Zepter

Analyze, Fix and Format features in your Rust workspace. The goal of this tool is to have this CI ready to prevent common errors with Rust features.

Install

cargo install zepter -f --locked

Commands

zepter

• : this is the same as run.
• run: Run a workflow from the config file. Uses default if none is specified.
• format
  • features: Format features layout and remove duplicates.
• trace: Trace dependencies paths.
• lint
  • propagate-features: Check that features are passed down.
  • never-enables: A feature should never enable another other.
  • never-implies (⚠️ unstable): A feature should never transitively imply another one.
  • only-enables (⚠️ unstable): A features should exclusively enable another one.
  • why-enables (⚠️ unstable): Find out why a specific feature is enables.
• debug: (⚠️ unstable) just for quick debugging some stuff.
• transpose (⚠️ unstable)
  • dependency
    • lift-to-workspace: Lifts crate dependencies to the workspace.

Example - Feature Formatting

To ensure that your features are in canonical formatting, just run:

zepter format features
# Or shorter:
zepter f f

The output will tell you which features are missing formatting:

Found 37 crates with unformatted features:
  polkadot-cli
  polkadot-runtime-common
  polkadot-runtime-parachains
  ...
Run again with `--fix` to format them.

Re-running with --fix/-f:

Found 37 crates with unformatted features:
  polkadot-cli
  polkadot-parachain
  polkadot-core-primitives
  polkadot-primitives
  ...
Formatted 37 crates (all fixed).

Looking at the diff that this command produces; Zepter assumes a default line width of 80. For one-lined features they will just be padded with spaces:

-default = [
-       "static_assertions",
-]
+default = [ "static_assertions" ]

Entries are sorted, comments are kept and indentation is one tab for your convenience 😊

-       # Hi
-       "xcm/std",
        "xcm-builder/std",
+       # Hi
+       "xcm/std",

Example - Fixing feature propagation

Let's check that the runtime-benchmarks feature is properly passed down to all the dependencies of the frame-support crate in the workspace of Substrate. You can use commit 395853ac15 to verify it yourself:

zepter lint propagate-feature --feature runtime-benchmarks -p frame-support --workspace

The output reveals that some dependencies expose the feature but don't get it passed down:

crate 'frame-support'
  feature 'runtime-benchmarks'
    must propagate to:
      frame-system
      sp-runtime
      sp-staking
Found 3 issues and fixed 0 issues.

Without the -p it will detect many more problems. You can verify this for the frame-support which is indeed missing the feature for sp-runtime while sp-runtime clearly supports it 🤔.

This can be fixed by appending the --fix flag, which results in this diff:

-runtime-benchmarks = []
+runtime-benchmarks = [
+       "frame-system/runtime-benchmarks",
+       "sp-runtime/runtime-benchmarks",
+       "sp-staking/runtime-benchmarks",
+]

The auto-fix can be configured to enable specific optional dependencies as non-optional via --feature-enables-dep="runtime-benchmarks:frame-benchmarking" for example. In this case the frame-benchmarking dependency would enabled as non-optional if the runtime-benchmarks feature is enabled.

Example - Feature tracing

Let's say you want to ensure that specific features are never enabled by default. For this example, we will use the try-runtime feature of Substrate. Check out branch oty-faulty-feature-demo and try:

zepter lint never-implies --precondition default --stays-disabled try-runtime --offline --workspace

The precondition defines the feature on the left side of the implication and stays-disabled expressing that the precondition never enables this.

Errors correctly with:

Feature 'default' implies 'try-runtime' via path:
  frame-benchmarking/default -> frame-benchmarking/std -> frame-system/std -> frame-support/wrong -> frame-support/wrong2 -> frame-support/try-runtime

Only the first path is shown in case there are multiple.

Example - Dependency tracing

Recently there was a build error in the Substrate master CI which was caused by a downstream dependency snow. To investigate this, it is useful to see how Substrate depends on it.

Let's find out how node-cli depends on snow (example on commit dd6aedee3b8d5):

zepter trace node-cli snow

It reports that snow is pulled in from libp2p - good to know. In this case, all paths are displayed.

node-cli -> try-runtime-cli -> substrate-rpc-client -> sc-rpc-api -> sc-chain-spec -> sc-telemetry -> libp2p -> libp2p-webrtc -> libp2p-noise -> snow

Config Files

⚠️ the syntax for workflows is highly experimental and bound to change.

The first step is that Zepter checks that it is executed in a rust workspace. Otherwise it fails directly. Then a workflow file is located as follows:

• $WORKSPACE/zepter.yaml
• $WORKSPACE/.zepter.yaml
• $WORKSPACE/.cargo/zepter.yaml
• $WORKSPACE/.cargo/.zepter.yaml
• $WORKSPACE/.config/zepter.yaml
• $WORKSPACE/.config/.zepter.yaml

It uses the first file that is found and errors if none is found. Currently it not possible to overwrite the config in a sub-folder.

Workflows

[!NOTE] A production example can be found in the Polkadot-SDK or in the presets.

It is possible to aggregate the long commands into workflows instead of typing them each time. Zepter tries to locate a config file and run the default workflow when it is bare invoked without any arguments.
Alternately, it is possible to use zepter run default, or any other workflow name.

Config files can contain workflows like this:

workflows:
  default:
    - [ 'propagate-features', ... ]
    - ...

It is also possible to extend previous steps:

workflows:
  check:
    - ...
  default:
    - [ $check.0, '--fix' ]
    - ...

CI Usage

Zepter is currently being used in the Polkadot-SDK CI to spot missing features.
When these two experiments proove the usefulness and reliability of Zepter for CI application, then a more streamlined process will be introduced (possibly in the form of CI actions).

Testing

Unit tests: cargo test UI and downstream integration tests: cargo test -- --ignored

Environment overwrites exist for the UI tests to:

• OVERWRITE: Update the UI diff locks.
• UI_FILTER: Regex to selectively run UI test.
• KEEP_GOING: Print FAILED but don't abort on the first failed UI test.

Planned Features

• Add feature information to the enabled deps
• Optimize shortest_path function
• Add support for config files
• Feature sorting and deduplication