Fix check-publish-compile job

[skunert]

One more red job turned green. Did not investigate how misformated prdoc 6689 made it in, but this fixes the check-publish-compile job on CI.

Failure example: https://github.com/paritytech/polkadot-sdk/actions/runs/12925909945/job/36047953474

[skunert]

Okay looks like this PR just unblocks the stage that was failing before, now reveiling that the compile stage of the job also fails 🤦

[skunert]

With #7187 the compilation succeeds. But we can not just remove bandersnatch-experimental from the master branch.

[ggwpez]

Gonna request changes to not accidentally merge it now.
Turns out the bandersnatch crate is still causing issues. I tried to publish it now, lets see if it works.

[skunert]

So now the job fails with:

error: failed to select a version for `sp-application-crypto`.
    ... required by package `sp-consensus-sassafras v0.3.4 (/__w/polkadot-sdk/polkadot-sdk/substrate/primitives/consensus/sassafras)`
    ... which satisfies path dependency `sp-consensus-sassafras` of package `pallet-sassafras v0.3.5 (/__w/polkadot-sdk/polkadot-sdk/substrate/frame/sassafras)`
versions that meet the requirements `^39.0.0` are: 39.0.0

the package `sp-consensus-sassafras` depends on `sp-application-crypto`, with features: `bandersnatch-experimental` but `sp-application-crypto` does not have these features.

So it is depending on a version of [email protected] which does not have the bandersnatch-experimental feature. Since we have it locally, it must be the one from crates.io which is also at 39.0.0.

Looking at the Cargo.toml that is generated by parity-publish, it has sp-application-crypto without a path set, maybe because it is the same version number as on crates.io?

sp-application-crypto = { default-features = false, version = "39.0.0" }

So maybe problem is that sp-application-crypto is at 39 on stable2412 which has been published without bandersnatch-experimental, but on master we are also at 39 but with the feature. We would need to bump the version on master for every crate that has been published without the feature. In general, are these versions from stable supposed to get ported to master? They must be right?

[ggwpez]

So maybe problem is that sp-application-crypto is at 39 on stable2412 which has been published without bandersnatch-experimental, but on master we are also at 39 but with the feature.

Yes that sounds likely. We stripped that feature AFAIK before publishing.
I guess we can only fix this in the next release; 2503 by publishing it with the feature.

In general, are these versions from stable supposed to get ported to master? They must be right?

Not sure what you mean. The feature stripping is done right before publish, we may not even have the code that was published... still trying to get new ppl into the release team to clean up this mess.