Arch Linux User Repository

@M4G0: That's caused by archlinux rolling out python 3.7. I'll try to just move the python 3.6 files of modeller into the appropriate folders for python 3.7, but I can't promise anything since only python 3.6 is officially supported by upstream. Since modeller is distributed as binary-only we might have to wait for upstream to release a version compatible with python 3.7

UPDATE: Files are in place now using the latest release of this package. However it may not function as intended with python 3.7!

Thanks for your fix, it works fine. I have been toying around with the python33 configure flags to see if a bug with unicode (ucs2 vs ucs4) explained the error, but it is more complex than that.

Ok, seems as the current build of modeller is incompatible with python 3.3/3.4. At least python 3.3 used to work to some point in time, but I haven't figured out yet, what changes in either one of the minor versions of python or the way the package in archlinux is built causes this error. As a workaround, use python32 from the AUR (there is an updated PKGBUILD in the respective comment section) and reinstall modeller (modeller-9.13-3). From a terminal run: export MODINSTALL9v13=/usr/lib/modeller python3.2 and 'import modeller' should work.

Using modeller 9.13 and python3.3 (from AUR) on x86_64, I get this when trying import modeller in a script: ImportError: /usr/lib/python3.3/site-packages/_modeller.so: undefined symbol: PyUnicodeUCS4_Concat Any idea why? What is your set-up? Thanks a lot

This is a *very* late heads-up to your previous comment: Simply using "import modeller" works. I've used that version throughout the entire last year, so I'm pretty sure of it. :)

Both should work now. In theory at least, I just tested if importing the modeller package into python scripts works without errors. I did no further testing though.

Thanks a lot! Hm… Either I'm missing something, or your comment and the instructions in modeller.install contradict each other: Do I still have to run MODELLER scripts with modpy.sh, or can I simply do "import modeller" and am good to go?

Updated to version 9.11. Packages for Python 2.7 and 3.2 are now automatically installed if available. You can use "import _modeller" from within your python scripts.

Hey, Would you consider including the patch that fixes memory leaks in loop modeling that is posted on the MODELLER wiki [1]? Thanks in advance! [1] http://salilab.org/modeller/wiki/Patches

Thanks for updating! I tried to use the symlinks at the end of the PKGBUILD to make Python aware of the module, but I get the following error message: ---> 77 import _modeller 78 if hasattr(config, 'license'): 79 _modeller.mod_license_key_set(config.license) ImportError: libmodeller.so.5: cannot open shared object file: No such file or directory This means that the module "modeller" gets imported, but its twin _modeller (which calls the shared libraries) cannot be imported because of wrong paths. Now, libmodeller.so.5 is in /usr/lib/modeller/lib/x86_64-intel8 (or equivalent for 32 bits), shall we symlink it somewhere else?

The new release 9.10-2 solved previous naming scheme problems. I suggest two further improvements: 1. change the function build() to package() in the PKGBUILD, since nothing is actually being built. 2. add a mock config.py file, e.g. via #Create config.py is it not exists if [ ! -f "${MODINSTALL}/modlib/modeller/config.py" ]; then echo "license = 'XXXX'" > "${MODINSTALL}/modlib/modeller/config.py" fi at the end of the PKGBUILD. This makes it slightly easier for the user to fill out the license thing, since a skeleton is already in place. Finally, I have a question. The modeller developer do not know about the user's LD_LIBRARY_PATH, so they created the modpy.sh scripts which sets this variable for the python session. However, the packager (we) should know about these paths in ArchLinux, right? Why don't we copy the relevant libraries into a system folder so that people are not required to use modpy.sh anymore?