lbuild | lbuild : a generic , modular code generator in Python

I'm stuck compiling a bitbake recipe for an Allwinner H2 SoC. It seems to be a problem of floating point unit compatibility. This is the compilation error log (abbreviated paths and added line breaks for a little better readability):

I followed these topics:

1. https://devtalk.nvidia.com/default/topic/1044958/jetson-agx-xavier/scikit-learn-for-python-3-on-jetson-xavier/
2. https://devtalk.nvidia.com/default/topic/1049684/jetson-nano/errors-during-install-sklearn-/
3. https://github.com/scikit-learn/scikit-learn/issues/12707

python version: 3.6.9

Here are all commands I run:

I'm trying to install sklearn on top of a Docker image (FROM astronomerinc/ap-airflow:master-1.10.5-onbuild). Environment coming with the source image:

• Alpine Linux v3.10 (kernel 4.9.93-linuxkit-aufs)
• Python 3.7.3
• numpy==1.17.2
• pandas==0.25.1
• pandas-gbq==0.11.0
• ...

I had scipy==1.3.1 in my requirements.txt and had no issues installing it with pip, however when I added scikit-learn to requirements.txt and rebuilt again, I got this error saying a numpy header is missing:

Usually when the linker doesn't find a library that actually exists in the path it's because some binary incompatibilities (i.e. about 32/64 bit). In my situation it seems different and I didn't find a question that answer the issue.

Rapsberry Pi 3B+, Raspbian Stretch Lite, trying to compile the Microchip SDK for 3DTouchPad.

The compilation fails in this way:

I have seen a number of different people post this problem (example: Tried to guess R's HOME but no R command in the PATH. OsX 10.6 and Installing rpy2 on Mac OSX 10.8.5), but I have yet to find a viable solution.

I have ensured that I have Python-2.7 installed in my terminal, and since I had recently upgraded by RStudio, I thought that my R was also updated. But everytime I tried to run either:

I'm working on a Raspberry-based project that needs SciPy, NumPy and scikit-learn. And we need to package our virtual environment in a .deb for distribution. For that, we use dh_virtualenv, which up until now has worked just fine.

When I just install our requirements on the venv, like so:

I'm getting some warnings from ld about libraries that aren't found, but as far as I can tell it should be finding them. For example:

/usr/bin/ld: warning: libleaf_util.so, needed by build/libleaf_lang.so, not found (try using -rpath or -rpath-link)

This happens when I link the executable build/unit_test that includes build/libleaf_lang.so, as part of the command line(1). The linking is adding rpath=$ORIGIN to the executable and all the libraries, including the library build/libleaf_lang.so. If I do ldd on build/libleaf_lang.so it is able to find the library in question:

libleaf_util.so => /home/src/leaf/misc/build/libleaf_util.so (0x00007fd7c2f90000)

That would seem to indicate the required library is found. So why do I get the warning?

Note this appears to only happen when I link using the path to the shared library, build/libleaf_lang.so. If I link by name -lleaf_lang, as I do for another executable, I do not get the warning.

(1) g++ -o build/unit_test -z origin -Wl,-rpath=\$ORIGIN build/boost_test_main.o build/test/expr_conversion_test.o build/test/statement_test.o build/test/expr_type_test.o build/test/full_type_test.o build/test/gmp_test.o build/test/intr_type_parse_test.o build/test/lambda_test.o build/test/number_test.o build/test/object_holder_test.o build/test/parse_test.o build/test/scope_test.o build/test/source_test.o build/test/type_converter_cost.o build/test/type_converter_fixate.o build/test/type_converter_function_call.o build/test/type_converter_match_function.o build/test/type_converter_parameterize_type.o build/test/type_converter_test.o build/test/type_converter_unify.o build/test/type_identifier_constrain.o build/test/type_identifier_determine.o build/test/type_identifier_expand.o build/test/type_identifier_get_spec.o build/test/type_identifier_infer.o build/test/unicode_test.o build/libleaf_lang.so build/libleaf_parser.so build/libleaf_util.so build/libleaf_runner.so build/libleaf_ir.so build/libleaf_ir_llvm.so -Lbuild/build/lib -Lsrc/build/lib -L/usr/lib -Lbuild -Lsrc -L/opt/llvm/install/lib -L/usr/lib -lboost_unit_test_framework -lboost_program_options -lrt -ldl -lboost_regex -lLLVM-3.8 -lgmp -lgmpxx -lboost_filesystem -lboost_system

I'm currently running Red Hat 7.3 and installed Python 3.5 from the SCL (www.softwarecollections.org/en/scls/rhscl/rh-python35/). When I attempt to pip install C intensive packages such as numpy and pandas, the install process on Python 3.5 is taking significantly longer than when I attempt to install the same packages in the native Python 2.7 installation (6 minutes per package versus ~10 seconds).

I have some automated processes that are building and rebuilding virtual environments on a frequent basis, so this is having a huge impact on the overall performance. Does anyone know why these installations are taking significantly longer in Python 3.5? Any help would be greatly appreciated.

Here's a snippet of the 'pip install numpy -v' on both versions. The obvious thing that jumps out at me is the GCC building that occurs in 3.5 and not in 2.7 but I'm not sure why...

Native Python 2.7:

I'm on a cluster on which I have no rights, and I'm trying to pip install mpi4py. Since I cannot install the python3-devel package, I downloaded it and placed it in ~/.local/. Regardless of whether this has a chance of succeeding, the following confuses me.

If I simply run pip3.4 install --user mpi4py, I am met with this error: