Migration Guide

This document covers breaking changes between manifest schema versions and how to upgrade.

v1.7.8 to v1.8.0 — Scoped undo restore (cross-trunk isolation)

Qleany version: v1.8.0

What changed

Undo/redo snapshot/restore are now scoped to the subtree they target instead of operating on the whole store. The v1.7.0 switch to im::HashMap made snapshot capture O(1) by cloning the entire store, but restore then replaced the entire store — so undoing an operation on one undoable trunk reverted every other trunk and all non-undoable data (the savepoint behaviour the snapshot system was meant to replace). This release fixes that while keeping the O(1) capture.

snapshot(ids) still clones the whole store (O(1), im structural sharing) but now also records root_ids. EntityTreeSnapshot gains a pub root_ids: Vec<EntityId> field.
restore walks the subtree rooted at root_ids (strong relationships only), in both the snapshot and the live store, and reconciles only that subtree: in-scope rows and the forward junctions they own are restored wholesale; the subtree’s placement in external owners/referrers is reconciled surgically (membership only, preserving sibling edits made on other undo stacks); entities created after the snapshot are deleted. New generated methods: Repository::restore_subtree and per-backward-relationship reconcile_backref_*. HashMapStoreSnapshot’s table/junction fields are now pub(crate).
UndoableCreateUseCase no longer wholesale-resets the owner relationship on undo/redo (it relied on set_relationships_in_owner, which clobbered concurrent siblings); it now leans on the already-surgical remove_multi plus the scoped restore.
UndoableSetRelationshipUseCase / UndoableMoveRelationshipUseCase no longer take a whole-store savepoint. They capture the affected junction row’s prior value and restore just that row (a surgical inverse). WriteRelUoW<RF> gains a get_relationship method to read the pre-image.

Behavioral changes

Cross-trunk isolation: undoing an operation on one undoable trunk no longer reverts other trunks or non-undoable data (settings, caches, etc.). This is the headline fix.
Precise restore events: restore now emits Created/Updated/Removed for exactly the affected ids (three-way diff of snapshot vs live), instead of a Created event for every entity in the store. Set/move-relationship undo emits a scoped Updated for the touched row instead of a whole-store AllEvent::Reset. UIs that relied on the old “everything changed” storm to force a full refresh may need to listen to the precise events.
Performance: capture stays O(1); restore is now O(subtree) plus an O(n) scan per weak backward relationship (only when one exists), instead of an O(store) replace.
Unchanged: snapshots remain in-memory only (store_snapshot is still #[serde(skip)]); id counters are still preserved across restore.

How to upgrade

Regenerate affected files: snapshot.rs, hashmap_store.rs, entity repository files (*_repository.rs), entity unit-of-work files (*_units_of_work.rs), the use-case traits file, and the generated create.rs, set_relationship.rs, move_relationship.rs use cases.
Custom feature use cases: no API change. Code following the documented pattern (self.snap = uow.snapshot(ids); … uow.restore(&snap) on undo) becomes scoped automatically as long as it passes the correct ids — undo no longer reverts unrelated data.
Hand-written WriteRelUoW implementations (rare): add the new get_relationship method (delegate to the entity repository’s get_relationship).
If you relied on undo reverting the whole store (the old behaviour): that no longer happens. For an explicit whole-database rollback, use create_savepoint / restore_to_savepoint (or restore_store) directly — do not route it through undo.

v1.7.3 to v1.7.4 — Event-hub shutdown via channel wakeup

Qleany version: v1.7.4

What changed

AppContext::quit_signal: Arc<AtomicBool> is removed. Background event-hub threads (EventHub::start_event_loop, EventHubClient::start, mobile bridge start_event_dispatch) no longer poll a flag every 100–500 ms. They now block on a flume::Selector that waits on either the event channel or a shutdown receiver, so idle wake-ups drop to 0 instead of 5–10 per thread per second.

AppContext gains two new fields:

pub shutdown_rx: Receiver<()> — handed to every spawned event-loop thread.
shutdown_tx: Arc<Mutex<Option<Sender<()>>>> (private) — the only live Sender clone. AppContext::shutdown() .take()s it, which makes every cloned receiver see Disconnected within microseconds.

EventHub::start_event_loop, EventHubClient::start, and start_event_dispatch now take Receiver<()> instead of Arc<AtomicBool>.

Behavioral changes

Idle CPU: ~0 wake-ups per event-loop thread (previously 5/s for EventHubClient, 10/s for EventHub::start_event_loop, 2/s for the mobile dispatcher). On a host app with many widgets each owning their own backend, savings scale linearly.
Shutdown latency: microseconds (was up to one polling interval).
AppContext::shutdown() is still &self and idempotent — second call is a no-op because the sender has already been taken.

How to upgrade

Regenerate the affected files: crates/common/src/event.rs, crates/frontend/src/app_context.rs, crates/frontend/src/event_hub_client.rs, crates/slint_ui/src/main.rs (if using Slint), and the mobile bridge events.rs / backend.rs (if using the mobile bridge).
Update hand-written callers: any code calling event_hub_client.start(ctx.quit_signal.clone()) must become event_hub_client.start(ctx.shutdown_rx.clone()). Same for start_event_loop and start_event_dispatch.
Hand-written code that read ctx.quit_signal directly (e.g. to coordinate other shutdown work) must switch to a different mechanism — the field no longer exists. The shutdown channel is single-purpose; if you need a broader shutdown bus, either subscribe to shutdown_rx from your own thread (you’ll see Disconnected when shutdown fires) or layer your own signal alongside.

v1.6.3 to v1.7.0 — redb replaced by in-memory HashMap store

Qleany version: v1.7.0

What changed

The Rust storage backend has been replaced. The redb embedded database and postcard serialization are gone. The new backend is an in-memory store using im::HashMap (persistent data structure with structural sharing), giving O(1) snapshots for undo/redo.

Behavioral changes

Rollback-safe transactions: Write transactions now automatically create a savepoint on begin_transaction(). If the transaction is dropped without commit() (e.g., on error), Drop restores the savepoint — undoing all partial mutations. Previously with redb, this was handled by redb’s own transaction abort on drop.
Faster snapshots: Undo/redo snapshots are O(1) instead of O(n) deep clones, thanks to im::HashMap structural sharing.
No serialization: Entities are stored as plain Rust types. No postcard encoding/decoding overhead.

How to upgrade

Regenerate affected files: Use the Qleany UI or CLI to regenerate the storage-related files: Cargo.toml (common crate), database.rs, db_context.rs, hashmap_store.rs, transactions.rs, snapshot.rs, error.rs, repository_factory.rs, setup.rs, entity table files (*_table.rs), entity repository files (*_repository.rs), and test files (transaction_tests.rs). The old redb_tests.rs and snapshot_tests.rs can be deleted — their tests have been merged into transaction_tests.rs.
Update your workspace Cargo.toml: Remove redb and postcard from [workspace.dependencies] if present. The im crate is added automatically by the generated common Cargo.toml.
Custom feature use cases: No changes needed — the UoW trait interface (begin_transaction, commit, rollback, create_savepoint, restore_to_savepoint) is unchanged. Your use case code works as before, now with automatic rollback on error.

v1.6.0 to v1.6.1 — Crate renaming and publishing metadata

Qleany version: v1.6.1

What changed

No manifest schema changes. These are generated Cargo.toml and template improvements.

Workspace publishing metadata

Generated Cargo.toml files now include workspace-level metadata and enable publishing:

# Before (v1.6.0)
[package]
name = "my-app-common"
version.workspace = true
publish = false

# After (v1.6.1)
[package]
name = "my-app-common"
description = "Shared infrastructure for My App"
authors.workspace = true
documentation.workspace = true
keywords.workspace = true
categories.workspace = true
version.workspace = true
readme = "../../README.md"
publish = true

The workspace root Cargo.toml now requires a [workspace.package] section with shared metadata (authors, documentation, keywords, categories). Generated crates inherit from it.

Prompt templates

Prompt templates have been slimmed down — they now point to source files for DTOs and entities instead of inlining full definitions.

How to upgrade

Regenerate all Cargo.toml files (infrastructure and feature crates) to pick up the new metadata fields.
If publishing to crates.io, ensure your workspace root has a [workspace.package] section with proper metadata (homepage, repository, license, etc.).
No code changes required — this is purely a packaging/metadata update.

v1.5.3 to v1.6.0 — Event publishing moves to UoW layer

Qleany version: v1.5.4 through v1.6.0

What changed

No manifest schema changes. The major change is that event publishing responsibility has moved from controllers into the Unit of Work layer. All UoW factories now receive event_hub, and each use case publishes its own event after commit.

Event publishing (v1.6.0)

UoW factory constructor: Both read-only and read-write use cases now take (db_context, event_hub). Previously, read-only use cases took only (db_context).

