM_SPEED $CXXFLAGS_WARNINGS
>> 3) run R CMD INSTALL (only warnings)
> Phylobase-devl mailing list
> Phylobase-devl at lists.r-forge.r-project.org
More information about the Phylobase-devl