mz_adapter/catalog/

migrate.rs

// Copyright Materialize, Inc. and contributors. All rights reserved.
//
// Use of this software is governed by the Business Source License
// included in the LICENSE file.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0.

use std::collections::{BTreeMap, BTreeSet};
use std::time::Duration;

use base64::prelude::*;
use maplit::btreeset;
use mz_catalog::builtin::{BUILTINS, BuiltinTable};
use mz_catalog::durable::objects::{SystemObjectDescription, SystemObjectMapping};
use mz_catalog::durable::{MOCK_AUTHENTICATION_NONCE_KEY, Transaction};
use mz_catalog::memory::objects::{StateUpdate, StateUpdateKind};
use mz_ore::collections::CollectionExt;
use mz_ore::now::NowFn;
use mz_persist_types::ShardId;
use mz_proto::RustType;
use mz_repr::{CatalogItemId, Diff, Timestamp};
use mz_sql::ast::display::AstDisplay;
use mz_sql::ast::{
    CreateSinkOptionName, CreateViewStatement, CteBlock, DeferredItemName, IfExistsBehavior, Query,
    SetExpr, SqlServerConfigOptionName, ViewDefinition,
};
use mz_sql::catalog::{CatalogItemType, SessionCatalog};
use mz_sql::names::{FullItemName, QualifiedItemName};
use mz_sql::normalize;
use mz_sql::session::vars::{FORCE_SOURCE_TABLE_SYNTAX, Var, VarInput};
use mz_sql_parser::ast::{Raw, Statement};
use mz_storage_client::controller::StorageTxn;
use mz_storage_types::sources::SourceExportStatementDetails;
use mz_storage_types::sources::load_generator::LoadGeneratorOutput;
use prost::Message;
use semver::Version;
use tracing::info;
use uuid::Uuid;

// DO NOT add any more imports from `crate` outside of `crate::catalog`.
use crate::catalog::open::into_consolidatable_updates_startup;
use crate::catalog::state::LocalExpressionCache;
use crate::catalog::{BuiltinTableUpdate, CatalogState, ConnCatalog};
use crate::coord::catalog_implications::parsed_state_updates::ParsedStateUpdate;

/// Catalog key of the `migration_version` setting.
///
/// The `migration_version` tracks the version of the binary that last successfully performed and
/// committed all the catalog migrations (including builtin schema migrations). It can be used by
/// migration logic to identify the source version from which to migrate.
///
/// Note that the durable catalog also knows a `catalog_content_version`. That doesn't work for
/// this purpose as it is already bumped to the current binary version when the catalog is opened
/// in writable mode, before any migrations have run.
const MIGRATION_VERSION_KEY: &str = "migration_version";

pub(crate) fn get_migration_version(txn: &Transaction<'_>) -> Option<Version> {
    txn.get_setting(MIGRATION_VERSION_KEY.into())
        .map(|s| s.parse().expect("valid migration version"))
}

pub(crate) fn set_migration_version(
    txn: &mut Transaction<'_>,
    version: Version,
) -> Result<(), mz_catalog::durable::CatalogError> {
    txn.set_setting(MIGRATION_VERSION_KEY.into(), Some(version.to_string()))
}

fn rewrite_ast_items<F>(tx: &mut Transaction<'_>, mut f: F) -> Result<(), anyhow::Error>
where
    F: for<'a> FnMut(
        &'a mut Transaction<'_>,
        CatalogItemId,
        &'a mut Statement<Raw>,
    ) -> Result<(), anyhow::Error>,
{
    let mut updated_items = BTreeMap::new();

    for mut item in tx.get_items() {
        let mut stmt = mz_sql::parse::parse(&item.create_sql)?.into_element().ast;
        f(tx, item.id, &mut stmt)?;

        item.create_sql = stmt.to_ast_string_stable();

        updated_items.insert(item.id, item);
    }
    tx.update_items(updated_items)?;
    Ok(())
}

fn rewrite_items<F>(
    tx: &mut Transaction<'_>,
    cat: &ConnCatalog<'_>,
    mut f: F,
) -> Result<(), anyhow::Error>
where
    F: for<'a> FnMut(
        &'a mut Transaction<'_>,
        &'a &ConnCatalog<'_>,
        CatalogItemId,
        &'a mut Statement<Raw>,
    ) -> Result<(), anyhow::Error>,
{
    let mut updated_items = BTreeMap::new();
    let items = tx.get_items();
    for mut item in items {
        let mut stmt = mz_sql::parse::parse(&item.create_sql)?.into_element().ast;

        f(tx, &cat, item.id, &mut stmt)?;

        item.create_sql = stmt.to_ast_string_stable();

        updated_items.insert(item.id, item);
    }
    tx.update_items(updated_items)?;
    Ok(())
}

pub(crate) struct MigrateResult {
    pub(crate) builtin_table_updates: Vec<BuiltinTableUpdate<&'static BuiltinTable>>,
    pub(crate) catalog_updates: Vec<ParsedStateUpdate>,
    pub(crate) post_item_updates: Vec<(StateUpdateKind, Timestamp, Diff)>,
}

/// Migrates all user items and loads them into `state`.
///
/// Returns the builtin updates corresponding to all user items.
pub(crate) async fn migrate(
    state: &mut CatalogState,
    tx: &mut Transaction<'_>,
    local_expr_cache: &mut LocalExpressionCache,
    item_updates: Vec<StateUpdate>,
    _now: NowFn,
    _boot_ts: Timestamp,
) -> Result<MigrateResult, anyhow::Error> {
    let catalog_version = get_migration_version(tx).unwrap_or(Version::new(0, 0, 0));

    info!(
        "migrating statements from catalog version {:?}",
        catalog_version
    );

    rewrite_ast_items(tx, |tx, _id, stmt| {
        // Add per-item AST migrations below.
        //
        // Each migration should be a function that takes `stmt` (the AST
        // representing the creation SQL for the item) as input. Any
        // mutations to `stmt` will be staged for commit to the catalog.
        //
        // Migration functions may also take `tx` as input to stage
        // arbitrary changes to the catalog.
        ast_rewrite_create_sink_partition_strategy(stmt)?;
        ast_rewrite_sql_server_constraints(stmt)?;
        ast_rewrite_add_missing_index_ids(tx, stmt)?;
        ast_rewrite_add_missing_doc_on_ids(tx, stmt)?;
        ast_rewrite_kafka_metadata_refresh_intervals(stmt)?;
        ast_rewrite_small_commit_intervals(stmt)?;
        ast_rewrite_strip_builtin_version_pins(stmt)?;
        ast_rewrite_strip_unsupported_column_collations(stmt)?;
        Ok(())
    })?;

    // Load items into catalog. We make sure to consolidate the old updates with the new updates to
    // avoid trying to apply unmigrated items.
    let commit_ts = tx.upper();
    let mut item_updates = into_consolidatable_updates_startup(item_updates, commit_ts);
    let op_item_updates = tx.get_and_commit_op_updates();
    let op_item_updates = into_consolidatable_updates_startup(op_item_updates, commit_ts);
    item_updates.extend(op_item_updates);
    differential_dataflow::consolidation::consolidate_updates(&mut item_updates);

