-
Notifications
You must be signed in to change notification settings - Fork 378
Missing feature for parachain runtime? #2230
Comments
fixes KILTprotocol/ticket#2464 Fixes #481 -> see issue description for a complete list of PRs. ## Relevant changes - paritytech/polkadot#4097 (XCM v3) - paritytech/polkadot#6490 (root origin can issue grants on behalf of the treasury) - paritytech/substrate#13214 (we used in the the staking pallet but we never called the `set` function, so we don't need to do anything) - paritytech/substrate#13216 (removes all calls and intrinsics from the authorship pallet) Release analysis: https://forum.polkadot.network/t/polkadot-release-analysis-v0-9-38/2122 ## Open issues If any of the runtimes is complied with just the `runtime-benchmarks` features, there is a compilation issue due to some nested dependencies. Linked issue: paritytech/cumulus#2230. --------- Co-authored-by: Tino Rusch <tino@kilt.io>
If I pull the latest version of the
|
Yea I looked into it locally but could not pinpoint where a feature is missing or erroneously being propagated… |
hmm, I have the similiar/same problem here: https://gitlab.parity.io/parity/mirrors/cumulus/-/jobs/2490428 |
It is currently breaking our runtimes compilation whenever the feature is enabled. The node compilation is fine, probably because the default |
@bkchr cool, the fix works when compiling the template runtime with the benchmark feature only. Could this be backported to at least the 0.9.38 branch? Or will it be only included in the 0.9.39? |
I already backported it to the 0.9.38 branch. (this is the commit I linked above) |
Can we close this @ntn-x2? |
Yes, issue seems fixed also for our runtime compilation. Thanks for the quick fix! |
Fixes compilation for the runtimes when the `runtime-benchmarks` feature was enabled. For more context, see the [Cumulus GH issue](paritytech/cumulus#2230). Before, `cargo check -p peregrine-runtime --features runtime-benchmarks` was broken. Not anymore.
fixes KILTprotocol/ticket#2464 Fixes #481 -> see issue description for a complete list of PRs. - paritytech/polkadot#4097 (XCM v3) - paritytech/polkadot#6490 (root origin can issue grants on behalf of the treasury) - paritytech/substrate#13214 (we used in the the staking pallet but we never called the `set` function, so we don't need to do anything) - paritytech/substrate#13216 (removes all calls and intrinsics from the authorship pallet) Release analysis: https://forum.polkadot.network/t/polkadot-release-analysis-v0-9-38/2122 If any of the runtimes is complied with just the `runtime-benchmarks` features, there is a compilation issue due to some nested dependencies. Linked issue: paritytech/cumulus#2230. --------- Co-authored-by: Tino Rusch <tino@kilt.io>
Fixes compilation for the runtimes when the `runtime-benchmarks` feature was enabled. For more context, see the [Cumulus GH issue](paritytech/cumulus#2230). Before, `cargo check -p peregrine-runtime --features runtime-benchmarks` was broken. Not anymore.
After upgrading from 0.9.37 to 9.9.38, our parachain runtimes compiled only with the benchmarking flag have started failing.
The error is generated in the
polkadot-runtime
crate, which I am not sure how deep inside the dependency tree is, and it seems an issue with theruntime-benchmarks
flag not being properly set by one of the crates that our runtime depends on.On a related note, I was not even able to compile the statemine runtime on its own with just the
runtime-benchmarks
feature either, but with a different error regarding the newtry_success_origin
syntax, which gives me the hint that not all feature combinations are working for the cumulus repo.EDIT: this is the PR to update from 0.9.37 to 0.9.38. If our parachain runtimes are compiled with
cargo build -p peregrine-runtime --features runtime-benchmarks
, then we have the compilation issue.EDIT 2: compiling the whole
kilt-parachain
node with just theruntime-benchmarks
feature works, so something in the client side is causing the feature from being correctly enabled by the dependencies.The text was updated successfully, but these errors were encountered: