DerGut commented on code in PR #2120: URL: https://github.com/apache/iceberg-rust/pull/2120#discussion_r2905366096
########## crates/iceberg/src/transaction/update_schema.rs: ########## @@ -0,0 +1,1200 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and limitations +// under the License. + +use std::collections::{HashMap, HashSet}; +use std::sync::Arc; + +use async_trait::async_trait; + +use crate::spec::{ + ListType, Literal, MapType, NestedField, NestedFieldRef, Schema, StructType, Type, +}; +use crate::table::Table; +use crate::transaction::action::{ActionCommit, TransactionAction}; +use crate::{Error, ErrorKind, Result, TableRequirement, TableUpdate}; + +/// Sentinel parent ID representing the table root (top-level columns). +const TABLE_ROOT_ID: i32 = -1; + +/// A pending column addition, recording the parent path and the field to add. +struct PendingAdd { + /// `None` means a root-level addition; `Some("person")` or `Some("person.address")` + /// identifies the nested struct to add the column to. + parent: Option<String>, + /// The field to add. Uses placeholder ID `0` which is auto-assigned at commit time. + field: NestedFieldRef, +} + +/// Schema evolution API modeled after the Java `SchemaUpdate` implementation. +/// +/// This action accumulates schema modifications (column additions and deletions) +/// via builder methods. At commit time, it validates all operations against the +/// current table schema, auto-assigns field IDs from `table.metadata().last_column_id()`, +/// builds a new schema, and emits `AddSchema` + `SetCurrentSchema` updates with a +/// `CurrentSchemaIdMatch` requirement. +/// +/// # Example +/// +/// ```ignore +/// let tx = Transaction::new(&table); +/// let action = tx.update_schema() +/// .add_column("new_col", Type::Primitive(PrimitiveType::Int)) +/// .add_column_to("person", "email", Type::Primitive(PrimitiveType::String)) +/// .delete_column("old_col"); +/// let tx = action.apply(tx).unwrap(); +/// let table = tx.commit(&catalog).await.unwrap(); +/// ``` +pub struct UpdateSchemaAction { + additions: Vec<PendingAdd>, + deletes: Vec<String>, + auto_assign_ids: bool, +} + +impl UpdateSchemaAction { + /// Creates a new empty `UpdateSchemaAction`. + pub(crate) fn new() -> Self { + Self { + additions: Vec::new(), + deletes: Vec::new(), + auto_assign_ids: true, + } + } + + // --- Root-level additions --- + + /// Add a `NestedFieldRef` column to the table root. + pub fn add_field(self, field: NestedFieldRef) -> Self { + self.add_field_internal(None, field) + } + + /// Add an optional column to the table root. + /// + /// The field ID is a placeholder (`0`) and will be auto-assigned at commit time. + pub fn add_column(self, name: impl ToString, field_type: Type) -> Self { + self.add_field(Arc::new(NestedField::optional(0, name, field_type))) + } + + /// Add an optional column with a doc string to the table root. + /// + /// The field ID is a placeholder (`0`) and will be auto-assigned at commit time. + pub fn add_column_with_doc( + self, + name: impl ToString, + field_type: Type, + doc: impl ToString, + ) -> Self { + self.add_field(Arc::new( + NestedField::optional(0, name, field_type).with_doc(doc), + )) + } + + /// Add a required column to the table root. + /// + /// An `initial_default` value is required per the Iceberg spec: it is used to populate + /// this field for all records that were written before the field was added. + /// The field ID is a placeholder (`0`) and will be auto-assigned at commit time. + pub fn add_required_column( + self, + name: impl ToString, + field_type: Type, + initial_default: Literal, + ) -> Self { + self.add_field(Arc::new( + NestedField::required(0, name, field_type) + .with_initial_default(initial_default.clone()) + .with_write_default(initial_default), + )) + } + + /// Add a required column with a doc string to the table root. + /// + /// An `initial_default` value is required per the Iceberg spec: it is used to populate + /// this field for all records that were written before the field was added. + /// The field ID is a placeholder (`0`) and will be auto-assigned at commit time. + pub fn add_required_column_with_doc( + self, + name: impl ToString, + field_type: Type, + initial_default: Literal, + doc: impl ToString, + ) -> Self { + self.add_field(Arc::new( + NestedField::required(0, name, field_type) + .with_initial_default(initial_default.clone()) + .with_write_default(initial_default) + .with_doc(doc), + )) + } + + // --- Nested additions --- + + /// Add a `NestedFieldRef` column under a parent struct identified by name. + /// + /// If the parent is a map, the column is added to the