    // Since some migrations might introduce non-item 'post-item' updates, we sequester those
    // so they can be applied with other post-item updates after migrations to avoid
    // accumulating negative diffs.
    let (post_item_updates, item_updates): (Vec<_>, Vec<_>) = item_updates
        .into_iter()
        // The only post-item update kind we currently generate is to
        // update storage collection metadata.
        .partition(|(kind, _, _)| matches!(kind, StateUpdateKind::StorageCollectionMetadata(_)));

    let item_updates = item_updates
        .into_iter()
        .map(|(kind, ts, diff)| StateUpdate {
            kind,
            ts,
            diff: diff.try_into().expect("valid diff"),
        })
        .collect();

    let force_source_table_syntax = state.system_config().force_source_table_syntax();
    // When this flag is set the legacy syntax is denied. But here we are about to perform a
    // migration which requires that we parse the current catalog state. To proceed we temporarily disable
    // the flag and then reset it after migrations are done.
    if force_source_table_syntax {
        state
            .system_config_mut()
            .set(FORCE_SOURCE_TABLE_SYNTAX.name(), VarInput::Flat("off"))
            .expect("known parameter");
    }

    let (mut ast_builtin_table_updates, mut ast_catalog_updates) =
        state.apply_updates(item_updates, local_expr_cache).await;

    info!("migrating from catalog version {:?}", catalog_version);

    let conn_cat = state.for_system_session();

    // Special block for `ast_rewrite_sources_to_tables` migration
    // since it requires a feature flag needs to update multiple AST items at once.
    if force_source_table_syntax {
        rewrite_sources_to_tables(tx, &conn_cat)?;
    }

    rewrite_items(tx, &conn_cat, |_tx, _conn_cat, _id, _stmt| {
        let _catalog_version = catalog_version.clone();
        // Add per-item, post-planning AST migrations below. Most
        // migrations should be in the above `rewrite_ast_items` block.
        //
        // Each migration should be a function that takes `item` (the AST
        // representing the creation SQL for the item) as input. Any
        // mutations to `item` will be staged for commit to the catalog.
        //
        // Be careful if you reference `conn_cat`. Doing so is *weird*,
        // as you'll be rewriting the catalog while looking at it. If
        // possible, make your migration independent of `conn_cat`, and only
        // consider a single item at a time.
        //
        // Migration functions may also take `tx` as input to stage
        // arbitrary changes to the catalog.
        Ok(())
    })?;

    if force_source_table_syntax {
        state
            .system_config_mut()
            .set(FORCE_SOURCE_TABLE_SYNTAX.name(), VarInput::Flat("on"))
            .expect("known parameter");
    }

    // Add whole-catalog migrations below.
    //
    // Each migration should be a function that takes `tx` and `conn_cat` as
    // input and stages arbitrary transformations to the catalog on `tx`.

    let op_item_updates = tx.get_and_commit_op_updates();
    let (item_builtin_table_updates, item_catalog_updates) =
        state.apply_updates(op_item_updates, local_expr_cache).await;

    ast_builtin_table_updates.extend(item_builtin_table_updates);
    ast_catalog_updates.extend(item_catalog_updates);

    info!(
        "migration from catalog version {:?} complete",
        catalog_version
    );

    Ok(MigrateResult {
        builtin_table_updates: ast_builtin_table_updates,
        catalog_updates: ast_catalog_updates,
        post_item_updates,
    })
}

// Add new migrations below their appropriate heading, and precede them with a
// short summary of the migration's purpose and optional additional commentary
// about safety or approach.
//
// The convention is to name the migration function using snake case:
// > <category>_<description>_<version>
//
// Please include the adapter team on any code reviews that add or edit
// migrations.

/// Migrates all sources to use the new sources as tables model
///
/// Suppose we have an old-style source named `source_name` with global id `source_id`. The source
/// will also have an associated progress source named `progress_name` (which is almost always
/// `source_name` + "_progress") with global id `progress_id`.
///
/// We have two constraints to satisfy. The migration:
///   1. should not change the schema of a global id *if that global id maps to a
///      durable collection*. The reason for this constraint is that when a durable collection (i.e
///      backed by a persist shard) is opened persist will verify that the schema is the expected
///      one. If we change the Create SQL of a global id to a non-durable definition (e.g a view)
///      then we are free to also change the schema.
///   2. should make it such that the SQL object that is constructed with a new-style `CREATE
///      SOURCE` statement contains the progress data and all other objects related to the
///      old-style source depend on that object.
///
/// With these constraints we consider two cases.
///
/// ## Case 1: A multi-output source
///
/// Multi-output sources have a dummy output as the contents of `source_name` that is useless. So
/// we re-purpose that name to be the `CREATE SOURCE` statement and make `progress_name` be a view
/// of `source_name`. Since the main source is a durable object we must move `source_name` and the
/// corresponding new-style `CREATE SOURCE` statement under `progress_id`. Then `progress_name` can
/// move to `source_id` and since it becomes a view we are free to change its schema.
///
/// Visually, we are changing this mapping:
///
/// |  Global ID  |  SQL Name     | Create SQL                 | Schema   | Durable |
/// +-------------+---------------+----------------------------+----------+---------|
/// | source_id   | source_name   | CREATE SOURCE (old-style)  | empty    | yes     |
/// | progress_id | progress_name | CREATE SUBSOURCE .."       | progress | yes     |
///
/// to this mapping:
///
/// |  Global ID  |  SQL Name     | Create SQL                | Schema        | Durable |
/// +-------------+---------------+---------------------------+---------------+---------+
/// | source_id   | progress_name | CREATE VIEW               | progress data | no      |
/// | progress_id | source_name   | CREATE SOURCE (new-style) | progress data | yes     |
///
/// ## Case 2: A single-output source
///
/// Single-output sources have data as the contents of `source_name` and so we can't repurpose that
/// name to be the `CREATE SOURCE` statement. Here we leave everything intact except for the
/// Create SQL of each object. Namely, the old-style `CREATE SOURCE` statement becomes a `CREATE
/// TABLE FROM SOURCE` and the old-style `CREATE SUBSOURCE .. PROGRESS` becomes a new-style `CREATE
/// SOURCE` statement.
///
/// Visually, we are changing this mapping:
///
/// |  Global ID  |  SQL Name     | Create SQL                 | Schema      | Durable |
/// +-------------+---------------+----------------------------+-------------+---------|
/// | source_id   | source_name   | CREATE SOURCE (old-style)  | source data | yes     |
/// | progress_id | progress_name | CREATE SUBSOURCE .."       | progress    | yes     |
///
/// to this mapping:
///
/// |  Global ID  |  SQL Name     | Create SQL                 | Schema      | Durable |
/// +-------------+---------------+----------------------------+-------------+---------|
/// | source_id   | source_name   | CREATE TABLE FROM SOURCE   | source data | yes     |
/// | progress_id | progress_name | CREATE SOURCE (new-style)  | progress    | yes     |
///
/// ## Subsource migration
///
/// After the migration goes over all the `CREATE SOURCE` statements it then transforms each
/// non-progress `CREATE SUBSOURCE` statement to be a `CREATE TABLE FROM SOURCE` statement that
/// points to the original `source_name` but with the altered global id (which is now
/// `progress_id`).
fn rewrite_sources_to_tables(
    tx: &mut Transaction<'_>,
    catalog: &ConnCatalog<'_>,
) -> Result<(), anyhow::Error> {
    use mz_sql::ast::{
        CreateSourceConnection, CreateSourceStatement, CreateSubsourceOptionName,
        CreateSubsourceStatement, CreateTableFromSourceStatement, Ident,
        KafkaSourceConfigOptionName, LoadGenerator, MySqlConfigOptionName, PgConfigOptionName,
        RawItemName, TableFromSourceColumns, TableFromSourceOption, TableFromSourceOptionName,
        UnresolvedItemName, Value, WithOptionValue,
    };

