Re: GRC version 3.9.0.0 git master (python 3.8.2) + OOT blocks

[David Taylor (manx.net)]

Josh,

 

Sorry to go over old ground yet again.

 

In order to remove any additional problems caused by the use of external libraries, I have created a separate OOT branch that is void of external libraries.

Stripped to the bone python and C++ blocks have been created using gr_modtool. Test codes validate the blocks code content as functional and they import correctly into the GRC with simple .yml.

The top level CMakeLists.txt and others are unmodified from a clean new module.

 

The current issue has been traced so far to the __init__.py file and its inability to resolve ‘__path__’

Consequently the pybind11 link fails to find my ‘test’ directory and the subsequent imported python block fails with no known parent package.

I am unclear how and where this is initialised under python3? I have attended to the usual library and path ~./bashrc changes to no avail.

 

The GRC outputs the AttributeError: message for the python and C++ OOT block.

 

Once resolved, I can move forward again with the external library inclusion.

 

Many thanks,

David GD4FMB

 

 

 

From: Josh

Sent: Monday, August 17, 2020 11:45 AM

To: David Taylor (manx.net)

Cc: GNURadio Discussion List

Subject: Re: GRC version 3.9.0.0 git master (python 3.8.2) + OOT blocks

 

David,

 

I've found most of the time I get the "No module named ..." error it is due to C++ linkage issues in setting up the CMake.  There was a big jump in CMake modernization from GR 3.7 to 3.8, so be sure to use gr_modtool (from 3.9) to create a new module and copy your blocks in from there is usually the easiest way.  Porting guide is here: https://wiki.gnuradio.org/index.php/GNU_Radio_3.8_OOT_Module_Porting_Guide and here: https://wiki.gnuradio.org/index.php/GNU_Radio_3.9_OOT_Module_Porting_Guide

 

the one that usually gets me and causes the ModuleNotFoundError is this

https://wiki.gnuradio.org/index.php/GNU_Radio_3.8_OOT_Module_Porting_Guide#GNU_Radio_Components

 

Does your library reference any GR modules, or any other external libraries?

 

Josh

 

 

On Mon, Aug 17, 2020 at 3:58 AM David Taylor (manx.net) <drtaylor@manx.net> wrote:

Hi,

 

I have been developing a number of direct spread spectrum OOT blocks as part of a research project.

 

Working blocks were originally developed using GRC 3.7.11, however I wish to move forward and have installed and persevered so far with GRC 3.9 from the master branch.

 

The GRC, UHD, CMAKE (3.16) pybind11 (2.4.3) and other dependencies have been installed and build correctly. 

 

I have chosen to migrate the (3.7.11) C++ blocks and for completeness of the build process checking, have included a simple python OOT.

The C++ to python code binding, make and install under Ubuntu 20.04 all work and the new blocks import correctly to flow-graph using modified .yml descriptors.

 

1). GRC 9.0 works standalone from git-master install and with the UHD, in my case a B210.

2). OOT blocks including the aforementioned python OOT block all fail at import. In my case ModuleNotFoundError: No module named ‘development’

i.e. failure of ‘import development’ in the flow-graph python script

 

3). I have tried and retained the library workarounds with PYTHONPATH and LD_LIBRARY_PATH, but these now seem irrelevant as the GRC basically loads and runs.

 

4). I have looked at the gr_modtool  __init_.py file for indicators as to why both python and C++ blocks (using python bindings) both fail.

The inability of python 3.8.2 in my case to resolve the import is clearly at its core.

 

5). The OOT GRC blocks themselves import correctly into the flow-graph produce error free python script and all have relatively primitive interfaces.

 

Many thanks,

 

David Taylor