It seems the sanest way (from a human perspective) of modelling an OPC-UA server is to create a (relatively) simple XML file describing data and relations (the "model design" format), convert it into another more machine-oriented XML file (the "NodeSet2" format) and then generate from it the C code to be linked against other server code.

The conversion from model design to NodeSet2 is performed by UA-ModelCompiler, a tool officially endorsed by the OPC foundation, while the generation of the C code from NodeSet2 is performed by nodeset_compiler.py, a Python script provided by the awesome Open62541 project.

While nodeset_compiler.py, apart from being included in PATH, does not require any particular processing, the UA-ModelCompiler is strongly geared towards Windows/DotNet and it is in general quite unpleasant to work with. In other words, compiling model designs on GNU/Linux systems is a nightmare.

The only documentation I found is the OPC-UA information model tutorial, which provides a strong foundation for understanding the whole process but the commands reported there for converting the model design simply do not work on my system (ArchLinux current).

After a lot of trials and errors, I came up with a solution based on docker that does not require building and installing UA-ModelCompiler natively. See the Makefile for the gory details.

The sample model design used here is an adaptation of the work done by Industry40tv on its YouTube tutorial: check it out for a really good introduction.