wasm-dbms 0.9.0

0.9.0

Released on 2026-04-28

⚠ Breaking Changes

• acl: granular per-identity permissions (closes #87)

  granular per-identity permissions (closes #87)

Added

• query: add DISTINCT support to query API

  Adds distinct_by field on Query and .distinct(&[..]) builder method.
  The select pipeline deduplicates rows by the listed columns before applying
  ORDER BY / OFFSET / LIMIT, matching standard SQL semantics. Closes #85.

• query: add aggregates, GROUP BY, and HAVING (#86)

  Introduce Database::aggregate with COUNT/SUM/AVG/MIN/MAX, GROUP BY,
  HAVING, and aggregate-aware ORDER BY/LIMIT/OFFSET. Expose the new flow
  through the IC canister macro (aggregate_<table>) and the
  ic-dbms-client Client trait + all three client impls.

  • AggregateFunction / AggregatedRow / AggregatedValue types in
    wasm-dbms-api; HAVING / ORDER BY reference aggregate outputs by
    synthetic agg{N} names.
  • Plan-time validation: SUM/AVG numeric, unknown col/agg, LIKE/JSON
    rejected in HAVING, joins/eager relations rejected in aggregate.
  • Reject group_by/having on non-aggregate select paths via new
    QueryError::AggregateClauseInSelect (mirrors JoinInsideTypedSelect).
  • Fix Query candid serialization order after group_by/having insertion
    (idl_hash-sorted: eager_relations, distinct_by, joins, offset, limit,
    filter, group_by, having, order_by, columns).
  • Tests: 29 unit tests, 3 select-guard tests, 6 pocket-ic integration
    tests including round-trip via wrapper canister.
  • Docs: guides/querying.md aggregations section, reference/query.md
    Aggregate Types + Errors, ic/reference/schema.md endpoint listing,
    ic/guides/client-api.md client.aggregate example, errors.md updates.
  • Database trait method docs rewritten with proper Errors sections.

• migrations: add schema snapshot, Migrate trait, and macro support (#34)

  Lay the groundwork for schema migrations:

  • Snapshot types (TableSchemaSnapshot, ColumnSnapshot, IndexSnapshot,
    ForeignKeySnapshot, OnDeleteSnapshot, DataTypeSnapshot) with
    versioned binary Encode/Decode, Serialize/Deserialize, and
    feature-gated CandidType derives.
  • TableSchema::schema_snapshot() with default impl assembling from
    table_name / primary_key / columns / indexes and
    Encode::ALIGNMENT.
  • dbms::migration module: Migrate trait, MigrationOp, ColumnChanges,
    MigrationPolicy, MigrationError, plus DbmsError::Migration and
    prelude re-exports.
  • ColumnDef gains default: Option<fn() -> Value> (kept fn-pointer to
    preserve Copy) and renamed_from: &'static [&'static str].
  • #[derive(Table)] parses #[default = ...], #[renamed_from(...)], and
    #[migrate]; emits impl Migrate for T {} unless #[migrate] is set.
  • DatabaseSchema trait and #[derive(DatabaseSchema)] gain
    migrate_default, migrate_transform, compiled_snapshots dispatch
    methods (object-safe via Self: Sized).
  • WIT: new migration-error(string) variant; guest maps
    DbmsError::Migration.
  • Docs: schema reference covers the three new attributes; new
    docs/reference/migrations.md and docs/guides/migrations.md; errors
    reference lists every MigrationError variant.
  • Tests: 13 new macro tests cover #[default], #[renamed_from],
    #[migrate], dispatch fall-through, unknown-table behaviour, and
    multi-table snapshot ordering.

  Memory layer integration, the migration engine, Database wiring, IC
  endpoints, client surface, integration tests, and full WIT migrate APIs
  are tracked in the issue checklist.

• migrations: name-hash fingerprint + schema snapshot ledger

  Switch TableSchema::fingerprint to hash table_name() instead of
  TypeId so table identity stays stable across rebuilds and schema
  evolution. SchemaRegistry::register_table now detects name-hash
  collisions eagerly: when the fingerprint slot is occupied, it loads the
  persisted snapshot and compares names, returning the new
  MemoryError::NameCollision variant on mismatch.