    let mut updated_items = BTreeMap::new();

    let mut sources = vec![];
    let mut subsources = vec![];

    for item in tx.get_items() {
        let stmt = mz_sql::parse::parse(&item.create_sql)?.into_element().ast;
        match stmt {
            Statement::CreateSubsource(stmt) => subsources.push((item, stmt)),
            Statement::CreateSource(stmt) => sources.push((item, stmt)),
            _ => {}
        }
    }

    let mut pending_progress_items = BTreeMap::new();
    let mut migrated_source_ids = BTreeMap::new();
    // We first go over the sources, which depending on the kind determine what happens with the
    // progress statements.
    for (mut source_item, source_stmt) in sources {
        let CreateSourceStatement {
            name,
            in_cluster,
            col_names,
            mut connection,
            include_metadata,
            format,
            envelope,
            if_not_exists,
            key_constraint,
            with_options,
            external_references,
            progress_subsource,
        } = source_stmt;

        let (progress_name, progress_item) = match progress_subsource {
            Some(DeferredItemName::Named(RawItemName::Name(name))) => {
                let partial_name = normalize::unresolved_item_name(name.clone())?;
                (name, catalog.resolve_item(&partial_name)?)
            }
            Some(DeferredItemName::Named(RawItemName::Id(id, name, _))) => {
                let gid = id.parse()?;
                (name, catalog.get_item(&gid))
            }
            Some(DeferredItemName::Deferred(_)) => {
                unreachable!("invalid progress subsource")
            }
            None => {
                info!("migrate: skipping already migrated source: {name}");
                continue;
            }
        };
        let raw_progress_name =
            RawItemName::Id(progress_item.id().to_string(), progress_name.clone(), None);

        // We need to jump through some hoops to get to the raw item name of the source
        let catalog_item = catalog.get_item(&source_item.id);
        let source_name: &QualifiedItemName = catalog_item.name();
        let full_source_name: FullItemName = catalog.resolve_full_name(source_name);
        let source_name: UnresolvedItemName = normalize::unresolve(full_source_name.clone());

        // First, strip the connection options that we no longer need
        match &mut connection {
            CreateSourceConnection::Postgres { options, .. } => {
                options.retain(|o| match o.name {
                    PgConfigOptionName::Details
                    | PgConfigOptionName::Publication
                    | PgConfigOptionName::Failover => true,
                    PgConfigOptionName::TextColumns | PgConfigOptionName::ExcludeColumns => false,
                });
            }
            CreateSourceConnection::SqlServer { options, .. } => {
                options.retain(|o| match o.name {
                    SqlServerConfigOptionName::Details => true,
                    SqlServerConfigOptionName::TextColumns
                    | SqlServerConfigOptionName::ExcludeColumns => false,
                });
            }
            CreateSourceConnection::MySql { options, .. } => {
                options.retain(|o| match o.name {
                    MySqlConfigOptionName::Details => true,
                    MySqlConfigOptionName::TextColumns | MySqlConfigOptionName::ExcludeColumns => {
                        false
                    }
                });
            }
            CreateSourceConnection::Kafka { .. } | CreateSourceConnection::LoadGenerator { .. } => {
            }
        }

        // Then, figure out the new statements for the progress and source.
        let (new_progress_name, new_progress_stmt, new_source_name, new_source_stmt) =
            match connection {
                connection @ (CreateSourceConnection::Postgres { .. }
                | CreateSourceConnection::MySql { .. }
                | CreateSourceConnection::SqlServer { .. }
                | CreateSourceConnection::LoadGenerator {
                    generator:
                        LoadGenerator::Tpch | LoadGenerator::Auction | LoadGenerator::Marketing,
                    ..
                }) => {
                    // Assert the expected state of the source
                    assert_eq!(col_names, &[]);
                    assert_eq!(key_constraint, None);
                    assert_eq!(format, None);
                    assert_eq!(envelope, None);
                    assert_eq!(include_metadata, &[]);
                    assert_eq!(external_references, None);

                    // This is a dummy replacement statement for the source object of multi-output
                    // sources. It is describing the query `TABLE source_name`. This ensures that
                    // whoever was used to run select queries against the `source_name` + "_progress"
                    // object still gets the same data after the migration. This switch does
                    // changes the schema of the object with `source_item.id` but because we're turning
                    // it into a view, which is not durable, it's ok. We'll never open a persist shard
                    // for this global id anymore.
                    let dummy_source_stmt = Statement::CreateView(CreateViewStatement {
                        if_exists: IfExistsBehavior::Error,
                        temporary: false,
                        definition: ViewDefinition {
                            name: progress_name,
                            columns: vec![],
                            query: Query {
                                ctes: CteBlock::Simple(vec![]),
                                body: SetExpr::Table(RawItemName::Id(
                                    progress_item.id().to_string(),
                                    source_name.clone(),
                                    None,
                                )),
                                order_by: vec![],
                                limit: None,
                                offset: None,
                            },
                        },
                    });

                    let new_progress_stmt = CreateSourceStatement {
                        name: source_name.clone(),
                        in_cluster,
                        col_names: vec![],
                        connection,
                        include_metadata: vec![],
                        format: None,
                        envelope: None,
                        if_not_exists,
                        key_constraint: None,
                        with_options,
                        external_references: None,
                        progress_subsource: None,
                    };

                    migrated_source_ids.insert(source_item.id, progress_item.id());

                    (
                        full_source_name.item,
                        new_progress_stmt,
                        progress_item.name().item.clone(),
                        dummy_source_stmt,
                    )
                }
                CreateSourceConnection::Kafka {
                    options,
                    connection,
                } => {
                    let constraints = if let Some(_key_constraint) = key_constraint {
                        // Primary key not enforced is not enabled for anyone
                        // TODO: remove the feature altogether
                        vec![]
                    } else {
                        vec![]
                    };

                    let columns = if col_names.is_empty() {
                        TableFromSourceColumns::NotSpecified
                    } else {
                        TableFromSourceColumns::Named(col_names)
                    };