#![allow(unused)]
fn main() {
// Before (v1.5.3) — read-only use cases
let uow_context = MyUseCaseUnitOfWorkFactory::new(db_context);

// After (v1.6.0) — all use cases
let uow_context = MyUseCaseUnitOfWorkFactory::new(db_context, event_hub);
}

UoW trait: All feature use case traits now require a publish_*_event method:

#![allow(unused)]
fn main() {
pub trait MyUseCaseUnitOfWorkTrait: QueryUnitOfWork + Send + Sync {
    fn publish_my_use_case_event(&self, ids: Vec<EntityId>, data: Option<String>);
}
}

Event publishing in use cases: The use case now calls uow.publish_*_event() after commit/end_transaction, instead of the controller calling event_hub.send_event() directly:
```
#![allow(unused)]
fn main() {
// In execute():
uow.commit()?; // or uow.end_transaction()? for read-only
uow.publish_my_use_case_event(vec![], None);
}
```
Controllers simplified: Controllers no longer contain event-sending code. The event_hub.send_event(Event { origin, ids, data }) block has been removed from controller templates.
UoW structs: All UoW structs (including read-only) now carry event_hub: Arc<EventHub>.

Float type support (v1.5.6)

Generated entities and DTOs now exclude Eq from derive traits when float fields are present. This is automatic on regeneration.

Entity test improvements (v1.5.5)

Generated entity controller tests now include ownership chain validation. No API changes.

How to upgrade

Regenerate infrastructure files (nature: Infra) to pick up the new controller and UoW templates.
If you have custom feature use cases, update:
- Change UnitOfWorkFactory::new(db_context) to UnitOfWorkFactory::new(db_context, event_hub) for read-only use cases.
- Add the publish_{use_case}_event method to your UoW trait and implementation:
```
#![allow(unused)]
fn main() {
// In the trait:
fn publish_my_use_case_event(&self, ids: Vec<EntityId>, data: Option<String>);

// In the implementation:
fn publish_my_use_case_event(&self, ids: Vec<EntityId>, data: Option<String>) {
    self.event_hub.send_event(Event {
        origin: Origin::MyFeature(MyUseCase),
        ids,
        data,
    });
}
}
```
- Move event publishing from your controller into the use case’s execute() method.
- Add event_hub: Arc<EventHub> to your UoW struct and accept it in the factory constructor.

v1.5.0 to v1.5.3 — Error handling and robustness improvements

Qleany version: v1.5.1 through v1.5.3

What changed

No manifest schema changes. These are generated code improvements that affect regenerated projects.

Error handling (v1.5.1–v1.5.2)

Transactions: get_read_transaction() and get_write_transaction() now return Result instead of panicking on wrong transaction type or consumed state. commit(), rollback(), create_savepoint(), and restore_to_savepoint() return descriptive errors instead of panicking on double-commit or missing begin_transaction().
Repository factory: Factory functions return Result, so all unit of work call sites must use ? to propagate errors. If you have custom UoW implementations, update repository creation calls from repository_factory::write::create_*_repository(transaction) to repository_factory::write::create_*_repository(transaction)?.
Undo/redo: begin_composite() now returns Result<()> instead of panicking on mismatched stack IDs. cancel_composite() now undoes any already-executed sub-commands before clearing state. Failed undo() and redo() operations re-push the command to its original stack instead of dropping it.
Table constraints: One-to-one constraint violations return RepositoryError::ConstraintViolation instead of panicking.
New error variants: RepositoryError gains ConstraintViolation(String) and Other(anyhow::Error).
Proc macros: #[macros::uow_action] with missing arguments now emits a compile error instead of panicking.
DTO enums: Enum imports in generated DTO files are now pub use instead of use, making them accessible to external crates.

Event loop and long operations (v1.5.3)

Event loop: start_event_loop now returns thread::JoinHandle<()> and uses recv_timeout(100ms) so the stop signal is checked even when no events arrive. This fixes unresponsive shutdown. (Superseded in v1.7.4 — the recv_timeout poll is gone, replaced by a true blocking flume::Selector wakeup. See the v1.7.3 → v1.7.4 entry.)
Long operations: A lock_or_recover helper handles mutex poisoning gracefully in LongOperationManager and OperationHandle, replacing all .lock().unwrap() calls.

Mobile bridge (v1.5.1)

Feature method naming: Feature use case methods now include the feature prefix (e.g., handling_manifest_save() instead of save()). Swift/Kotlin async wrappers follow suit (handlingManifestSaveAsync()).
Cross-module types: A mobile_types module re-exports entity types across command modules.
Entity conversions: From<Entity> for MobileEntityDto and reverse conversions are now generated.

How to upgrade

Regenerate your project’s infrastructure files (nature: Infra) to pick up the new error handling patterns.
If you have custom UoW implementations (feature use cases), update:
- Replace .take().unwrap() on transaction Options with .take().ok_or_else(|| anyhow!("No active transaction"))?
- Add ? after repository_factory::write::create_*_repository(...) and repository_factory::read::create_*_repository(...) calls
- Update begin_composite() call sites to handle the new Result<()> return type
If you use the mobile bridge, update Swift/Kotlin call sites to use the new feature-prefixed method names.

Cargo workspace dependencies

Generated Cargo.toml templates now use workspace-level dependency declarations. Regenerate your Cargo files to pick up this change.

Schema v4 to v5 — `is_list` for entity fields

Qleany version: v1.4.0

What changed

Entity fields now support is_list: true, the same way DTO fields already did. This allows declaring list/array fields of primitive types (string, integer, uinteger, float, boolean, uuid, datetime) directly on entities.

Constraints

is_list cannot be used with entity or enum field types.
is_list and optional are mutually exclusive on the same field.

Example

entities:
  - name: Project
    inherits_from: EntityBase
    fields:
      - name: title
        type: string
      - name: labels
        type: string
        is_list: true
      - name: scores
        type: float
        is_list: true

Automatic migration

Qleany auto-migrates v2+ manifests on load. When you open a v4 manifest, the migrator bumps the version to 5 before validation. No manual editing is required.

Manual migration

Change the schema version:

schema:
  version: 5    # was 4

No other manifest changes are needed — is_list defaults to false when omitted.

Storage

Rust: list fields are stored as Vec<T> in the entity struct, held as plain Rust types in the in-memory HashMap store.
C++/Qt: list fields are stored as QList<T> in the entity struct, serialized as JSON arrays in SQLite TEXT columns.

Schema v3 to v4

Qleany version: v1.0.31

What changed

The validator use case property has been removed.

Reasons for the change

Validation is the responsibility of the developer.

Automatic migration

Qleany auto-migrates v2+ manifests on load. When you open a v3 manifest, the migrator strips all validator fields and bumps the version to 4 before validation. No manual editing is required to load an old manifest.

If you save the manifest afterwards (from the UI), the file is written as v4.

From the CLI, it’s the same: if you run qleany generate on a v3 manifest, it will be auto-migrated to v4 before generation. To only migrate the manifest, use qleany migrate instead.

Manual migration

If you prefer to update the file yourself:

Change the schema version:

schema:
  version: 4    # was 3

Remove every validator: line from your entities:

 feature:
   - name : my_feature
     use_cases:
       - name: my_use_case
-        validator: true

No other manifest changes are needed.

Behavioral differences

None

Code generation templates

Never used.

Schema v2 to v3

Qleany version: v1.0.29

What changed

The allow_direct_access entity property has been removed. Every entity that isn’t heritage-only now always gets its direct_access/ files generated.

Reasons for the change

The direct_access/ is an internal API. allow_direct_access: true skipped generation of the files for an entity. Yet, this entity could have needed to offer a list_model or a single model, which wouldnt be possible without direct_access/ files. So, from now on, all non-heritage entities always get their direct_access/ files generated. At compilation time, unused C++ functions (static libraries) are stripped from the binary. Same for Rust. In shared C++ libraries, C++ unused functions are compiled, yet the overweight is negligible.

Automatic migration

Qleany auto-migrates v2+ manifests on load. When you open a v2 manifest, the migrator strips all allow_direct_access fields and bumps the version to 3 before validation. No manual editing is required to load an old manifest.

If you save the manifest afterwards (from the UI), the file is written as v3.

From the CLI, it’s the same: if you run qleany generate on a v2 manifest, it will be auto-migrated to v3 before generation. To only migrate the manifest, use qleany migrate instead.

Manual migration

If you prefer to update the file yourself:

Change the schema version:

schema:
  version: 3    # was 2

Remove every allow_direct_access: line from your entities:

 entities:
   - name: EntityBase
     only_for_heritage: true
-    allow_direct_access: false
     fields:
       ...

   - name: Car
     inherits_from: EntityBase
-    allow_direct_access: true
     fields:
       ...

That’s it. No other manifest changes are needed.

Behavioral differences

Before (v2)	After (v3)
`allow_direct_access: false` hid an entity from `direct_access/` generation	Use `only_for_heritage: true` instead (which also skips generation)
`allow_direct_access: true` (the default) generated files	All non-heritage entities always generate files

If you had entities with allow_direct_access: false that were not only_for_heritage: true, those entities will now generate direct_access/ files. If you don’t want that, mark them only_for_heritage: true.

Code generation templates

Tera templates that referenced ent.inner.allow_direct_access now use not ent.inner.only_for_heritage. If you’ve written custom templates that check this field, update them accordingly.

Keyboard shortcuts

Qleany Documentation