Re: testing with python 3.6 3.7

On 10/30/18 7:49 AM, Jason Kenny wrote:
> Hi,
> I was wondering if SCons had noticed any testing issues with python 3.6 or newer. I noticed as I was porting Parts to python3 I was having some testing issues related to testing the output of some values that had been controlled by values in dict object. I discovered that with python 3.6 and made official in 3.7 the insert order is preserved. This caused some false positive failures for me as values. I was just curious if this was seen with the scons move to python3 as well.
> Jason

the CI infrastructure (travis and appveyor) builds everything against
3.5, 3.6 and 3.7, so those are seen.  There have been several cleanups
as a result, but mostly those problems have been noise - 3.6 and 3.7
issuing deprecation warnings by default now, and if those end up in the
"expected" stderr stream of a test, then it causes problems for that
test.  There's also a lot of noise from unclosed file descriptors when
launching other processes via subprocess.  I don't recall seeing any
test affected by dictionary order... some of the code was already using
OrderedDict, and I did send a PR through to remove the internal custom
OrderedDict in favor of the standard library one.

Others may recall something additional...