5 posts tagged with "ddd"

Domain-driven development breaks down when the domain language disappears from the code path.

That happens easily in data-heavy systems. The model may be discussed in design sessions, but implementation code becomes SQL strings, mapper files, repository methods, DTOs, and service glue.

TeaQL uses generated APIs to keep the domain model visible.

Model First​

TeaQL starts from a domain model:

• entities;
• scalar fields;
• object relations;
• child lists;
• constants;
• validation expectations;
• data-service targets.

The generator turns that model into APIs. Application code then works with the generated vocabulary instead of reconstructing it by hand.

Queries as Domain Language​

Generated query methods make intent explicit:

Q.orders()
    .filterByMerchant(ctx.getMerchant())
    .selectCustomer(Q.customers().selectName())
    .selectLineItemList(Q.lineItems().selectSku().selectQuantity())
    .countLineItems()
    .executeForList(ctx);

The code names the domain structure:

• orders;
• merchant;
• customer;
• line items;
• SKU;
• quantity.

That is more reviewable than a service method that hides several mapper calls and response transformations.

Relations Matter​

DDD systems are rarely flat. Relations carry meaning:

• an order has line items;
• a merchant belongs to a platform;
• an employee belongs to a department;
• a ticket has comments and assignments.

TeaQL relation loading APIs give those relationships a generated shape. The runtime can still decide how to execute the load efficiently.

Graph Writes​

Domain changes often arrive as object graphs:

• create an order with lines;
• update a parent and merge children;
• remove a child row;
• attach an existing reference;
• keep missing children for a relation that should not delete absent rows.

TeaQL Rust models this through graph nodes, graph operation state, mutation planning, and transaction boundaries. Java TeaQL carries the same high-level idea through entity save and graph-oriented runtime behavior.

Validation and Runtime Policy​

Domain logic is not only data shape. It also includes:

• required fields;
• status transitions;
• tenant rules;
• permission policy;
• audit metadata;
• language-aware validation messages.

TeaQL keeps those concerns near runtime context and generated model hooks instead of scattering them across controllers.

Why Generate?​

Handwritten domain APIs are possible. The problem is consistency.

Generation gives every entity the same baseline:

• field selectors;
• filters;
• relation selectors;
• aggregate helpers;
• request builders;
• checker hooks;
• behavior hooks;
• runtime registration.

Teams can then focus on the parts of the domain that are actually special.

The Short Version​

Generated APIs are not a shortcut around domain modeling.

They are a way to make the domain model executable, reviewable, and reusable across application code, runtime providers, and AI coding tools.

It is tempting to describe TeaQL as an ORM because TeaQL knows about entities, relations, repositories, and databases.

That description is incomplete.

TeaQL is a generated business API layer. Persistence is one part of the system, but the main value is the generated domain language that sits above persistence.

What an ORM Usually Optimizes​

Most ORM discussions focus on mapping:

• classes to tables;
• fields to columns;
• relations to joins;
• objects to rows;
• transactions to persistence sessions.

Those are real problems. TeaQL also needs to solve them. But large business systems have another repeated problem: the same business request is rebuilt again and again in controllers, repositories, DTOs, SQL, validators, and frontend response logic.

What TeaQL Optimizes​

TeaQL optimizes the business API surface.

It generates request APIs that can express:

• selection;
• nested relation loading;
• filters;
• list-existence queries;
• pagination;
• grouped statistics;
• relation aggregates;
• graph writes;
• validation hooks;
• runtime customization.

The generated API is meant to be read by backend engineers, domain engineers, reviewers, and AI coding tools.

Generated Business APIs​

An ORM might make it easy to load an Order.

TeaQL aims to make it clear how a complete order page is assembled:

Q.orders()
    .filterByMerchant(ctx.getMerchant())
    .selectCustomer(Q.customers().selectName())
    .selectLineItemList(Q.lineItems().selectSku().selectQuantity())
    .countLineItems()
    .orderByCreateTimeDescending()
    .page(1, 20)
    .executeForList(ctx);

The key is not whether this compiles to SQL. It does. The key is that the API names the business shape before the runtime turns it into storage operations.

Runtime Boundary​

TeaQL also treats runtime behavior as part of the model:

• user context;
• tenant and permission policy;
• cache behavior;
• distributed locks;
• logging and metrics;
• validation;
• event dispatch;
• database provider selection.

That is why TeaQL has a runtime layer instead of only a mapper layer.

Why This Helps AI​

AI tools should not need to guess SQL, join rules, table names, tenant filters, and response shapes from scattered code.

Generated APIs give AI tools a deterministic vocabulary:

• call this field selector;
• use this relation loader;
• compose this query fragment;
• execute through this runtime context;
• do not bypass the provider.

