6.1.4 again

[djsutherland]

Trying to further #18 here, since that doesn't have "allow edits by maintainers" checked.

@vigsterkr: the reason it was picking up the old ones is because the new shogun-cpp builds have had their arpack and hdf5 migrated, while #18 hasn't. I applied those migrations here, and now it's getting the new builds; let's see if that worked.

Closes #18, closes #13, closes #15, closes #17, closes #20, closes #19.

[conda-forge-linter]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

[djsutherland]

Failing on Linux with

gmake[3]: *** No rule to make target '/home/conda/feedstock_root/build_artifacts/shogun-cpp_1587164780022/_build_env/x86_64-conda_cos6-linux-gnu/sysroot/usr/lib/librt.so', needed by 'src/interfaces/python/_shogun.so'. Stop.


Something wonky is going on with the build system here...any ideas @vigsterkr?

[djsutherland]

Also, hmm – as is, I think it's using the generic blas version of shogun-cpp, which means it's not generating any Python bindings for the GPL components. This should probably not alter the ABI of shogun itself, so that you can install the Python bindings with shogun and run stuff with either version of shogun-cpp, but this is maybe worth verifying....

[vigsterkr]

@djsutherland yep.... still the same problem... see #18 (comment) and messages after

[djsutherland]

Hmm, I missed all of that. I'll try a new shogun-cpp build with the newest conda-build, just in case that helps, since there's an update I wanted to make anyway (conda-forge/shogun-cpp-feedstock#38)....

[djsutherland]

Okay, new builds didn't help ☹️ . Looks like Python 3.8 is also failing even on Mac with what looks like something unrelated:

shogunPYTHON_wrap.cxx:651976:5: error: cannot initialize a member subobject
of type 'Py_ssize_t' (aka 'long')
with an lvalue of type 'int (PyObject *, FILE *, int)' (aka 'int (_object *, __sFILE *, int)')
    print_sgobject,                           /* tp_print */
    ^~~~~~~~~~~~~~

[djsutherland]

Sorry to bug, @isuruf / @mingwandroid, but I missed the whole thing with path rewriting in 3.18.12 and am not sure how to proceed from here with

gmake[3]: *** No rule to make target '/home/conda/feedstock_root/build_artifacts/shogun-cpp_1587164780022/_build_env/x86_64-conda_cos6-linux-gnu/sysroot/usr/lib/librt.so', needed by 'src/interfaces/python/_shogun.so'. Stop.

even though the build of shogun-cpp in use was made today (conda-build 3.19.2). Any hints?

(Doesn't seem like it should be relevant, but I checked and hdf5, libgcc-ng, libgomp are all shogun-cpp deps that link to librt and were built with conda-build 3.18.11; the other deps which link to librt were all built with 3.19.2.)

[isuruf]

I haven't checked if conda-build has a bug here or not. A workaround would be to do merge_build_host: True # [linux] in shogun-cpp, but that's an ugly hack.

[vigsterkr]

@djsutherland the python3.8 looks like a typical swig problem.