[Draft] Icechunk CLI Design Document

.gitignore

@@ -12,3 +12,4 @@
 /devel
 
 .ipynb_checkpoints
+.vscode

design-docs/008-command-line-interface.md

@@ -0,0 +1,143 @@
+# Icechunk Command Line Interface
+
+This document outlines the design of the Icechunk command line interface.
+
+## Functionality
+
+Here is a list of tasks a user might want to do with Icechunk:
+
+- List my repositories
+- List a history of a repo
+- List branches in a repo
+- List tags in a repo
+- Create a new repository
+- Check configuration
+- Diff between two commits
+- Invoke administrative tasks (garbage collection, compaction, etc)
+
+## Interface
+
+General command structure
+
+```bash
+icechunk <object> <action> <args>
+```
+
+Examples
+
+```bash
+icechunk repo list
+
+icechunk repo create <repo>
+icechunk repo info <repo>
+icechunk repo tree <repo>
+icechunk repo delete <repo>
+
+icechunk branch list <repo>
+icechunk branch create <repo> <branch_name>
+icechunk snapshot list <repo>
+icechunk snapshot diff <repo> <snapshot_id_1> <snapshot_id_2>
+icechunk ref list <repo>
+
+icechunk config init   # init: interactive setup
+icechunk config list 
+icechunk config get <key>
+icechunk config set <key> <value>
+
+```
+
+### Git-like interface
+
+Alternative would be a more git-like structure (`git diff`, `git show`, ..).
+
+The git interface is familiar, but
+
+- The differences between git and Icechunk can be deceptive to new users
+- The git interface is (arguably) not very user-friendly if you're not familiar with it
+- This structure is more extensible
+  - Example: Docker [adopting](https://www.docker.com/blog/whats-new-in-docker-1-13/) this structure over time (`docker ps` -> `docker container ls`)
+
+## Configuration
+
+Two guiding use-cases
+
+- User just wants to `icechunk repo create s3://bucket/path`, get credentials from environment/aws config, and use default repo settings.
+- User wants to manage multiple repositories stored in different locations, with different credentials and settings.
+
+Following Icechunk's config module, there are four types of information needed to work with a repository:
+
+- Location: `bucket`, `path`
+- Credentials: `access_key_id`, `secret_access_key`, ..
+- Options: `region`, `endpoint_url`, ..
+- Repo configuration: `compression`, `caching`, `virtual_chunk_containers`, ..
+
+There are three ways to provide this information, in the standard order of precedence:
+
+1. Command line arguments
+2. Environment variables
+3. Configuration file
+
+
+### Repositories configuration
+
+The CLI repositories configuration file.
+
+> Note: This configuration could also be used by the library.
+
+A first draft of the structure:
+
+```rust
+use std::collections::HashMap;
+
+use crate::config::{RepositoryConfig, ObjectStoreConfig, Credentials}
+
+pub struct RepoLocation {
+    bucket: String,
+    prefix: String,
+}
+
+pub struct RepositoryDefinition {
+    location: RepoLocation,
+    object_store_config: ObjectStoreConfig,
+    credentials: Credentials,
+    config: RepositoryConfig,
+}
+
+pub struct RepositoryAlias(String);
+
+pub struct Repositories {
+    repos: HashMap<RepositoryAlias, RepositoryDefinition>,
+}
+```
+
+## Python packaging
+
+Following the [Python entrypoint](https://www.maturin.rs/bindings#both-binary-and-library) approach.
+
+- cli implemented in `icechunk/src/cli/`
+- cli exposed to Rust in `icechunk/src/bin/icechunk/`
+- cli exposed to Python through an entrypoint function, exposed in `pyproject.toml`
+
+```ini
+[project.scripts]
+icechunk = "icechunk._icechunk_python:cli_entrypoint"
+```
+
+The disadvantage is that Python users need to call Python to use the CLI, resulting in hundreds of milliseconds of latency.
+
+The user can also install the Rust binary directly through `cargo install`.
+
+## Implementation details
+
+Implemented with
+
+- `clap` for the CLI
+  - `clap_complete` for shell completion
+- `anyhow` for error handling
+- `serde_yaml_ng` for configuration
+
+## Optional features
+
+- Structured output option (e.g. JSON)
+- Short version of the command (e.g. `ic`)
+- Support for tab completion

icechunk/src/config.rs

@@ -203,7 +203,7 @@ impl ManifestConfig {
 
 #[derive(Clone, Debug, Serialize, Deserialize, PartialEq, Eq, Default)]
 pub struct RepositoryConfig {
-    /// Chunks smaller than this will be stored inline in the manifst
+    /// Chunks smaller than this will be stored inline in the manifest
     pub inline_chunk_threshold_bytes: Option<u16>,
     /// Unsafely overwrite refs on write. This is not recommended, users should only use it at their
     /// own risk in object stores for which we don't support write-object-if-not-exists. There is

icechunk/src/storage/mod.rs

@@ -244,7 +244,7 @@ pub trait Storage: fmt::Debug + private::Sealed + Sync + Send {
     ) -> StorageResult<Box<dyn AsyncRead + Unpin + Send>>;
     /// Returns whatever reader is more efficient.
     ///
-    /// For example, if processesed with multiple requests, it will return a synchronous `Buf`
+    /// For example, if processed with multiple requests, it will return a synchronous `Buf`
     /// instance pointing the different parts. If it was executed in a single request, it's more
     /// efficient to return the network `AsyncRead` directly
     async fn fetch_manifest_known_size(