  Wire SchemaSnapshotLedger into the registry: register_table
  initializes the snapshot on the dedicated page, and the encode/decode of
  SchemaRegistry now persists schema_snapshot_page. Add full module-,
  struct-, and method-level docs to SchemaSnapshotLedger plus coverage
  tests for init/load/write/get and page isolation.

  Update docs/technical/memory.md with the new per-table snapshot page,
  the name-hash fingerprint semantics, the collision-detection flow, the
  ledger API, and the TableSchemaSnapshot serialization layout.

• migrations: engine + Database wiring for schema migrations

  Wire the migration engine into the generic wasm-dbms crate so callers
  can detect drift, plan migrations, and apply structural ops through the
  existing Database trait surface.

  Engine layer (crates/wasm-dbms/wasm-dbms/src/database/migration/):

  • snapshots: drift hash via xxh3 (xxhash-rust), seeded with
    TableSchemaSnapshot::latest_version(). compute_drift compares
    persisted snapshots (loaded through SchemaRegistry::stored_snapshots)
    with S::compiled_snapshots() reachable through the boxed schema.
  • diff: pure stored-vs-compiled diff producing Vec<MigrationOp>.
    Renames resolved via the new DatabaseSchema::renamed_from_dyn
    dispatch, type changes routed through the widening whitelist or
    MigrationOp::TransformColumn.
  • plan: deterministic op ordering plus policy-driven validation
    (destructive-op gate, defense-in-depth missing-default check).
  • apply: journaled execution of structural ops (CreateTable,
    DropTable, AlterColumn, AddIndex, DropIndex); tightening
    validation (nullable: false, unique: true) scans existing rows
    through schema dispatch. Column-mutating ops (AddColumn,
    DropColumn, RenameColumn, WidenColumn, TransformColumn) return
    the new MigrationError::DataRewriteUnsupported until the
    snapshot-driven (de)serializer (issue #91) lands. DropTable leaks
    pages (issue #90).

  Database wiring:

  • DbmsContext gains Cell<Option<bool>> drift (lazy cache) and
    Cell<bool> migrating (apply-in-progress guard so internal reads
    bypass the gate).
  • WasmDbmsDatabase::ensure_no_drift is called as the first line of
    every Database method except rollback. ACL methods on
    DbmsContext continue to bypass.
  • Database trait gains has_drift, pending_migrations (renamed
    from the previous plan_migration), and migrate(policy). The IC
    adapter and any other Database implementor must now implement them;
    the test mock in wasm-dbms-api is updated.

  API additions (wasm-dbms-api):

  • fingerprint_for_name(&str) -> u64 (xxh3_64), so the engine can
    derive a registry key for tables it knows only by name.
  • MigrationError::DataRewriteUnsupported { op }.
  • xxhash-rust workspace dependency.

  Memory layer (wasm-dbms-memory):

  • SchemaRegistry::{stored_snapshots, register_table_from_snapshot, unregister_table, table_registry_page_by_name}.
  • IndexLedger::init_from_keys for snapshot-keyed init paths.
  • test_utils::write_dummy_schema_snapshot so the table-registry unit
    tests (which allocate raw pages and call TableRegistry::load) can
    satisfy the snapshot-ledger decode that landed in #34's prior commit.

  Macro additions (wasm-dbms-macros):

  • Emit compiled_snapshots_dyn, migrate_default_dyn,
    migrate_transform_dyn, and renamed_from_dyn as object-safe
    siblings of the existing Sized dispatch methods.

  Benches (ic-dbms-canister):

  • Replace hand-written DatabaseSchema impls in eager_relation and
    read_table with #[derive(DatabaseSchema)]. ~330 LOC of boilerplate
    removed; the derive macro covers the same dispatch.

• migrations: IC layer endpoints for schema migrations

  Wires the migration surface through the IC layer: has_drift and
  pending_migrations as queries, migrate as an update, all admin-gated
  via the existing ACL check. Adds matching Client trait methods with
  implementations for IcDbmsCanisterClient, IcDbmsAgentClient, and
  IcDbmsPocketIcClient, plus wrapper-canister bindings and PocketIC
  coverage exercising both the direct client and wrapper paths on a
  fresh canister (no drift, empty plan, migrate is a no-op).

