On 6/28/21 5:38 PM, John Snow wrote:
On Mon, Jun 28, 2021 at 4:31 PM Wainer dos Santos Moschetta
<waine...@redhat.com <mailto:waine...@redhat.com>> wrote:
On 6/25/21 12:45 PM, John Snow wrote:
> tox is already testing the most recent versions. Let's use pipenv to
> test the oldest versions we claim to support. This matches the
stylistic
> choice to have pipenv always test our oldest supported Python
version, 3.6.
Currently tox is testing with 3.6 too. Should we remove that version
from the test matrix?
Tox will test with 3.6, but against the very latest pylint/mypy. This
test will also use 3.6, but against the oldest pylint/mypy we claim to
support in the package metadata. I think they are both useful.
The pipenv test will make sure we don't accidentally increase our
dependency requirements without noticing. This has some value for
other QEMU developers that are used to running and managing their own
linters, such as we did for a long time. I don't want to break their
environments by accident. (It also makes sure that no 3.7+ isms sneak
into the code by accident, too. This has historically been, and I
anticipate it will continue to be, a prominent problem in the Python
library.)
The tox test gives us a heads up if there are incompatible changes
arriving in the bleeding edge for pylint/mypy et al that I will need
to be aware of before they are rolled out and wind up on random
developer's systems and start causing 'make check' to fail.
I think it's worth keeping both for now, provided the run cost isn't
too great. At the moment it seems rather minimal.
While I was reviewing the next patches I started to understand the tox
and pipenv testing strategies. I also think they are useful.
Thanks for the details explanation above.
- Wainer
>
> The effect of this is that the python-check-pipenv CI job on
gitlab will
> now test against much older versions of these linters, which
will help
> highlight incompatible changes that might otherwise go unnoticed.
>
> Update instructions for adding and bumping versions in
setup.cfg. The
> reason for deleting the line that gets added to Pipfile is
largely just
> to avoid having the version minimums specified in multiple places in
> config checked into the tree.
>
> (This patch was written by deleting Pipfile and Pipfile.lock, then
> explicitly installing each dependency manually at a specific
> version. Then, I restored the prior Pipfile and re-ran `pipenv lock
> --dev --keep-outdated` to re-add the qemu dependency back to the
pipenv
> environment while keeping the "old" packages. It's annoying,
yes, but I
> think the improvement to test coverage is worthwhile.)
>
> Signed-off-by: John Snow <js...@redhat.com
<mailto:js...@redhat.com>>
> ---
> python/Pipfile.lock | 113
+++++++++++++++++++++-----------------------
> python/setup.cfg | 4 +-
> 2 files changed, 56 insertions(+), 61 deletions(-)
Reviewed-by: Wainer dos Santos Moschetta <waine...@redhat.com
<mailto:waine...@redhat.com>>
Thanks!