map value's struct. + /// If the parent is a list, the column is added to the list element's struct. + pub fn add_field_to(self, parent: impl ToString, field: NestedFieldRef) -> Self { + self.add_field_internal(Some(parent.to_string()), field) + } + + /// Add an optional column under a parent struct identified by name. + /// + /// The `parent` can be a dotted path (e.g. `"person"` or `"person.address"`). + /// If the parent is a map, the column is added to the map value's struct. + /// If the parent is a list, the column is added to the list element's struct. + /// The field ID is a placeholder (`0`) and will be auto-assigned at commit time. + pub fn add_column_to( + self, + parent: impl ToString, + name: impl ToString, + field_type: Type, + ) -> Self { + self.add_field_to(parent, Arc::new(NestedField::optional(0, name, field_type))) + } + + /// Add an optional column with a doc string under a parent struct. + /// + /// See [`add_column_to`](Self::add_column_to) for parent path details. + pub fn add_column_to_with_doc( + self, + parent: impl ToString, + name: impl ToString, + field_type: Type, + doc: impl ToString, + ) -> Self { + self.add_field_to( + parent, + Arc::new(NestedField::optional(0, name, field_type).with_doc(doc)), + ) + } + + /// Add a required column under a parent struct. + /// + /// See [`add_column_to`](Self::add_column_to) for parent path details. + /// An `initial_default` value is required per the Iceberg spec. + pub fn add_required_column_to( + self, + parent: impl ToString, + name: impl ToString, + field_type: Type, + initial_default: Literal, + ) -> Self { + self.add_field_to( + parent, + Arc::new( + NestedField::required(0, name, field_type) + .with_initial_default(initial_default.clone()) + .with_write_default(initial_default), + ), + ) + } + + /// Add a required column with a doc string under a parent struct. + /// + /// See [`add_column_to`](Self::add_column_to) for parent path details. + /// An `initial_default` value is required per the Iceberg spec. + pub fn add_required_column_to_with_doc( + self, + parent: impl ToString, + name: impl ToString, + field_type: Type, + initial_default: Literal, + doc: impl ToString, + ) -> Self { + self.add_field_to( + parent, + Arc::new( + NestedField::required(0, name, field_type) + .with_initial_default(initial_default.clone()) + .with_write_default(initial_default) + .with_doc(doc), + ), + ) + } + + // --- Other builder methods --- + + /// Record a column deletion by name. + /// + /// At commit time, the column must exist in the current schema. + pub fn delete_column(mut self, name: impl ToString) -> Self { + self.deletes.push(name.to_string()); + self + } + + /// Disable automatic field ID assignment. When disabled, the placeholder IDs + /// provided in builder methods are used as-is. + pub fn disable_id_auto_assignment(mut self) -> Self { + self.auto_assign_ids = false; + self + } + + // --- Internal helpers --- + + fn add_field_internal(mut self, parent: Option<String>, field: NestedFieldRef) -> Self { + self.additions.push(PendingAdd { parent, field }); + self + } +} + +// --------------------------------------------------------------------------- +// ID assignment helpers +// --------------------------------------------------------------------------- + +/// Recursively assign fresh field IDs to a `NestedField` and all its nested sub-fields. +/// +/// This follows the same recursive pattern as `ReassignFieldIds::reassign_ids_visit_type` +/// from `crate::spec::schema::id_reassigner`, but operates on new fields with placeholder +/// IDs rather than reassigning an existing schema. `ReassignFieldIds` cannot be used +/// directly here because it rejects duplicate old IDs (all new fields share placeholder +/// ID `0`). +fn assign_fresh_ids(field: &NestedField, next_id: &mut i32) -> NestedFieldRef { + *next_id += 1; + let new_id = *next_id; + let new_type = assign_fresh_ids_to_type(&field.field_type, next_id); + + Arc::new(NestedField { + id: new_id, + name: field.name.clone(), + required: field.required, + field_type: Box::new(new_type), + doc: field.doc.clone(), + initial_default: field.initial_default.clone(), + write_default: field.write_default.clone(), + }) +} + +/// Recursively assign fresh field IDs to all nested fields within a `Type`. +fn assign_fresh_ids_to_type(field_type: &Type, next_id: &mut i32) -> Type { + match field_type { + Type::Primitive(_) => field_type.clone(), + Type::Struct(struct_type) => { + let new_fields: Vec<NestedFieldRef> = struct_type + .fields() + .iter() + .map(|f| assign_fresh_ids(f, next_id)) + .collect(); + Type::Struct(StructType::new(new_fields)) + } + Type::List(list_type) => { + let