• migrations: WIT surface and IC migration docs

  Add has-drift / pending-migrations / migrate to wit/dbms.wit plus the
  supporting snapshot, op, and policy records. Wire them through the
  example guest and host so the round-trip is exercised end-to-end.

  Document the IC migration flow: new docs/ic/guides/migrations.md, full
  Candid signatures in docs/ic/reference/schema.md, and a Schema
  Migrations section in docs/ic/guides/client-api.md.

• migrations: snapshot-driven record codec for column-mutating ops
• memory: page reclamation via unclaimed-pages ledger (closes #90)

  Reserve page 2 for an LIFO ledger of pages released by destructive ops.
  Rename MemoryAccess::allocate_page to claim_page and add unclaim_page,
  both built on new grow_one_page / zero_page primitives. claim_page
  pops the ledger before bumping the high-water mark; unclaim_page zeros
  the page and pushes it. JournaledWriter records the full pre-zero page so
  rollback restores ledger state and c...

wasm-dbms 0.8.2

0.8.2

Released on 2026-04-21

Changed

• derive Default on generated *UpdateRequest structs

  All fields on update requests are Option<T>, so Default is always
  derivable. This enables the ..Default::default() struct-update pattern
  when constructing partial updates, removing the need to write None
  for every untouched field.

wasm-dbms 0.8.1

0.8.1

Released on 2026-04-05

Added

• add select_join to Database trait and rename CandidColumnDef to JoinColumnDef

  Move select_join from a #[doc(hidden)] method on WasmDbmsDatabase to a
  proper public method on the Database trait, making it the official API
  for join queries. Rename CandidColumnDef to JoinColumnDef to better
  reflect its purpose as the column definition type that carries table
  provenance for join results. Update all docs to reference select_join
  instead of select_raw for join queries.

Fixed

• apply LIMIT and OFFSET after ORDER BY to match SQL semantics

  LIMIT and OFFSET were applied during the table scan (before sorting),
  so ORDER BY + LIMIT didn't produce correct "top N sorted" results.
  Now, when ORDER BY is present, LIMIT and OFFSET are deferred to after
  sorting, matching standard SQL evaluation order. The early-exit
  optimization is preserved when no ORDER BY is specified.

• align insert offset to RawRecord alignment for fixed-size records

  The second insert into a table with all fixed-size columns (e.g. Uint64 +
  Uint64 + 1-byte CustomDataType) failed with OffsetNotAligned because:

  1. table_registry::insert aligned the write offset using E::ALIGNMENT
     instead of RawRecord::ALIGNMENT (which includes the 2-byte header).
  2. page_ledger::commit double-wrapped the type in RawRecord, computing
     RawRecord<RawRecord>::ALIGNMENT instead of RawRecord::ALIGNMENT,
     causing incorrect free-space tracking.

wasm-dbms 0.8.0

0.8.0

Released on 2026-04-05

Added

• add select_join to Database trait and rename CandidColumnDef to JoinColumnDef

  Move select_join from a #[doc(hidden)] method on WasmDbmsDatabase to a
  proper public method on the Database trait, making it the official API
  for join queries. Rename CandidColumnDef to JoinColumnDef to better
  reflect its purpose as the column definition type that carries table
  provenance for join results. Update all docs to reference select_join
  instead of select_raw for join queries.

Fixed

• apply LIMIT and OFFSET after ORDER BY to match SQL semantics

  LIMIT and OFFSET were applied during the table scan (before sorting),
  so ORDER BY + LIMIT didn't produce correct "top N sorted" results.
  Now, when ORDER BY is present, LIMIT and OFFSET are deferred to after
  sorting, matching standard SQL evaluation order. The early-exit
  optimization is preserved when no ORDER BY is specified.

wasm-dbms 0.7.2

0.7.2

Released on 2026-04-02

Fixed

• export macros in wasm-dbms-api

  wasm-dbms-macros were not actually exported as the documentation example were showing

wasm-dbms 0.7.1

0.7.1

Released on 2026-04-01

Documentation

• build docs from mdbook

Fixed

• rename generated loop variables in Table macro to avoid shadowing user field names

  Fields named value caused compile errors because the generated code
  used value as both the loop binding and destructure binding, shadowing
  the field accumulator variable.

