This looks great - I have one concern with tools like this - what is the update flow when the template is updated?
This is a thing I've been struggling with - what is the best way of delivering such templates. Is a template generator such as this? Would it be best to copy a baseline repo with all options set? Should we fork that base repo and rebase when the baseline updates?
Great question because this has haunted me for over 10 years with cookiecutter. Most of the time I was copy-pasting newer features from running projects back into the template, and then copy-pasting in other running projects.
But Keel is based on copier which has a "pull" based approach. It means unlike cookiecutter where you generate project from template once, you can run `copier update` to update your project with the latest addition to the template. (It takes care of merging based on the initial config that you selected).
The fork approach came up at work - because all projects have a lot of boilerplate in common (Terraform, K8s configs, secrets, etc) that can receive improvements upstream from the ops team and it'd be nice if the improvements were easy to apply.
The `copier` thing is nice. I'll sure try it on my next Django project.
Definitely, if you find anything that can be improved (even if "docs are not super-clear here), please feel free to raise a PR or report it as an issue.
15 years of Django for me personally. I use it now in my SaaS as a data backend with Strawberry GraphQL and lean heavily in on the admin. My frontend is React Router 7 framework mode. I’m going to be checking into Keel for my new toy projects. I wish it would use ty but easy enough to add myself.
Python 3.14 is there in the repo, though not added in the docs since I haven't tested it with all the infra stuff. The idea is to keep it updated to at the least previous 2 versions of Python/Django at all times so that it's easier for people to keep their projects updated.
Quite honestly, I haven't. I've been playing with temporal, and since it's picking up pace, I thought to have an option to have Temporal in place of celery.
Yes, I maintained a cookie-cutter template for over a decade, and it was a pain for me to get it updated. I wanted to get something newer, like a copier, which eliminates the pain of updating, while I can focus on building things.
I have also added temporal (though still testing it) since it's getting a lot popular these days.
Also, to answer why ever be less than 100% is that it depends. Usually, I prefer writing behavioural & integration tests, over chasing per-line coverage which might give a false confidence.
I'm not saying getting 100% coverage is bad, but with my experience working on several different projects, I've seen that the last 10-20% is usually low-value surface.
For critical modules like auth, money, policies (business-logic), should aim at close to > 95%.
When you reach that 100% mark, in my opinion, you might have to write brittle tests which slows down your refactoring with little risk reduction.
Again, to summarize, it's not a black/white answer on having 100% or no coverage, but this differs from project to project, and generally having anything above 80% that covers the core-business logic of projects tends to work in my opinion.
This looks great - I have one concern with tools like this - what is the update flow when the template is updated?
This is a thing I've been struggling with - what is the best way of delivering such templates. Is a template generator such as this? Would it be best to copy a baseline repo with all options set? Should we fork that base repo and rebase when the baseline updates?
Great question because this has haunted me for over 10 years with cookiecutter. Most of the time I was copy-pasting newer features from running projects back into the template, and then copy-pasting in other running projects.
But Keel is based on copier which has a "pull" based approach. It means unlike cookiecutter where you generate project from template once, you can run `copier update` to update your project with the latest addition to the template. (It takes care of merging based on the initial config that you selected).
It's just running one command `copier update`. Docs here: https://django-keel.readthedocs.io/en/latest/?h=copier+updat...
You don't need to fork it unless you want to make some improvements or build upon the base template itself.
The fork approach came up at work - because all projects have a lot of boilerplate in common (Terraform, K8s configs, secrets, etc) that can receive improvements upstream from the ops team and it'd be nice if the improvements were easy to apply.
The `copier` thing is nice. I'll sure try it on my next Django project.
Definitely, if you find anything that can be improved (even if "docs are not super-clear here), please feel free to raise a PR or report it as an issue.
[dead]
Cruft solves exactly this problem https://cruft.github.io/cruft/
15 years of Django for me personally. I use it now in my SaaS as a data backend with Strawberry GraphQL and lean heavily in on the admin. My frontend is React Router 7 framework mode. I’m going to be checking into Keel for my new toy projects. I wish it would use ty but easy enough to add myself.
Interesting! I've loved Strawberry GraphQL.
If you end up adding anything, you may also choose to give back to the template so that others can use it.
I do not understand what you meant by "ty", though?
Ty is the typechecker by Astral. https://docs.astral.sh/ty/.
Reviewing the dependencies in the jinja file -- how difficult is it to keep these up to date? I see Django Debug Toolbar is still ^4.3.0.
I think it's ty the typechecker :D https://docs.astral.sh/ty/
Thanks, this seems interesting!
As someone who up to now has mainly watched Django from the sidelines, as it were, this looks great.
What's are your plans for supporting Django v6? (I appreciate it's just gone alpha now, and planned prod release is not for two months.)
Likewise for Python 3.14.
Happy to let you know that I've just updated the template to use Python3.14 by default, along with matrix tests that also tests the template against Python 3.14 in here: https://github.com/CuriousLearner/django-keel/commit/87eae7e...
Python 3.14 is there in the repo, though not added in the docs since I haven't tested it with all the infra stuff. The idea is to keep it updated to at the least previous 2 versions of Python/Django at all times so that it's easier for people to keep their projects updated.
Looks great, looking forward to trying it.
Interesting that you included Temporal for background task processing — did you ever use it in production in a Django project instead of Celery?
Quite honestly, I haven't. I've been playing with temporal, and since it's picking up pace, I thought to have an option to have Temporal in place of celery.
Nice. Batteries included for major frameworks is always great.
Yes, I maintained a cookie-cutter template for over a decade, and it was a pain for me to get it updated. I wanted to get something newer, like a copier, which eliminates the pain of updating, while I can focus on building things.
I have also added temporal (though still testing it) since it's getting a lot popular these days.
> pytest - Comprehensive test suite (80% coverage minimum)
Why would this ever be less than 100%?
It's configurable. The minimum quality gate check is set at 80% here: https://github.com/CuriousLearner/django-keel/blob/main/temp...
Also, to answer why ever be less than 100% is that it depends. Usually, I prefer writing behavioural & integration tests, over chasing per-line coverage which might give a false confidence.
I'm not saying getting 100% coverage is bad, but with my experience working on several different projects, I've seen that the last 10-20% is usually low-value surface.
For critical modules like auth, money, policies (business-logic), should aim at close to > 95%.
When you reach that 100% mark, in my opinion, you might have to write brittle tests which slows down your refactoring with little risk reduction.
Again, to summarize, it's not a black/white answer on having 100% or no coverage, but this differs from project to project, and generally having anything above 80% that covers the core-business logic of projects tends to work in my opinion.
It was not at all clear to me that number was for code coverage. It sounded like it was permitting up to 20% of tests to be failing.
Interesting. Do you have any opinion on what this should be rephrased to, so that it's clearer?
Explicitly say code coverage or lines of code?
[dead]
encountered this post in reddit maybe an hour ago. the comment sections are very different. i'm gonna stick with HN for sure.
good project though. using copier instead of cookiecutter is a good choice.
It was all great on Reddit, though until some folks decided to downvote for no apparent reason and then deleted their accounts.
I've chosen copier because of all the pain I've gone through for years just to keep my template and projects running in parallel updated.
one small note, you should move quick start section in README to be seen earlier
Thank you for your suggestion. I'll update the README so that people can hit the ground running faster!
looks great
[dead]