textproc/py-sphinx: Fix build

[jrm]

Summary

poudirere testport and poudriere bulk -t were failing with

=>> Checking for filesystem violations... done
=>> Error: Filesystem touched during build:
extra: usr/local/lib/python3.9/site-packages/build/__pycache__
extra: usr/local/lib/python3.9/site-packages/packaging/__pycache__
extra: usr/local/lib/python3.9/site-packages/flit_core/vendor/__pycache__
extra: usr/local/lib/python3.9/site-packages/flit_core/vendor/tomli/__pycache__
extra: usr/local/lib/python3.9/site-packages/flit_core/__pycache__
extra: usr/local/lib/python3.9/site-packages/tomli/__pycache__
extra: usr/local/lib/python3.9/site-packages/pyproject_hooks/__pycache__
extra: usr/local/lib/python3.9/site-packages/pyproject_hooks/_in_process/__pycache__

Work around this by not writing byte code.

[matthew]

I believe there was a recent, possibly contraversial, change to generate .pyc files dynamically at pkg install time using a pkg script, rather than including those in the main package. This apparently parallels the way python wheels work, and has a number of advantages.

One side effect has been that all of the pycache directories now show up as "checksum mismatch" in the daily security e-mails.

[mandree]

In D39261#893754, @matthew wrote:

I believe there was a recent, possibly contraversial, change to generate .pyc files dynamically at pkg install time using a pkg script, rather than including those in the main package. This apparently parallels the way python wheels work, and has a number of advantages.

One side effect has been that all of the pycache directories now show up as "checksum mismatch" in the daily security e-mails.

That change is being discussed in https://reviews.freebsd.org/D34739.
It was not controversial but premature, it caused regressions more severe than that poudriere bickering about extra __pycache__ folders.

I would propose that we fix the Mk/Uses/python.mk to generate proper pkg lists, and NOT add code that causes checksum errors because something is listing files behind pkg's back. This may require some feature extension of pkg so that it can record proper checksums for post-install/trigger script modifications.

This change (setting MAKE_ENV+=PYTHONDONTWRITEBYTECODE=1 MAKE_ENV) however only passes the blame down to ports that use sphinx during their build, because they will then trigger fs violation warnings inside poudriere, so I find should not be made. We are really looking at a general flaw and fixing individual ports is wasted effort.

In the meanwhile, poudriere-devel (not sure if the stable version, too) supports a PORTTESTING_FATAL=no in /usr/local/etc/poudriere.conf or $PREFIX/etc/poudriere.conf for that matter.

[antoine]

Please don't do this.
This has to be fixed in the framework, not in individual ports.

[jrm]

No problem. I'll wait for the framework fix and ignore poudriere testport failures for now.