new_element = assign_fresh_ids(&list_type.element_field, next_id); + Type::List(ListType { + element_field: new_element, + }) + } + Type::Map(map_type) => { + let new_key = assign_fresh_ids(&map_type.key_field, next_id); + let new_value = assign_fresh_ids(&map_type.value_field, next_id); + Type::Map(MapType { + key_field: new_key, + value_field: new_value, + }) + } + } +} + +// --------------------------------------------------------------------------- +// Parent path resolution +// --------------------------------------------------------------------------- + +/// Resolve a parent path to the target struct's parent field ID and a reference +/// to its `StructType`. +/// +/// If the parent is a map, navigates to the value field. If a list, navigates to +/// the element field. The final target must be a struct type. +fn resolve_parent_target<'a>( + base_schema: &'a Schema, + parent: &str, +) -> Result<(i32, &'a StructType)> { + base_schema + .field_by_name(parent) + .ok_or_else(|| { + Error::new( + ErrorKind::PreconditionFailed, + format!("Cannot add column: parent '{parent}' not found"), + ) + }) + .and_then(|parent_field| match parent_field.field_type.as_ref() { + Type::Struct(s) => Ok((parent_field.id, s)), + Type::Map(m) => match m.value_field.field_type.as_ref() { + Type::Struct(s) => Ok((m.value_field.id, s)), + _ => Err(Error::new( + ErrorKind::PreconditionFailed, + format!("Cannot add column: map value of '{parent}' is not a struct"), + )), + }, + Type::List(l) => match l.element_field.field_type.as_ref() { + Type::Struct(s) => Ok((l.element_field.id, s)), + _ => Err(Error::new( + ErrorKind::PreconditionFailed, + format!("Cannot add column: list element of '{parent}' is not a struct"), + )), + }, + _ => Err(Error::new( + ErrorKind::PreconditionFailed, + format!("Cannot add column: parent '{parent}' is not a struct, map, or list"), + )), + }) +} + +// --------------------------------------------------------------------------- +// Schema tree rebuild +// --------------------------------------------------------------------------- + +/// Rebuild a slice of fields, applying deletions and additions at every level, +/// plus any root-level additions keyed by `TABLE_ROOT_ID`. +fn rebuild_fields( + fields: &[NestedFieldRef], + adds: &HashMap<i32, Vec<NestedFieldRef>>, + delete_ids: &HashSet<i32>, + root_id: i32, +) -> Vec<NestedFieldRef> { + fields + .iter() + .filter(|f| !delete_ids.contains(&f.id)) + .map(|f| rebuild_field(f, adds, delete_ids)) + .chain(adds.get(&root_id).into_iter().flatten().cloned()) + .collect() +} + +/// Recursively rebuild a single field. If the field (or any descendant) is a struct +/// that has pending additions, those additions are appended to the struct's fields. +/// Fields whose IDs appear in `delete_ids` are filtered out at every struct level. +fn rebuild_field( + field: &NestedFieldRef, + adds: &HashMap<i32, Vec<NestedFieldRef>>, + delete_ids: &HashSet<i32>, +) -> NestedFieldRef { + match field.field_type.as_ref() { + Type::Primitive(_) => field.clone(), + Type::Struct(s) => { + let new_fields = rebuild_fields(s.fields(), adds, delete_ids, field.id); + Arc::new(NestedField { + id: field.id, + name: field.name.clone(), + required: field.required, + field_type: Box::new(Type::Struct(StructType::new(new_fields))), + doc: field.doc.clone(), + initial_default: field.initial_default.clone(), + write_default: field.write_default.clone(), + }) + } + Type::List(l) => { + let new_element = rebuild_field(&l.element_field, adds, delete_ids); + Arc::new(NestedField { + id: field.id, + name: field.name.clone(), + required: field.required, + field_type: Box::new(Type::List(ListType { + element_field: new_element, + })), + doc: field.doc.clone(), + initial_default: field.initial_default.clone(), + write_default: field.write_default.clone(), + }) + } + Type::Map(m) => { Review Comment: The [Java](https://github.com/apache/iceberg/blob/b310cba27cefdc9c3ac0bd620dc36d1d1879ecb9/core/src/main/java/org/apache/iceberg/SchemaUpdate.java#L716-L717) and [pyiceberg](https://github.com/apache/iceberg-python/blob/main/pyiceberg/table/update/schema.py#L817-L818) implementations explicitly forbid deletion of map keys+values. IIUC this code just silently ignores them. TBH I don't think this is very clear from [the spec](https://iceberg.apache.org/spec/#schema-evolution)... apparently > **Any struct**, including a top-level schema, can evolve through deleting fields [...] means _any struct_ but no maps. So an explicit error is probably appropriate (same with list type btw) -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