                    // All source tables must have a `details` option, which is a serialized proto
                    // describing any source-specific details for this table statement.
                    let details = SourceExportStatementDetails::Kafka {};
                    let table_with_options = vec![TableFromSourceOption {
                        name: TableFromSourceOptionName::Details,
                        value: Some(WithOptionValue::Value(Value::String(hex::encode(
                            details.into_proto().encode_to_vec(),
                        )))),
                    }];
                    // The external reference for a kafka source is the just the topic name
                    let topic_option = options
                        .iter()
                        .find(|o| matches!(o.name, KafkaSourceConfigOptionName::Topic))
                        .expect("kafka sources must have a topic");
                    let topic = match &topic_option.value {
                        Some(WithOptionValue::Value(Value::String(topic))) => topic,
                        _ => unreachable!("topic must be a string"),
                    };
                    let external_reference = UnresolvedItemName::qualified(&[Ident::new(topic)?]);

                    let new_source_stmt =
                        Statement::CreateTableFromSource(CreateTableFromSourceStatement {
                            name: source_name,
                            constraints,
                            columns,
                            if_not_exists,
                            source: raw_progress_name,
                            include_metadata,
                            format,
                            envelope,
                            external_reference: Some(external_reference),
                            with_options: table_with_options,
                        });

                    let new_progress_stmt = CreateSourceStatement {
                        name: progress_name,
                        in_cluster,
                        col_names: vec![],
                        connection: CreateSourceConnection::Kafka {
                            options,
                            connection,
                        },
                        include_metadata: vec![],
                        format: None,
                        envelope: None,
                        if_not_exists,
                        key_constraint: None,
                        with_options,
                        external_references: None,
                        progress_subsource: None,
                    };
                    (
                        progress_item.name().item.clone(),
                        new_progress_stmt,
                        full_source_name.item,
                        new_source_stmt,
                    )
                }
                CreateSourceConnection::LoadGenerator {
                    generator:
                        generator @ (LoadGenerator::Clock
                        | LoadGenerator::Counter
                        | LoadGenerator::Datums
                        | LoadGenerator::KeyValue),
                    options,
                } => {
                    let constraints = if let Some(_key_constraint) = key_constraint {
                        // Should we ignore not enforced primary key constraints here?
                        vec![]
                    } else {
                        vec![]
                    };

                    let columns = if col_names.is_empty() {
                        TableFromSourceColumns::NotSpecified
                    } else {
                        TableFromSourceColumns::Named(col_names)
                    };

                    // All source tables must have a `details` option, which is a serialized proto
                    // describing any source-specific details for this table statement.
                    let details = SourceExportStatementDetails::LoadGenerator {
                        output: LoadGeneratorOutput::Default,
                    };
                    let table_with_options = vec![TableFromSourceOption {
                        name: TableFromSourceOptionName::Details,
                        value: Some(WithOptionValue::Value(Value::String(hex::encode(
                            details.into_proto().encode_to_vec(),
                        )))),
                    }];
                    // Since these load generators are single-output the external reference
                    // uses the schema-name for both namespace and name.
                    let external_reference = FullItemName {
                        database: mz_sql::names::RawDatabaseSpecifier::Name(
                            mz_storage_types::sources::load_generator::LOAD_GENERATOR_DATABASE_NAME
                                .to_owned(),
                        ),
                        schema: generator.schema_name().to_string(),
                        item: generator.schema_name().to_string(),
                    };

                    let new_source_stmt =
                        Statement::CreateTableFromSource(CreateTableFromSourceStatement {
                            name: source_name,
                            constraints,
                            columns,
                            if_not_exists,
                            source: raw_progress_name,
                            include_metadata,
                            format,
                            envelope,
                            external_reference: Some(external_reference.into()),
                            with_options: table_with_options,
                        });

                    let new_progress_stmt = CreateSourceStatement {
                        name: progress_name,
                        in_cluster,
                        col_names: vec![],
                        connection: CreateSourceConnection::LoadGenerator { generator, options },
                        include_metadata: vec![],
                        format: None,
                        envelope: None,
                        if_not_exists,
                        key_constraint: None,
                        with_options,
                        external_references: None,
                        progress_subsource: None,
                    };
                    (
                        progress_item.name().item.clone(),
                        new_progress_stmt,
                        full_source_name.item,
                        new_source_stmt,
                    )
                }
            };

        // The source can be updated right away but the replacement progress statement will
        // be installed in the next loop where we go over subsources.

        info!(
            "migrate: converted source {} to {}",
            source_item.create_sql, new_source_stmt
        );
        source_item.name = new_source_name.clone();
        source_item.create_sql = new_source_stmt.to_ast_string_stable();
        updated_items.insert(source_item.id, source_item);
        pending_progress_items.insert(progress_item.id(), (new_progress_name, new_progress_stmt));
    }

    for (mut item, stmt) in subsources {
        match stmt {
            // Migrate progress statements to the corresponding statement produced from the
            // previous step.
            CreateSubsourceStatement {
                of_source: None, ..
            } => {
                let Some((new_name, new_stmt)) = pending_progress_items.remove(&item.id) else {
                    panic!("encountered orphan progress subsource id: {}", item.id)
                };
                item.name = new_name;
                item.create_sql = new_stmt.to_ast_string_stable();
                updated_items.insert(item.id, item);
            }
            // Migrate each `CREATE SUBSOURCE` statement to an equivalent
            // `CREATE TABLE ... FROM SOURCE` statement.
            CreateSubsourceStatement {
                name,
                columns,
                constraints,
                of_source: Some(raw_source_name),
                if_not_exists,
                mut with_options,
            } => {
                let new_raw_source_name = match raw_source_name {
                    RawItemName::Id(old_id, name, None) => {
                        let old_id: CatalogItemId = old_id.parse().expect("well formed");
                        let new_id = migrated_source_ids[&old_id].clone();
                        RawItemName::Id(new_id.to_string(), name, None)
                    }
                    _ => unreachable!("unexpected source name: {raw_source_name}"),
                };
                // The external reference is a `with_option` on subsource statements but is a
                // separate field on table statements.
                let external_reference = match with_options
                    .iter()
                    .position(|opt| opt.name == CreateSubsourceOptionName::ExternalReference)
                {
                    Some(i) => match with_options.remove(i).value {
                        Some(WithOptionValue::UnresolvedItemName(name)) => name,
                        _ => unreachable!("external reference must be an unresolved item name"),
                    },
                    None => panic!("subsource must have an external reference"),
                };

