On 2023/09/20 17:44, Daniel Dickman wrote: > > > On Wed, 20 Sep 2023, Stuart Henderson wrote: > > > need runtime checks of dependents, these problems don't show up in build > > > > The good news is that I didn't notice anything new that broke at runtime. > > The bad news is tons of stuff in the Python tree is already broken at > runtime. This has nothing to do with click though, that's a pretty > generic comment. > > Just looking at the click dependencies alone... some examples of either > outright brokeness or missing test depends: > > devel/py-xdis -- broken when we went past python 3.10.9 (and so is > devel/py-uncompyle6 which has the same issue of hardcoding very specific > python versions). These can be fixed by someone who cares enough to update > to latest versions on github.
Those are broken with current python versions more often than not (though iirc still do work but can only handle bytecode files from older python versions). Would need to be git head as there are rarely releases. > www/tootstream -- broken at runtime (ImportError: cannot import name > 'attr' from 'colored'). Not sure if an update to py-colored or tootstream > itself broke runtime. I've fixed this with an update to py-colored and a patch to tootstream. > security/boofuzz -- missing TDEP on pytest_bdd > > security/yubico/yubikey-manager -- missing TDEP on makefun > > math/py-snuggs still has the same 3 failing tests with no activity > upstream in the last 4 years and users of geo/py-rasterio asking the > rasterio probject to find some way to migrate away from snuggs or fork it. > No change from a click update. > > Of the rest of the things I checked, I didn't see new breakage due to > click. Some of them even had test suites that ran, most things don't have > test suites or generate lots of errors when trying to run the tests. A lot of test in python ports are missing bits, I think a lot of people were running with network access and it was fetching missing deps on the fly so they didn't notice, also people running many other tests of python ports will tend to have a lot of the required things installed already so they can get missed from TEST_DEPENDS. The usual symptom of something breaking with click is that the scripts don't run (seeing as it's a command-line parser etc). Things used to break with nearly every click update but perhaps they've got a bit more stable now. (I suspect that running test suites of depending ports won't actually show up click problems as they usually run the code from python rather than via the cli script). > So from what I can tell, I think the update to click itself is ok. ok with me then.