Miscellaneous

• derive CustomDataType on principal

wasm-dbms 0.7.0

0.7.0

Released on 2026-03-30

⚠ Breaking Changes

• change MemoryProvider::read and MemoryAccess::read_at to take &mut self

  MemoryProvider::read signature changed from &self to &mut self.

Added

• bench: add benchmark comparison crate against other in-memory DBMS

  Add wasm-dbms-bench crate with Criterion benchmarks comparing wasm-dbms
  against SQLite (in-memory) and DuckDB (in-memory) across CRUD operations,
  bulk inserts, queries (filter, order, join), and transactions.

  Includes CI workflow for running benchmarks and uploading artifacts.

• B+ tree indexes for accelerated queries

  Add a complete B+ tree index system to wasm-dbms. Every table
  automatically gets an index on its primary key, and users can declare
  additional single-column or composite indexes with the #[index]
  attribute.

  Key changes:

  Memory layer (wasm-dbms-memory):

  • IndexLedger: per-table registry mapping column sets to B-tree roots
  • IndexTree: page-per-node B+ tree with variable-size keys, doubly-linked
    leaves for range scans, and automatic node splitting/merging
  • RecordAddress: lightweight (page, offset) pointer stored in leaf entries
  • SchemaRegistry/TableRegistryPage extended with index_registry_page
  • TableRegistry now owns and exposes an IndexLedger
  • INSERT/UPDATE/DELETE maintain all indexes eagerly

  DBMS layer (wasm-dbms):

  • FilterAnalyzer: extracts index plans (Eq, Range, In) from query filters
  • IndexReader: unified view merging base B-tree results with transaction
    overlay additions/removals
  • IndexOverlay: in-memory BTreeMap tracking uncommitted index changes per
    transaction, flushed on commit, discarded on rollback
  • SELECT, UPDATE, DELETE, and JOIN queries use indexes when a suitable
    plan is found; remaining filter conditions applied as residual checks

  Macro layer (wasm-dbms-macros):

  • #[index] attribute on fields for single-column indexes
  • #[index(group = "name")] for composite indexes
  • Automatic primary key index generation in TableSchema
  • Deduplicated shared macro logic from ic-dbms-macros into wasm-dbms-macros

  Also includes CI improvements, dependency updates, and documentation
  updates covering the index memory layout, query optimization, and
  architecture changes.

• add wasi-dbms-memory crate with file-backed MemoryProvider

  Implements WasiMemoryProvider backed by a single flat file,
  enabling wasm-dbms to persist data on any WASI-compliant runtime
  (Wasmer, Wasmtime, WasmEdge). The file layout is byte-for-byte
  equivalent to IC stable memory.

• add #[unique] attribute for table fields

  Add support for the #[unique] field attribute that enforces uniqueness
  constraints on non-primary-key columns. A unique field automatically
  gets a B+ tree index for efficient O(log n) duplicate detection.

  • Parse #[unique] in Table derive macro, set ColumnDef::unique and
    auto-generate an index for the field
  • Add UniqueConstraintViolation error variant to QueryError
  • Enforce uniqueness in InsertIntegrityValidator and
    UpdateIntegrityValidator (update allows keeping own value)
  • Add comprehensive tests for insert, update, and transaction scenarios
  • Update schema, errors, and IC reference documentation

• add #[autoincrement] attribute for table fields

  Add support for autoincrement columns in table schemas. Fields annotated
  with #[autoincrement] automatically generate sequential values on
  insert, starting from zero and incrementing by one.