                let with_options = with_options
                    .into_iter()
                    .map(|option| {
                        match option.name {
                            CreateSubsourceOptionName::Details => TableFromSourceOption {
                                name: TableFromSourceOptionName::Details,
                                // The `details` option on both subsources and tables is identical, using the same
                                // ProtoSourceExportStatementDetails serialized value.
                                value: option.value,
                            },
                            CreateSubsourceOptionName::TextColumns => TableFromSourceOption {
                                name: TableFromSourceOptionName::TextColumns,
                                value: option.value,
                            },
                            CreateSubsourceOptionName::ExcludeColumns => TableFromSourceOption {
                                name: TableFromSourceOptionName::ExcludeColumns,
                                value: option.value,
                            },
                            CreateSubsourceOptionName::RetainHistory => TableFromSourceOption {
                                name: TableFromSourceOptionName::RetainHistory,
                                value: option.value,
                            },
                            CreateSubsourceOptionName::Progress => {
                                panic!("progress option should not exist on this subsource")
                            }
                            CreateSubsourceOptionName::ExternalReference => {
                                unreachable!("This option is handled separately above.")
                            }
                        }
                    })
                    .collect::<Vec<_>>();

                let table = CreateTableFromSourceStatement {
                    name,
                    constraints,
                    columns: TableFromSourceColumns::Defined(columns),
                    if_not_exists,
                    source: new_raw_source_name,
                    external_reference: Some(external_reference),
                    with_options,
                    // Subsources don't have `envelope`, `include_metadata`, or `format` options.
                    envelope: None,
                    include_metadata: vec![],
                    format: None,
                };

                info!(
                    "migrate: converted subsource {} to table {}",
                    item.create_sql, table
                );
                item.create_sql = Statement::CreateTableFromSource(table).to_ast_string_stable();
                updated_items.insert(item.id, item);
            }
        }
    }
    assert!(
        pending_progress_items.is_empty(),
        "unexpected residual progress items: {pending_progress_items:?}"
    );

    tx.update_items(updated_items)?;

    Ok(())
}

// Durable migrations

/// Migrations that run only on the durable catalog before any data is loaded into memory.
pub(crate) fn durable_migrate(
    tx: &mut Transaction,
    _organization_id: Uuid,
    _boot_ts: Timestamp,
) -> Result<(), anyhow::Error> {
    // Migrate the expression cache to a new shard. We're updating the keys to use the explicit
    // binary version instead of the deploy generation.
    const EXPR_CACHE_MIGRATION_KEY: &str = "expr_cache_migration";
    const EXPR_CACHE_MIGRATION_DONE: u64 = 1;
    if tx.get_config(EXPR_CACHE_MIGRATION_KEY.to_string()) != Some(EXPR_CACHE_MIGRATION_DONE) {
        if let Some(shard_id) = tx.get_expression_cache_shard() {
            tx.insert_unfinalized_shards(btreeset! {shard_id})?;
            tx.set_expression_cache_shard(ShardId::new())?;
        }
        tx.set_config(
            EXPR_CACHE_MIGRATION_KEY.to_string(),
            Some(EXPR_CACHE_MIGRATION_DONE),
        )?;
    }

    // Migrate the builtin migration shard to a new shard. We're updating the keys to use the explicit
    // binary version instead of the deploy generation.
    const BUILTIN_MIGRATION_SHARD_MIGRATION_KEY: &str = "migration_shard_migration";
    const BUILTIN_MIGRATION_SHARD_MIGRATION_DONE: u64 = 1;
    if tx.get_config(BUILTIN_MIGRATION_SHARD_MIGRATION_KEY.to_string())
        != Some(BUILTIN_MIGRATION_SHARD_MIGRATION_DONE)
    {
        if let Some(shard_id) = tx.get_builtin_migration_shard() {
            tx.insert_unfinalized_shards(btreeset! {shard_id})?;
            tx.set_builtin_migration_shard(ShardId::new())?;
        }
        tx.set_config(
            BUILTIN_MIGRATION_SHARD_MIGRATION_KEY.to_string(),
            Some(BUILTIN_MIGRATION_SHARD_MIGRATION_DONE),
        )?;
    }

    if tx
        .get_setting(MOCK_AUTHENTICATION_NONCE_KEY.to_string())
        .is_none()
    {
        let mut nonce = [0u8; 24];
        openssl::rand::rand_bytes(&mut nonce).expect("failed to generate nonce");
        let nonce = BASE64_STANDARD.encode(nonce);
        tx.set_setting(MOCK_AUTHENTICATION_NONCE_KEY.to_string(), Some(nonce))?;
    }

    migrate_builtin_tables_to_mvs(tx)?;

    Ok(())
}

/// Update system object mappings for builtins whose type changed from table to materialized view.
///
/// Required for the work of making builtin tables views over `mz_catalog_raw`.
fn migrate_builtin_tables_to_mvs(tx: &mut Transaction) -> Result<(), anyhow::Error> {
    // Collect `(schema, name)` of all builtin MVs.
    let expected_mvs: BTreeSet<_> = BUILTINS::materialized_views()
        .map(|mv| (mv.schema, mv.name))
        .collect();

    // Find persisted mappings for builtin tables that must be migrated.
    let mut to_remove = BTreeSet::new();
    let mut to_add = Vec::new();
    for mapping in tx.get_system_object_mappings() {
        let desc = &mapping.description;
        if desc.object_type != CatalogItemType::Table {
            continue;
        }

        let key = (&*desc.schema_name, &*desc.object_name);
        if expected_mvs.contains(&key) {
            info!(
                "migrate: builtin {}.{} changed type from table to MV",
                desc.schema_name, desc.object_name,
            );
            to_remove.insert(desc.clone());
            to_add.push(SystemObjectMapping {
                description: SystemObjectDescription {
                    schema_name: desc.schema_name.clone(),
                    object_type: CatalogItemType::MaterializedView,
                    object_name: desc.object_name.clone(),
                },
                unique_identifier: mapping.unique_identifier,
            });
        }
    }

    if !to_remove.is_empty() {
        tx.remove_system_object_mappings(to_remove)?;
        tx.set_system_object_mappings(to_add)?;
    }

    Ok(())
}

// Add new migrations below their appropriate heading, and precede them with a
// short summary of the migration's purpose and optional additional commentary
// about safety or approach.
//
// The convention is to name the migration function using snake case:
// > <category>_<description>_<version>
//
// Please include the adapter team on any code reviews that add or edit
// migrations.

// Remove PARTITION STRATEGY from CREATE SINK statements.
fn ast_rewrite_create_sink_partition_strategy(
    stmt: &mut Statement<Raw>,
) -> Result<(), anyhow::Error> {
    let Statement::CreateSink(stmt) = stmt else {
        return Ok(());
    };
    stmt.with_options
        .retain(|op| op.name != CreateSinkOptionName::PartitionStrategy);
    Ok(())
}

// Migrate SQL Server constraint information from the columns to dedicated constraints field.
fn ast_rewrite_sql_server_constraints(stmt: &mut Statement<Raw>) -> Result<(), anyhow::Error> {
    use mz_sql::ast::{
        CreateSubsourceOptionName, TableFromSourceOptionName, Value, WithOptionValue,
    };
    use mz_sql_server_util::desc::{SqlServerTableConstraint, SqlServerTableConstraintType};
    use mz_storage_types::sources::ProtoSourceExportStatementDetails;
    use mz_storage_types::sources::proto_source_export_statement_details::Kind;

    let deets: Option<&mut String> = match stmt {
        Statement::CreateSubsource(stmt) => stmt.with_options.iter_mut().find_map(|option| {
            if matches!(option.name, CreateSubsourceOptionName::Details)
                && let Some(WithOptionValue::Value(Value::String(ref mut details))) = option.value
            {
                Some(details)
            } else {
                None
            }
        }),
        Statement::CreateTableFromSource(stmt) => stmt.with_options.iter_mut().find_map(|option| {
            if matches!(option.name, TableFromSourceOptionName::Details)
                && let Some(WithOptionValue::Value(Value::String(ref mut details))) = option.value
            {
                Some(details)
            } else {
                None
            }
        }),
        _ => None,
    };
    let Some(deets) = deets else {
        return Ok(());
    };

    let current_value = hex::decode(&mut *deets)?;
    let current_value = ProtoSourceExportStatementDetails::decode(&*current_value)?;

    // avoid further work if this isn't SQL Server
    if !matches!(current_value.kind, Some(Kind::SqlServer(_))) {
        return Ok(());
    };

    let SourceExportStatementDetails::SqlServer {
        mut table,
        capture_instance,
        initial_lsn,
    } = SourceExportStatementDetails::from_proto(current_value)?
    else {
        unreachable!("statement details must exist for SQL Server");
    };

    // Migration has already occured or did not need to happen.
    if !table.constraints.is_empty() {
        return Ok(());
    }

    // Relocates the primary key constraint information from the individual columns to the
    // constraints field. This ensures that the columns no longer hold constraint information.
    let mut migrated_constraints: BTreeMap<_, Vec<_>> = BTreeMap::new();
    for col in table.columns.iter_mut() {
        if let Some(constraint_name) = col.primary_key_constraint.take() {
            migrated_constraints
                .entry(constraint_name)
                .or_default()
                .push(col.name.to_string());
        }
    }

    table.constraints = migrated_constraints
        .into_iter()
        .map(|(constraint_name, column_names)| SqlServerTableConstraint {
            constraint_name: constraint_name.to_string(),
            constraint_type: SqlServerTableConstraintType::PrimaryKey,
            column_names,
        })
        .collect();

    let new_value = SourceExportStatementDetails::SqlServer {
        table,
        capture_instance,
        initial_lsn,
    };
    *deets = hex::encode(new_value.into_proto().encode_to_vec());

    Ok(())
}

/// Add missing item IDs to the ON clauses of CREATE INDEX statements.
fn ast_rewrite_add_missing_index_ids(
    tx: &Transaction<'_>,
    stmt: &mut Statement<Raw>,
) -> Result<(), anyhow::Error> {
    let Statement::CreateIndex(stmt) = stmt else {
        return Ok(());
    };

    let unresolved_name = match stmt.on_name.clone() {
        mz_sql::ast::RawItemName::Name(name) => name,
        // ID already present; nothing to do.
        mz_sql::ast::RawItemName::Id(..) => return Ok(()),
    };

    let parts = &unresolved_name.0;
    let (db_name, schema_name, item_name) = match parts.len() {
        3 => (Some(&parts[0]), &parts[1], &parts[2]),
        2 => (None, &parts[0], &parts[1]),
        _ => panic!("invalid unresolved name: {unresolved_name:?}"),
    };

    let db_id = db_name.map(|x| {
        let db = tx.get_databases().find(|db| db.name == x.as_str());
        let db = db.unwrap_or_else(|| panic!("missing database: {x}"));
        db.id
    });
    let schema_id = {
        let schema = tx
            .get_schemas()
            .find(|s| s.name == schema_name.as_str() && s.database_id == db_id);
        let schema = schema.unwrap_or_else(|| panic!("missing schema: {schema_name}, {db_id:?}"));
        schema.id
    };
    let item_id = {
        let item = tx
            .get_items()
            .find(|i| i.name == item_name.as_str() && i.schema_id == schema_id);
        let item = item.unwrap_or_else(|| panic!("missing item: {item_name}, {schema_id:?}"));
        item.id
    };

    stmt.on_name = mz_sql::ast::RawItemName::Id(item_id.to_string(), unresolved_name, None);

    Ok(())
}

/// Add missing item IDs to `DOC ON` references in CREATE SINK statements.
///
/// `DOC ON TYPE x` and `DOC ON COLUMN x.c` used to persist a type reference
/// as a bare qualified name, unlike a relation in the same position, because
/// name resolution suppressed ids for types. Resolution now prints the id
/// (see `NameResolver::resolve_doc_on_name`); this rewrites stored statements
/// to match, so `create_sql` reference extraction (`mz_object_dependencies`)
/// recovers the sink's edge to the type.
///
/// References that already carry an id are skipped, so this is idempotent and
/// safe to run every boot.
fn ast_rewrite_add_missing_doc_on_ids(
    tx: &Transaction<'_>,
    stmt: &mut Statement<Raw>,
) -> Result<(), anyhow::Error> {
    if !matches!(stmt, Statement::CreateSink(_)) {
        return Ok(());
    }
    rewrite_doc_on_ids(stmt, |name| {
        let parts = &name.0;
        let (db_name, schema_name, item_name) = match parts.len() {
            3 => (Some(&parts[0]), &parts[1], &parts[2]),
            // Names in ambient schemas denote builtin types, which are resolved
            // through the system object mappings.
            2 => (None, &parts[0], &parts[1]),
            _ => panic!("invalid doc on reference: {name:?}"),
        };
        let db_id = db_name.map(|db_name| {
            let db = tx.get_databases().find(|db| db.name == db_name.as_str());
            let db = db.unwrap_or_else(|| panic!("missing database in doc on reference: {name:?}"));
            db.id
        });
        let schema = tx
            .get_schemas()
            .find(|s| s.name == schema_name.as_str() && s.database_id == db_id);
        let schema =
            schema.unwrap_or_else(|| panic!("missing schema in doc on reference: {name:?}"));

        let user_type = tx.get_items().find(|i| {
            i.name == item_name.as_str()
                && i.schema_id == schema.id
                // A type may share its (schema, name) with a secret, connection, sink,
                // or function, so explicitly filter on Type.
                && i.item_type() == CatalogItemType::Type
        });
        if let Some(item) = user_type {
            return Some(item.id.to_string());
        }
        // A bare name in an ambient schema denotes a builtin type (an explicit
        // `DOC ON TYPE int4`, or a DOC ON injected by purification after mz_system
        // commented on one). Builtin types are not durable items, so the lookup
        // above cannot find them; their ids live in the system object mappings.
        let builtin_type = tx.get_system_object_mappings().find(|m| {
            m.description.schema_name == schema.name
                && m.description.object_type == CatalogItemType::Type
                && m.description.object_name == item_name.as_str()
        });
        let builtin_type =
            builtin_type.unwrap_or_else(|| panic!("missing type in doc on reference: {name:?}"));
        Some(builtin_type.unique_identifier.catalog_id.to_string())
    });
    Ok(())
}

/// Replaces every bare `DOC ON` reference in `stmt` for which `lookup`
/// returns an id with the `[id AS name]` form. `None` leaves the reference as
/// a name.
fn rewrite_doc_on_ids(
    stmt: &mut Statement<Raw>,
    lookup: impl FnMut(&mz_sql::ast::UnresolvedItemName) -> Option<String>,
) {
    use mz_sql::ast::visit_mut::{VisitMut, VisitMutNode};
    use mz_sql::ast::{ColumnName, DocOnIdentifier, RawItemName, UnresolvedItemName};

    struct Rewriter<F> {
        lookup: F,
    }

    impl<'ast, F> VisitMut<'ast, Raw> for Rewriter<F>
    where
        F: FnMut(&UnresolvedItemName) -> Option<String>,
    {
        fn visit_doc_on_identifier_mut(&mut self, node: &mut DocOnIdentifier<Raw>) {
            let name = match node {
                DocOnIdentifier::Type(name) => name,
                DocOnIdentifier::Column(ColumnName {
                    relation,
                    column: _,
                }) => relation,
            };
            if let RawItemName::Name(unresolved) = name {
                if let Some(id) = (self.lookup)(unresolved) {
                    let unresolved = unresolved.clone();
                    *name = RawItemName::Id(id, unresolved, None);
                }
            }
        }
    }

    stmt.visit_mut(&mut Rewriter { lookup });
}

/// Strips the `VERSION` qualifier from by-id references to non-user (builtin)
/// items.
///
/// A version pin on a builtin is meaningless, since builtins are not
/// user-versioned. Older binaries could still persist such a pin, and once the
/// builtin is converted to an item type without versions (e.g. a materialized
/// view) reparsing the pin fails with `InvalidVersion` and panics during
/// catalog open, wedging the upgrade. Stripping it makes the reference resolve
/// to the latest version.
///
/// The read-side resolver tolerates these pins too, so this is durable cleanup
/// rather than a correctness requirement. Safe to run every boot: stripping an
/// absent version is a no-op.
fn ast_rewrite_strip_builtin_version_pins(stmt: &mut Statement<Raw>) -> Result<(), anyhow::Error> {
    use mz_sql::ast::RawItemName;
    use mz_sql::ast::visit_mut::{VisitMut, VisitMutNode};

    struct StripBuiltinVersionPins;

    impl<'ast> VisitMut<'ast, Raw> for StripBuiltinVersionPins {
        fn visit_item_name_mut(&mut self, item_name: &mut RawItemName) {
            if let RawItemName::Id(id, _, version) = item_name {
                if version.is_some() {
                    if let Ok(parsed) = id.parse::<CatalogItemId>() {
                        if !parsed.is_user() {
                            *version = None;
                        }
                    }
                }
            }
        }
    }

    let mut visitor = StripBuiltinVersionPins;
    stmt.visit_mut(&mut visitor);
    Ok(())
}

fn ast_rewrite_kafka_metadata_refresh_intervals(
    stmt: &mut Statement<Raw>,
) -> Result<(), anyhow::Error> {
    use mz_sql::ast::{
        CreateSinkConnection, CreateSourceConnection, KafkaSinkConfigOptionName,
        KafkaSourceConfigOptionName, WithOptionValue,
    };
    // A user can persist the interval either as a string literal
    // (`WithOptionValue::Value`) or, if they wrote it as a double-quoted
    // value, as a lexed identifier (`WithOptionValue::UnresolvedItemName`).
    // Both shapes must be handled here.
    let interval: Option<&mut WithOptionValue<Raw>> = match stmt {
        Statement::CreateSource(stmt) => {
            if let CreateSourceConnection::Kafka { options, .. } = &mut stmt.connection {
                options.iter_mut().find_map(|option| {
                    if matches!(
                        option.name,
                        KafkaSourceConfigOptionName::TopicMetadataRefreshInterval
                    ) {
                        option.value.as_mut()
                    } else {
                        None
                    }
                })
            } else {
                None
            }
        }
        Statement::CreateSink(stmt) => {
            if let CreateSinkConnection::Kafka { options, .. } = &mut stmt.connection {
                options.iter_mut().find_map(|option| {
                    if matches!(
                        option.name,
                        KafkaSinkConfigOptionName::TopicMetadataRefreshInterval
                    ) {
                        option.value.as_mut()
                    } else {
                        None
                    }
                })
            } else {
                None
            }
        }
        _ => None,
    };

    let Some(interval) = interval else {
        return Ok(());
    };

    rewrite_interval_option_floor_1s(interval, "kafka metadata refresh interval")
}

/// Planning enforces a 1 second minimum `COMMIT INTERVAL`, but smaller
/// intervals used to be accepted (and a sub-millisecond one left the sink
/// unable to ever commit). Rewrite any persisted smaller interval to 1s so
/// the sink still plans after an upgrade.
///
/// `COMMIT INTERVAL` is a generic `CREATE SINK` option in the grammar, but only
/// Iceberg sinks accept it and only Iceberg planning enforces the 1s floor.
fn ast_rewrite_small_commit_intervals(stmt: &mut Statement<Raw>) -> Result<(), anyhow::Error> {
    use mz_sql::ast::{CreateSinkConnection, CreateSinkOptionName};

    let Statement::CreateSink(stmt) = stmt else {
        return Ok(());
    };
    if !matches!(stmt.connection, CreateSinkConnection::Iceberg { .. }) {
        return Ok(());
    }
    let interval = stmt.with_options.iter_mut().find_map(|o| {
        if matches!(o.name, CreateSinkOptionName::CommitInterval) {
            o.value.as_mut()
        } else {
            None
        }
    });
    let Some(interval) = interval else {
        return Ok(());
    };

    rewrite_interval_option_floor_1s(interval, "commit interval")
}

/// Planning rejects unsupported column collations, but `CREATE TABLE` used to
/// accept and ignore them. Strip them so the table still plans after an
/// upgrade. The planner never applied them, so the table's columns are
/// unchanged.
fn ast_rewrite_strip_unsupported_column_collations(
    stmt: &mut Statement<Raw>,
) -> Result<(), anyhow::Error> {
    let Statement::CreateTable(stmt) = stmt else {
        return Ok(());
    };
    for column in &mut stmt.columns {
        column
            .collation
            .take_if(|collation| !mz_sql::plan::is_supported_collation(collation));
    }
    Ok(())
}

/// Rewrites an interval option value to `'1s'` if it is below 1 second.
fn rewrite_interval_option_floor_1s(
    value: &mut mz_sql::ast::WithOptionValue<Raw>,
    label: &str,
) -> Result<(), anyhow::Error> {
    use mz_sql::ast::{Value, WithOptionValue};
    use mz_sql::plan::TryFromValue;

    let dur = Duration::try_from_value(value.clone())
        .map_err(|e| anyhow::anyhow!("invalid value for {label}: {value:?}: {e}"))?;

    if dur < Duration::from_secs(1) {
        *value = WithOptionValue::Value(Value::String("1s".to_string()));
    }

    Ok(())
}

#[cfg(test)]
mod tests {
    use super::*;

    fn strip(sql: &str) -> String {
        let mut stmt = mz_sql::parse::parse(sql)
            .expect("test sql parses")
            .into_element()
            .ast;
        ast_rewrite_strip_builtin_version_pins(&mut stmt).expect("rewrite succeeds");
        stmt.to_ast_string_stable()
    }

    #[mz_ore::test]
    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `rust_psm_stack_pointer` (SQL parser stack growth)
    fn strips_version_from_builtin_reference() {
        // A system (builtin) id must lose its version pin: builtins are never
        // user-versioned, so a persisted `VERSION 0` is a stale artifact that
        // wedges catalog open once the builtin is converted to a non-table.
        let out = strip(
            r#"CREATE VIEW "materialize"."public"."v" AS SELECT 1 FROM [s518 AS "mz_catalog"."mz_audit_events" VERSION 0]"#,
        );
        assert!(!out.contains("VERSION"), "version not stripped: {out}");
        // The reference itself is preserved, only the version is dropped.
        assert!(out.contains("s518"), "reference dropped: {out}");
    }

    #[mz_ore::test]
    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `rust_psm_stack_pointer` (SQL parser stack growth)
    fn preserves_version_on_user_reference() {
        // A user table legitimately carries versions (`ALTER TABLE ... ADD
        // COLUMN`), so the pin must survive.
        let out = strip(
            r#"CREATE VIEW "materialize"."public"."v" AS SELECT 1 FROM [u5 AS "materialize"."public"."t" VERSION 1]"#,
        );
        assert!(out.contains("VERSION"), "user version stripped: {out}");
        assert!(out.contains("u5"), "reference dropped: {out}");
    }

    #[mz_ore::test]
    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `rust_psm_stack_pointer` (SQL parser stack growth)
    fn leaves_unpinned_builtin_reference_untouched() {
        let out = strip(
            r#"CREATE VIEW "materialize"."public"."v" AS SELECT 1 FROM [s518 AS "mz_catalog"."mz_audit_events"]"#,
        );
        assert!(!out.contains("VERSION"), "unexpected version: {out}");
        assert!(out.contains("s518"), "reference dropped: {out}");
    }

    #[mz_ore::test]
    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `rust_psm_stack_pointer` (SQL parser stack growth)
    fn strips_unsupported_column_collations() {
        let mut stmt = mz_sql::parse::parse(
            r#"CREATE TABLE "materialize"."public"."t" ("a" [s20 AS "pg_catalog"."text"] COLLATE "unicode", "b" [s21 AS "pg_catalog"."int4"] COLLATE "nonexistent", "c" [s20 AS "pg_catalog"."text"] COLLATE "pg_catalog"."default" NOT NULL)"#,
        )
        .expect("test sql parses")
        .into_element()
        .ast;
        ast_rewrite_strip_unsupported_column_collations(&mut stmt).expect("rewrite succeeds");
        assert_eq!(
            stmt.to_ast_string_stable(),
            r#"CREATE TABLE "materialize"."public"."t" ("a" [s20 AS "pg_catalog"."text"], "b" [s21 AS "pg_catalog"."int4"], "c" [s20 AS "pg_catalog"."text"] COLLATE "pg_catalog"."default" NOT NULL)"#,
        );
    }

    fn add_doc_on_ids(
        sql: &str,
        lookup: impl FnMut(&mz_sql::ast::UnresolvedItemName) -> Option<String>,
    ) -> String {
        let mut stmt = mz_sql::parse::parse(sql)
            .expect("test sql parses")
            .into_element()
            .ast;
        rewrite_doc_on_ids(&mut stmt, lookup);
        stmt.to_ast_string_stable()
    }

    #[mz_ore::test]
    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `rust_psm_stack_pointer` (SQL parser stack growth)
    fn adds_ids_to_bare_doc_on_references() {
        // The persisted shape of a sink from before types printed ids in DOC
        // ON positions: relations carry ids, the type and its column carry
        // bare names.
        let out = add_doc_on_ids(
            r#"CREATE SINK "materialize"."public"."s" FROM [u1 AS "materialize"."public"."t"] INTO KAFKA CONNECTION [u2 AS "materialize"."public"."kc"] (TOPIC = 'top') FORMAT AVRO USING CONFLUENT SCHEMA REGISTRY CONNECTION [u3 AS "materialize"."public"."csr"] (DOC ON TYPE "materialize"."public"."point" = 'p', DOC ON COLUMN "materialize"."public"."point"."x" = 'x', DOC ON TYPE "pg_catalog"."int4" = 'i') ENVELOPE UPSERT"#,
            |name| match name.0.len() {
                3 => {
                    assert_eq!(name.0[2].as_str(), "point", "unexpected lookup: {name:?}");
                    Some("u9".into())
                }
                2 => {
                    assert_eq!(name.0[1].as_str(), "int4", "unexpected lookup: {name:?}");
                    Some("s23".into())
                }
                _ => panic!("unexpected lookup: {name:?}"),
            },
        );
        assert!(
            out.contains(r#"DOC ON TYPE [u9 AS "materialize"."public"."point"]"#),
            "type reference not rewritten: {out}"
        );
        assert!(
            out.contains(r#"DOC ON COLUMN [u9 AS "materialize"."public"."point"]."x""#),
            "column reference not rewritten: {out}"
        );
        assert!(
            out.contains(r#"DOC ON TYPE [s23 AS "pg_catalog"."int4"]"#),
            "builtin type reference not rewritten: {out}"
        );
    }

    #[mz_ore::test]
    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `rust_psm_stack_pointer` (SQL parser stack growth)
    fn leaves_id_doc_on_references_untouched() {
        let out = add_doc_on_ids(
            r#"CREATE SINK "materialize"."public"."s" FROM [u1 AS "materialize"."public"."t"] INTO KAFKA CONNECTION [u2 AS "materialize"."public"."kc"] (TOPIC = 'top') FORMAT AVRO USING CONFLUENT SCHEMA REGISTRY CONNECTION [u3 AS "materialize"."public"."csr"] (DOC ON TYPE [u4 AS "materialize"."public"."point"] = 'p', DOC ON COLUMN [u1 AS "materialize"."public"."t"]."c1" = 'c') ENVELOPE UPSERT"#,
            |name| panic!("id reference must not be resolved: {name:?}"),
        );
        assert!(
            out.contains(r#"DOC ON TYPE [u4 AS "materialize"."public"."point"]"#),
            "id reference altered: {out}"
        );
    }
}