Thanks Jarek. That looks like a nice improvement for sure. > and I guess many people just cancelled
That feels like a personal attack haha! Thanks & Regards, Amogh Desai On Tue, Oct 21, 2025 at 1:14 PM Jarek Potiuk <[email protected]> wrote: > Also another - related QOL improvement/suggestion. I noticed that often > when I push my change to my fork, I have many more files detected with > `pre-push` enabled. This is because often `main` of my fork is not synced > with the main of Airflow. > > The way how I solved that I created a small alias (with the help of `gh`) > to sync my fork (origin is the one I use for my private fork) > > ``` > alias sync-repo='gh repo sync potiuk/airflow --branch main && git fetch > origin' > ``` > > On Sun, Oct 19, 2025 at 2:43 AM Jarek Potiuk <[email protected]> wrote: > > > Replace airflow mypy with one of those of course :) > > > > > > On Sun, Oct 19, 2025 at 2:42 AM Jarek Potiuk <[email protected]> wrote: > > > >> Oh absolutely. At last PyCon I met the Pyrefly team and we even ran it > >> together on Airflow, and of course we know the team at Astral and ty. > The > >> issue at that time was that neither was close to handle Airflow (Pyrefly > >> out of the box was >10 thousands issues) and ty was very early (stil is) > >> but mid-term goal is to replace airflow with one of those. > >> > >> One of the problems we have is that some of our type-checking depends on > >> custom mypy plugins and none of those has any support for it (did not > check > >> zuban -but it seems one-man show, which is a bit worrying for such a > >> complex thing like Python typechecker). > >> > >> But yeah. If someone would like to give it a shot and see how far we are > >> and maybe lead the effort of switching to one of those... Absolutely :) > >> > >> J. > >> > >> On Sun, Oct 19, 2025 at 2:34 AM Dev iL <[email protected]> wrote: > >> > >>> Thanks Jarek, I know this will definitely make a big difference for me! > >>> > >>> It might be a good opportunity to recommend to those of us who want > >>> dev-time type checking to look into Astral's ty ( > >>> https://github.com/astral-sh/ty), Meta's pyrefly ( > >>> https://github.com/facebook/pyrefly), or David Halter's zuban ( > >>> https://github.com/zubanls/zuban/) - all of which are written in Rust > >>> and > >>> are much faster than mypy. > >>> > >>> On Sat, 18 Oct 2025, 23:13 Jarek Potiuk, <[email protected]> wrote: > >>> > >>> > Hello Everyone, > >>> > > >>> > I just merged something that might turn into quite a QOL improvement > >>> for > >>> > all contributors (it looks like a significant QOL improvement for my > >>> > workflows at least): https://github.com/apache/airflow/pull/56829 > >>> > > >>> > This PR moves `mypy` checks to the "pre-push" stage of prek hoos from > >>> > "pre-commit". You might want to run `prek install --refresh` now. to > >>> take > >>> > advantage of it. And follow up with `prek install --hook-stage > >>> pre-push` to > >>> > complete it (opt-in). > >>> > > >>> > I guess a number of people refrained from `prek install` so far > because > >>> > `mypy` checks were generally slow and required `breeze ci-image build > >>> > --python 3.10` to be run not only once but also kept up to date. > While > >>> we > >>> > have also other hooks that require `ci-image` - mypy is one that was > >>> > triggered by pretty much any python file change - and I guess many > >>> people > >>> > just cancelled it or even disabled prek auto-hooks because of the > >>> waiting > >>> > time at commit time. > >>> > > >>> > With this change - mypy hooks are moved to `pre-push` stage - this > >>> stage is > >>> > not enabled when you run `prek install`, you need to explicitly > enable > >>> it > >>> > by `prek install --hook-type pre-push`. So you might opt-in to it (I > >>> did) > >>> > while not impacting regular pre-commit stage hooks. > >>> > > >>> > What it really means for contributor's workflows: > >>> > > >>> > * If you did `prek install`, the `git commit` operation should be > >>> faster > >>> > now and will not run mypy checks, If you did not, good idea is to run > >>> it > >>> > now. I highly recommend doing so. Lots of things are auto-fixed when > >>> you do > >>> > and you save precious push -> fix -> push cycle. > >>> > > >>> > * if you did (or will do) `prek install --hook-type pre-push` - all > >>> > pre-commit and mypy hooks will run at the "git push" time - but this > is > >>> > more opt-in now, You can always do `git push --no-verify` as with > >>> commit to > >>> > skip that step though > >>> > > >>> > * if you did not run `prek install` at all beforr, but keep on > running > >>> > `prek` manually, this will also speed up plain `prek` execution. The > >>> > default for `prek` is to run `pre-comit` hook stage - so if you run > >>> `prek` > >>> > locally - it will run all the checks for your staged changes - but > not > >>> mypy > >>> > any more. You will need to run `prek run --hook-stage pre-push` to > run > >>> also > >>> > mypy checks or run "manual" (full folder) version of checks `prek run > >>> > --hook stage manual` (optionally wiht --all-files to force full > check). > >>> > > >>> > I hope this will improve QOL and iteration speed for a number of > >>> > contributors. Also If there are any ideas, problems, obstacles, > >>> > difficulties that the current setup causes - discussing it in > devlist, > >>> > #contributors channel in slack is a good idea. Also the CI/DEV stuff > >>> is not > >>> > as hard, and many people already contribute - regularly or casually, > >>> and I > >>> > think all of us in the ci/dev team are happy to get feedback and > ideas > >>> from > >>> > everyone contributing. > >>> > > >>> > Just a side comment - honestly I have no idea why I had not thought > >>> about > >>> > such setup before with `pre-push` hooks for mypy. It seems pretty > >>> "obvious" > >>> > when you ask now, I think it's partially caused by blind-spot > developed > >>> > from years of doing stuff "this way". > >>> > > >>> > So any fresh and out-of-the box ideas are more than welcome and we > >>> have now > >>> > a strong team of CI/Dev peoople who will hear it and respond. I think > >>> in > >>> > the CI/DEV team we like to do stuff our ways, but we like even more > if > >>> we > >>> > can learn new tricks (speaking as an Old Dog). > >>> > > >>> > J. > >>> > > >>> > >> >