  Implementation across all layers:

  Memory layer (wasm-dbms-memory):

  • AutoincrementLedger: per-table ledger storing current counter values
    for each autoincrement column, persisted to a dedicated memory page
  • AutoincrementRegistry: HashMap-based registry mapping column names to
    their current Value, with custom Encode implementation
  • SchemaRegistry: conditionally allocates an autoincrement page when a
    table has autoincrement columns (Option in TableRegistryPage)
  • TableRegistry: integrates AutoincrementLedger as an optional field,
    exposes autoincrement_next() to get the next value for a column

  API layer (wasm-dbms-api):

  • ColumnDef: add auto_increment field to column definitions
  • MemoryError::AutoincrementOverflow: new error variant returned when
    a column reaches its type's maximum value (uses checked_add)
  • Filter: support autoincrement columns in query filters

  Macro layer (wasm-dbms-macros):

  • Table derive macro: parse #[autoincrement] attribute on fields,
    propagate auto_increment flag to generated TableSchema impl

  DBMS layer (wasm-dbms):

  • Database: wire autoincrement through insert operations
  • Transaction overlay: support autoincrement in transactional context

  Supported types: Int8, Int16, Int32, Int64, Uint8, Uint16, Uint32,
  Uint64. Overflow returns AutoincrementOverflow error to prevent
  duplicate key generation.

CI

• run workflow only once in pr (branches main)
• run ci workflow against x.y.z branches
• install nightly and check format before installing stable toolchain
• install ic-wasm with curl

Changed

• remove duplicated macros from ic-dbms-macros

  Remove Encode, Table, CustomDataType, and DatabaseSchema derive macros
  from ic-dbms-macros, keeping only DbmsCanister. These macros were
  duplicated from wasm-dbms-macros with the only differences being crate
  path prefixes and Candid/Serde derives on generated types.

  IC crates now re-export the wasm-dbms-macros versions through their
  preludes. To support the IC requirement of Candid-serializable generated
  types, a #[candid] attribute is added to wasm-dbms-macros' Table derive:
  when present, generated Record, InsertRequest, and UpdateRequest types
  derive CandidType, Serialize, and Deserialize.

• 💥 change MemoryProvider::read and MemoryAccess::read_at to take &mut self

  File-backed providers need mutable access to seek before reading.
  Previously this was worked around with try_clone() on every read.
  Making the trait honest about mutation removes that overhead and
  simplifies implementations.

Documentation

• add WASI documentation
• update project description to better match the project identity

Fixed

• prevent PK from being indexed twice

  using #[index] on the primary key lead to duplicated index for the primary key

• track PK changes in overlay patch_row to chain subsequent operations
• add missing Int8, Int16, Uint8, Uint16 variants to DataTypeKind and CandidDataTypeKind

  Value enum already had these variants but DataTypeKind did not,
  causing compile errors when using 8-bit or 16-bit integer types
  in table field definitions via the derive macro.

• autoincrement macro codegen and DBMS integration

  Fix InsertRequest codegen for autoincrement fields:

  • from_values: wraps found values in Autoincrement::Value, absent ones
    in Autoincrement::Auto
  • into_values: skips Autoincrement::Auto fields, includes Value fields
  • into_record: unwraps Autoincrement::Value to inner type for schema

  Fix insert_contract test helper using wrong column index (order vs
  user_id). Add full coverage tests for autoincrement at the DBMS layer:
  sequential generation, explicit override, no recycle after delete,
  transaction commit/rollback counter behavior, from_values/into_values
  variants, and filter on autoincrement column.

  Wire autoincrement_next into TableRegistry as a public method.

Miscellaneous

• remove repeated compare benchmarks
• add Rust logo to crates.io badges in all READMEs
• removed kofi badge
• ignore .DS_Store

Build

• update dependencies

wasm-dbms 0.6.0

0.6.0

Released on 2026-03-02

⚠ Breaking Changes

• migrate Principal from built-in to CustomDataType

  Value::Principal and DataTypeKind::Principal removed.
  Principal fields in tables must now use #[custom_type] annotation.
  Existing stable memory schemas are incompatible (fingerprint change).

• restructure workspace into wasm-dbms and ic-dbms layers

  restructure workspace into wasm-dbms and ic-dbms layers

Added

• ic-dbms-api: add CustomValue struct with comparison and hashing
• ic-dbms-api: add CustomDataType trait
• ic-dbms-api: add Value::Custom variant and accessors
• ic-dbms-api: add DataTypeKind::Custom variant and CandidDataTypeKind

  Add Custom(&'static str) variant to DataTypeKind for user-defined types.
  Remove CandidType/Serialize/Deserialize derives from DataTypeKind since
  it no longer needs to cross API boundaries directly. Introduce
  CandidDataTypeKind as the Candid-serializable mirror with Custom(String)
  for the canister API layer. Update CandidColumnDef to use the new type.

• ic-dbms-macros: add #[derive(CustomDataType)] macro

  Add a proc-macro derive that generates impl CustomDataType (with
  TYPE_TAG constant) and impl From<T> for Value for user-defined types.
  The attribute #[type_tag = "..."] is required and uses the same
  NameValue parsing pattern as the existing #[table = "..."] attribute.

• ic-dbms-macros: add #[custom_type] support to Table derive macro

  When a field is annotated with #[custom_type], the generated code uses
  Value::Custom(CustomValue { ... }) instead of Value::FieldType(field)
  for to_values/from_values in TableSchema, Record, InsertRequest, and
  UpdateRequest. This allows user-defined types implementing CustomDataType
  to be used as table columns.

• 💥 migrate Principal from built-in to CustomDataType
• add WIT interface definition for wasm-dbms Component Model API
• add WIT guest crate with FileMemoryProvider and example schemas

  Create the wasm-dbms-example-guest crate scaffolding with:

  • FileMemoryProvider: file-backed MemoryProvider implementation with
    > persistence across process restarts and full test coverage
  • Example table schemas (User, Post) with ExampleDatabaseSchema
    > implementing the generic DatabaseSchema trait
  • register_tables helper for DBMS context initialization

  Fix wasm-dbms-macros to use DbmsError/DbmsResult instead of
  IcDbmsError/IcDbmsResult and remove IC-specific candid/serde derives
  from generated insert, update, and record structs, making the
  generic macro layer truly runtime-agnostic.

• implement WIT guest bridge layer for Component Model exports
• add Wasmtime host binary for WIT Component Model example

  Create the host-side binary that loads the guest WASM component via
  Wasmtime, provides WASI filesystem access, and exercises every exported
  database operation: insert, select, transactional commit, and rollback.

• add wasm-dbms dependency to ic-dbms-canister
• add #[derive(DatabaseSchema)] macro for automatic schema dispatch

  Add a DatabaseSchema derive macro that auto-generates the
  DatabaseSchema trait implementation from a #[tables(...)] attribute,
  eliminating ~130+ lines of boilerplate per schema. Two variants exist:
  a generic one in wasm-dbms-macros and an IC-specific one in
  ic-dbms-macros with IC crate paths. Update examples, tests, and docs.

• add AccessControl trait with associated Id type for runtime-agnostic ACL

  Introduce the AccessControl trait in wasm-dbms-memory to abstract access
  control behind a generic interface. Different runtimes can use different
  identity types: Vec (AccessControlList), Principal (IcAccessControlList),
  or () (NoAccessControl). The A: AccessControl generic parameter is propagated
  through DbmsContext, WasmDbmsDatabase, DatabaseSchema, integrity validators,
  join engine, and both derive macros. Default type parameters preserve backward
  compatibility.

• add journaling-based atomicity to MemoryManager

  Replace panic-based rollback in atomic() with a write-ahead journal in
  MemoryManager. All writes via write_at and zero are recorded when a
  journal is active, enabling byte-level rollback on error. This makes
  atomicity runtime-agnostic, removing the dependency on IC's
  trap-reverts-stable-memory semantics.

  Key changes:

  • Add JournalEntry, begin/commit/rollback_journal to MemoryManager
  • Refactor atomic() to use journal with nested-call awareness
  • Refactor commit() to use a single journal spanning all operations
  • Fix self vs db inconsistency in delete closure
  • Fix pre-existing clippy is_multiple_of lint
  • Add 14 journal unit tests and 1 commit-rollback integration test
  • Add docs/technical/atomicity.md

Changed

• 💥 restructure workspace into wasm-dbms and ic-dbms layers

  Split the monolithic ic-dbms crates into a two-layer architecture:

  • wasm-dbms (generic layer): runtime-agnostic DBMS engine (wasm-dbms-api,
    > wasm-dbms-memory, wasm-dbms, wasm-dbms-macros)
  • ic-dbms (IC layer): thin adapter for Internet Computer canister
    > integration (ic-dbms-api, ic-dbms-canister, ic-dbms-macros,
    > ic-dbms-client, example, integration-tests)

  Also fixes integration test wasm paths to account for the new directory
  depth and updates CI, docs, and build scripts accordingly.

• consolidate IC thread-locals into DbmsContext
• remove duplicated IC database engine module
• update ic-dbms-canister prelude to re-export from wasm-dbms
• update IC API layer to use wasm-dbms database engine
• slim down DbmsCanister macro to IC API only
• update IC canister tests to use wasm-dbms engine
• update CHANGELOG, docs, and API for custom data types and AccessControl trait

  Update CHANGELOG with custom data types, AccessControl, and DatabaseSchema entries.
  Remove CallerContext in favor of AccessControl trait. Update IC macros to use
  generic-layer AccessControl. Update example guest, Cargo.toml dependencies, and
  documentation across wasm-dbms and ic-dbms crates.

• remove IC-specific documentation from wasm-dbms crates

  The generic wasm-dbms layer should not reference IC-specific concepts.
  Remove all doc comments mentioning IC, canister, Principal, Candid,
  IcDbmsError, and IcDbmsResult from the wasm-dbms crates.

• make error types runtime-agnostic and replace ACL panic with error

  Rename IC-specific error variants to runtime-agnostic names
  (StableMemoryError → ProviderError, PrincipalError → IdentityDecodeError),
  add ConstraintViolation variant, replace panic in ACL last-identity removal
  with a proper error, simplify get_referenced_tables by removing thread-local
  cache, and add DbmsContext threading documentation.

• move journal from MemoryManager to transaction module

  Extract the write-ahead journal from the memory layer into the DBMS
  layer where it belongs as a transaction concern. Introduce MemoryAccess
  trait so memory-crate functions are generic over the writer, allowing
  JournaledWriter to intercept writes for rollback support.

Documentation

• add custom data types design document

  Design for issue #35: type-erased CustomValue approach with
  CustomDataType trait, no generic propagation through core API.
  Principal becomes a CustomDataType impl in 0.6, prerequisite
  for wasm-dbms extraction (#48) in 0.7.

• update CHANGELOG for 0.6.0 custom data types
• add custom data types guide and update references
• New website for wasm-dbms
• add Wasmtime WIT Component Model example documentation
• update architecture docs after IC deduplication

Fixed

• move design doc to .claude/plans, add convention to CLAUDE.md

  Design docs and plans belong in .claude/plans/ (gitignored),
  not in docs/plans/. Added this convention to CLAUDE.md.

• ic-dbms-macros: fix nullable custom type codegen using inner type

  When a custom type field is declared as Nullable, the macro now
  correctly uses the inner type T (not Nullable) for trait lookups
  like CustomDataType::TYPE_TAG and Encode::decode in all codegen paths.

• address code review findings

  • Replace String::leak() with OnceLock-based static cache in
    > Value::type_name() for Custom variants to prevent unbounded leaks
  • Add compile-time error when #[custom_type] and #[foreign_key] are
    > combined on the same field

• harden custom data types and add CustomValue constructor

  • Add cache size guard (max 64 entries) to Value::type_name() to
    > prevent unbounded memory leaks on IC
  • Replace panicking .expect() with non-panicking if-let-Ok decode
    > in macro codegen for custom types (record, insert, update)
  • Add CustomValue::new() constructor enforcing consistency between
    > type_tag, encoded bytes, and display string
  • Add Project table with #[custom_type] owner field to example canister
  • Add PocketIC integration tests for custom type CRUD and filtering

• exclude guest crate from native tests and fix clippy warning

  The guest crate targets wasm32-wasip2 and cannot link on native targets.
  Exclude it from just test using --workspace --exclude. Also fix a
  redundant_closure clippy warning in the host binary.

