Unable to build R-modular interface #2654
Comments
|
Some thoughts:
@vigsterkr @matthuska should be able to help, too |
|
I just checked out the latest development version and it compiles fine, though I'm using Ubuntu 14.10. I don't know if it will make any difference but the HDF5 header in that warning (H5public.h) has been updated on my system so that I don't get the same warning you're getting. Here's my version: $ aptitude show libhdf5-dev |
|
Hi Matthuska, My hdf5 is a local self build because shogun doesn't build with hdf5-openmpi, which I installed for the whole system as I need it for other things. See this thread |
|
Just to rule it out, can you modify the H5public.h file so that it doesn't trigger that warning? The details are here: https://gitorious.org/kitware/itk/commit/f46f541516a33aa12be8e390b59f4b5871d57ab9 --- a/Modules/ThirdParty/HDF5/src/itkhdf5/src/H5public.h define H5_PRINTF_HADDR_FMT "%lu"#elif H5_SIZEOF_HADDR_T ==H5_SIZEOF_LONG_LONG error "nothing appropriate for H5_PRINTF_HADDR_FMT"#endif |
|
Just updated my hdf5 to 1.8.12, the hdf5 warnings disappear but still cannot build R-Modular (shogun development from git)
|
|
Are there any earlier errors? Perhaps related to swig? Maybe you can do a fresh build and post the entire build output into a gist so I can look through it for other possible problems. Swig should be generating modshogun.R and for some reason it doesn't seem to be happening. |
|
Hi, here is the complete build log
|
|
@beew would you mind not putting such long outputs here directly? You can paste them at https://gist.github.com/ @matthuska thanks for helping! |
|
I don't see anything obvious in the build output. Can you give me the parameters you used when building hdf5? I'll build a local copy as well and see if I can reproduce your problem. Also, what version of swig are you using? |
|
Hi, It is hdf5-1.8.12 The configure line was ./configure --prefix=$HOME/opt/hdf5 --with-pthread --enable-unsupported --enable-shared --enable-production=yes --enable-parallel=yes --enable-largefile=yes --with-default-api-version=v18" I copied these options from FENICS' dorsal install script but disabled parallel. Swig version 2.0.11-1ubuntu2 from the Ubuntu repository |
|
I tried with the latest git version today, Karl has added a patch to disable hdf5. I was able to do that and shogun did build successfully without hdf5, but R-modular still failed with the same errors. So they are apparently not hdf5 related. I currently do have shogun R-Static installed, do I need to remove it first? |
|
Hi, tried the latest with HDF5 turned OFF but still get errors
But I found modshogun.R in ~/Downloads/shogun/build/src/interfaces/r_modular and it is 7.6 Mb |
|
Could you try to copy the modshogun.R to the place where cmake is looking for it? |
|
It is not clear from the error where cmake tries to look for it. |
|
I just built r-modular locally and it went fine, my modshogun.R is in |
|
Works for me either locally and on Fedora's builders… can you please give us the output of I suspect issues with permissions, UMASK and/or Type of Filesystem… |
|
Hi, here are the outputs
|
|
Hi, I figured it out! I have set my default R work directory to be ~/R in my .Rprofile. I think the installation was looking for path relative to the default R work directory, so that would be ~/R/Downloads/shogun.. which doesn't exist. After commenting out setwd('~/R') in .Rprofile, R-modular interface builds successfully. Thanks everyone for helping. |
|
Great, @beew ! |
|
Any idea where #2395 might come from? I'm still seeing that on Mavericks. |
Getting back a little late since I have not been using shogun's R interface I don't know how to fix the codes or make a patch but the problem I think is in shogun/src/interfaces/r_modular/CMakeLists More precisely these lines What follow COMMAND echo are basically commands to be executed in R. When R runs its work directory is given by the function getwd(), in my case this will return the value in ~/.RProfile, which is not my $HOME but a subdirectory of it. I can imagine others may want to set their default R work directory like this too instead of dumping R outputs all over their $HOME The CMAKE option WORKING_DIRECTORY apparently cannot override the value in ~/.RProfie, so when the shogun build process executes R, it is looking for modshogun.R in WORK_DIRECTORY, which is it's default work directory (defined by .RProfie) rather than ${CMAKE_CURRENT_BINARY_DIR} Here is my take on it, maybe someone who knows his C++ can just edit these lines so that people don't get tripped over on this little thing in the future (I tried but didn't work) |
|
@beew: Can you confirm the referenced patch fixes this issue? |
|
Just saw the link to your change so I manually edited the file. It works. Thanks for fixing this. |
set working-dir properly when running R (#2654)
|
Issue resolved. |
Hi,
I tried to build the R-modular interface but make failed
Tried both shogun 3.2.0 and developement version. R-static interface built fine though.
R version 3.1.2, OS is Ubuntu 14.04 64 bit.
Thanks for help in advance.
The text was updated successfully, but these errors were encountered: