[Scons-users] Possible python 3 issue?

[Mats Wichmann]

On 10/16/19 10:04 AM, Peter Diener wrote:
> Hi,
> 
> I have been successfully using scons 3.1.0 with python 2.7.12 to build a 
> mostly Fortran based software project on Linux (multiple distributions). 
> I have the source tree separated from the build directory and are 
> therefore using variantdir and an SConscript file. Recently I tried to 
> run this with python 3.7.4 and found that it did not work as scons 
> failed to correctly find the dependencies on module files.
> 
> I was able to reproduce this with a smaller test example that mimics the 
> setup in my larger software project. I have attached a zip file with 
> that source tree as well as build log files (in build_log.2.7.12 and 
> build_log.3.7.4) with all errors and output when building with
> scons --tree=all. As can be seen from the build log files, when building 
> with python 3, scons does not seem to find the dependencies of the main
> program on the module files.
> 
> I have also tried with the latest release version of scons 3.1.1 with
> the same results. It works with python 2 but not with python 3.
> 
> So my question is: am I doing something wrong in my SConstruct and
> SConscript files that just happens to work with python 2 and not with 
> python 3? Or is this a problem with scons and python 3?

Possibly a bit of both.  There's an existing scons issue that points out 
some weakness in submodule support (#3366), and given that Fortran isn't 
the most heavily used language in the scons universe, there may still be 
untested code paths in the conversion to Python 3, that don't get 
exercised by what's in the test suite (there are some fairly new tests 
in that area, which you know since you contributed there :) )