• update MSRV to 1.91.1, fix ACL persist-before-panic, fix clippy warnings

  • Set rust-version to 1.91.1 (actual MSRV per cargo msrv) across
    > workspace Cargo.toml, CLAUDE.md, and all docs
  • Replace is_multiple_of (Rust 1.87+) with modulo check for MSRV compat
  • Fix ACL remove_identity to check emptiness before persisting, preventing
    > corrupted state on non-IC runtimes
  • Add #[allow(clippy::approx_constant)] to JSON test modu...

ic-dbms 0.5.0

0.5.0

Released on 2026-02-27

⚠ Breaking Changes

• Remove generic T from Query, since it's unnecessary

  Remove T from Query and QueryBuilder

Added

• 💥 Remove generic T from Query, since it's unnecessary

  The T: TableSchema argument from Query and QueryBuilder was actually unnecessary, because it didn't provide
  any meaningful information. The T argument has just been moved to the dbms select method, in order to bring
  information to the selected entity.

• add generic select endpoint for untyped table queries (#10)

  Add a select_raw method to the Database trait and a select canister
  endpoint that returns Vec<Vec<(CandidColumnDef, Value)>>, enabling
  table queries by name without compile-time type information. This lays
  the groundwork for future SQL and JOIN support.

• ic-dbms-client: add select_raw method to allow selecting untyped columns
• implement JOIN support (INNER, LEFT, RIGHT, FULL) (#47)

  Add user-facing join guide content to the querying and relationships
  docs, create a technical deep-dive for the join engine, and update the
  architecture overview and index with join-related entries.
  Add cross-table join queries with nested-loop join engine, qualified
  column resolution, NULL padding for outer joins, and filter support
  on joined rows. Joins are available through the untyped select_raw
  path and the generated select canister endpoint.

Performance

• batch fetch foreign keys in eager relation loading (#41)

  Replace per-record N+1 foreign key fetching with a batched approach
  using Filter::In queries. Adds ForeignFetcher::fetch_batch trait method,
  HashSet-based FK deduplication, benchmarks, and uses the existing
  TableColumns type alias throughout.

ic-dbms 0.4.0

0.4.0

Released on 2026-02-06

• New features:
  • Issue 13: Added JSON filtering capabilities for querying JSON columns.
    • JsonFilter::Contains for PostgreSQL @> style structural containment checks
    • JsonFilter::Extract for extracting values at JSON paths with comparison operations
    • JsonFilter::HasKey for checking path existence in JSON structures
    • Path syntax supports dot notation with bracket array indices (e.g., user.items[0].name)
  • Issue 22: Added AgentClient for the ic-dbms-canister to interact with
    the IC from an IC Agent.
• Performance improvements:
  • Issue 11: Implemented in-place update instead of delete+insert strategy
    (#37).
    • Records whose size is unchanged are now overwritten directly in stable memory, avoiding unnecessary reallocation.
    • Records whose size changes still fall back to delete+reinsert.
    • Added UpdateIntegrityValidator that allows keeping the same primary key during updates.
    • Cascade primary key changes to referencing tables via update_pk_referencing_updated_table.
    • Extracted shared validation logic into integrity::common module.
  • Replaced the external like crate with a custom SQL LIKE pattern engine (#42).
    • The new iterative two-pointer algorithm runs in O(n*m) worst-case with O(1) space and zero heap allocation, replacing the previous recursive approach that had exponential worst-case complexity.
• Bug fixes:
  • Fixed an issue with the IcCanisterClient which called update with the wrong amount of arguments.
  • Fixed multi-column order_by applying sorts in the wrong order, causing only the last column's sort to survive
    (#39).
• Refactoring:
  • Moved workspace crates into crates/ directory for better project organization
    (#38).
  • Cleaned up dbms.rs with extracted helpers, immutable borrow fixes, and moved tests to a separate file
    (#40).
  • Reorganized and expanded project documentation (#31).
  • Increased test coverage for ic-dbms-api, ic-dbms-canister, and ic-dbms-client.
• Dependencies:
  • Issue 12: Bump pocket-ic to 12.0.0.