meta should be last.

This is not how we do such things in pythonPackages. I assume the packages share the module name which means we can't do it like this and need to overwrite it for every end application that we want to fix. This is due to limitations in python which we did not come around yet.

Also if we would add another version it would be in top-level named wtforms_2.

feel free to suggest an alternative, but I'd rather revert #155221 and override it for the single package that needs 3.0 (which hasn't even been merged yet?) rather than copy-paste this into like 10 different packages :(

I agree that this isn't great, but it's meant as a stop-gap to be removed once the rest of the ecosystem has been updated. Copying it all would hurt me too much to submit, though feel free to amend this PR accordingly (I think that's how that github checkbox works?)

This is is forbidden. We cannot use it as a stop gap. It causes massive issues when the two versions end up in a single python env. We can also revert the update of wtforms and override it for the one package that actually needs it.

I think we should check if we can fetch spiral-project/ihatemoney#916.

I don't think we even need this.

Well unless you're applying an overlay to the python package set, anything that indirectly depends on wtforms must also be overridden to keep it out of the closure.

But yes, this can instead be overridden manually everywhere flask_wtf is used, if that's what you mean.

You want to use something like that https://github.com/NixOS/nixpkgs/blob/master/pkgs/applications/misc/privacyidea/default.nix#L6-L39 in the final package.

We should check if dpgaspar/Flask-AppBuilder#1734 is already enough.

airflow is so often broken, I don't really care anymore at this point. I would suggest to just leave it like it is and let someone else care about it.

We shouldn't do this. It will cause issues which will be hard to debug.

so... leave it broken or what do you want here?
buku appears to be the only user of this module, so the override could stay isolated to that.

yes for python modules (as opposed to leaf packages) it's not possible to override their dependencies. So by bumping wtforms to 3, flask-admin was broken without solution. The only solution to prevent that is not to merge breaking updates in the first place. Please consider that next time before merging.

Would it help, if we add the version bump of WTForms as an extra package wtforms3? So everything that still relies on WTForms2 could use the original package, any any package needing the version 3, could use wtforms3?

Would it help, if we add the version bump of WTForms as an extra package wtforms3? So everything that still relies on WTForms2 could use the original package, any any package needing the version 3, could use wtforms3?

If they both end up in the same pythonEnv things still break because they share the same module name internally.

I'm not sure what an "empty" package means.

A package that contains no files. That happens when pip installs a package to $out which is already in the environment and then nothing happens except that $out is created and some directories and that ends up being the package.

of course once flask-admin is properly updated then this override would be obviously unneeded and removed.

Lets be real, there is a high chance someone will forget.

I'm not really aware of python version overrides being uncommon practice in nixpkgs

We nuked most of them from the tree a few months ago so except pytest there should be almost none.

I'm not sure what an "empty" package means.

A package that contains no files. That happens when pip installs a package to $out which is already in the environment and then nothing happens except that $out is created and some directories and that ends up being the package.

Well that's unfortunate as a failure case, but is a much more extraordinary case that isn't relevant here - plus could easily be guarded against if it were a real concern. A package that accidentally depends on itself if it weren't overridden would cause infinite recursion on evaluation, no? A situation where that is narrowly avoided due to a stray override seems like it would be an incredibly exceptional one. You'd also hope installtests would catch it, but .-.

of course once flask-admin is properly updated then this override would be obviously unneeded and removed.

Lets be real, there is a high chance someone will forget.

Sorry, I want to say I highly doubt your statement here, as it would be immediately obvious to anyone looking at the file to update it. Perhaps not to newcomers, though? and then if github doesn't show it in the diff... though, you could apply that as an argument to reject literally any change in nixpkgs, so it doesn't feel very useful?

I'm not really aware of python version overrides being uncommon practice in nixpkgs

We nuked most of them from the tree a few months ago so except pytest there should be almost none.

Gotcha, well it sucks that this was apparently done without any alternatives proposed. Is any of this policy documented somewhere?
(but really, the fact that nixpkgs allows merges to break any package is honestly the real problem here, unfortunately)

is a much more extraordinary case that isn't relevant here

I already got this once into master and unstable for 3 weeks and no one noticed that the checkInput poluted the environment and the package ended up being empty. 🙃

A package that accidentally depends on itself if it weren't overridden would cause infinite recursion on evaluation, no?

For nix they are two different packages but for python not and we currently do not have any safeguard for this.

Gotcha, well it sucks that this was apparently done without any alternatives proposed

Limitation of python which cannot easily and universally overcome.

Is any of this policy documented somewhere?

IDK, probably not.

Flask-admin got updated to version 1.6.0 which includes an update to WTForms 3:

1.6.0
Dropped Python 2 support
WTForms 3.0 support
Various fixes

I build it locally and it works flawlessly with WTForms3. The only thing I needed to change was to add flask_admin/tests/sqla/test_inlineform.py to disabledTestPaths.

Yeah it seems good here, replaced the override with the update!