That is a different goal from a traditional ORM.

The Short Version​

TeaQL uses persistence mapping, but it is not defined by persistence mapping.

It is a generated business API platform that keeps domain intent visible while allowing the runtime provider to change underneath.

TeaQL is a data layer for applications where the domain model is the center of the system rather than a thin mapping over tables. The current Rust implementation carries over ideas from the Java TeaQL stack, but with a smaller scope: PostgreSQL, SQLite, a Rust-native query AST, generated typed APIs, and no web framework dependency.

The goal is direct: instead of writing most application data access as handmade repository methods or raw SQL fragments, a domain model can generate a Rust crate with entity types, relation metadata, query builders, checker hooks, behavior hooks, and graph-save entrypoints.

Application code then works with a high-level API:

let platforms = Q::platforms()
    .select_merchant_list_with(
        Q::merchants()
            .select_name()
            .which_names_contain("TeaQL"),
    )
    .execute_for_list(&ctx)
    .await?;

This is not meant to replace every way of using SQL from Rust. If a service is mostly carefully tuned SQL, direct sqlx is probably a better fit. If the preferred abstraction is an ORM with a large Rust ecosystem, Diesel or SeaORM will be more familiar. TeaQL is aimed at a different case: large domain models where repeated relation loading, graph persistence, validation, and statistics queries become their own layer of application logic.

Why Generate the API?​

The original pressure came from systems where the same entity model needed to support several kinds of behavior:

• ordinary list and detail queries;
• nested relation loading, including paths such as merchant.platform or platform.merchant_list;
• graph writes where a parent object and children are committed together;
• additive schema bootstrap for development and tests;
• checker and validation logic that can inspect and fix entities;
• simple and grouped statistics;
• JSON serialization and JSON-expression style search.

You can build all of that by hand, but the code tends to become repetitive in two places. First, relation names and field names are repeated across query methods, repository code, and validation code. Second, application developers end up switching between typed domain objects and untyped row maps. TeaQL tries to keep the generated surface typed while letting the runtime keep a generic query and graph model internally.

For example, a generated service crate exposes Q::platforms() and Q::merchants() rather than asking application code to construct SelectQuery::new("Platform") directly. Low-level query objects still exist, but they are not the normal application-level API.

Runtime Pieces​

The Rust workspace is split into small crates:

• teaql-core: values, records, entity descriptors, query AST, expressions, commands, and SmartList<T>;
• teaql-sql: SQL compilation and dialect-neutral compiled query types;
• teaql-runtime: UserContext, repository resolution, behavior hooks, checker hooks, graph writes, relation enhancement, events, and optional SQLx executors;
• teaql-macros: #[derive(TeaqlEntity)] for descriptors and typed record/entity mapping;
• teaql-provider-sqlx-postgres, teaql-provider-sqlx-sqlite, teaql-provider-sqlx-mysql, and teaql-provider-rusqlite: provider adapters and dialect-specific execution paths.

Generated crates sit above those runtime crates. A generated CRM or ERP service, for example, exports entities such as Platform and Merchant, a Q query facade, behavior skeletons, checker skeletons, repository registration, and runtime module assembly helpers.

Query Construction​

TeaQL has a generic query AST, but generated code provides a domain-specific facade. Instead of this in application code:

let query = SelectQuery::new("Merchant")
    .project("id")
    .project("name")
    .filter(Expr::contains("name", "tea"));

the generated API can expose:

let merchants = Q::merchants()
    .select_name()
    .which_names_contain("tea")
    .order_by_create_time_desc()
    .page(1, 20)
    .execute_for_list(&ctx)
    .await?;

Relation loading uses the same style:

let platforms = Q::platforms()
    .select_merchant_list_with(
        Q::merchants()
            .select_name()
            .select_platform(),
    )
    .execute_for_list(&ctx)
    .await?;

One implementation detail mattered here: when a child is attached to a parent relation list, the reverse object relation should be populated too. In the example above, each merchant in platform.merchant_list should have its platform relation set. Otherwise, the result is typed but not really a domain object graph.

Graph Writes​

TeaQL has a save_graph path for committing complex objects. In generated crates, application code can call a typed save helper:

merchant
    .update_name("TeaQL Merchant")
    .update_platform_id(1_u64)
    .save(&ctx)
    .await?;

Internally, the runtime turns that object into a graph plan. The plan classifies nodes by entity and operation: create, update, delete/remove, or reference. It then batches compatible work where possible and runs the graph write inside a transactional executor.

The interesting part is not only inserting children. Updating a graph means answering questions like:

• is this child new, already present, a reference, or explicitly removed?
• should missing children be soft-deleted or left alone?
• should a relation write attach a foreign key, or is it detached?
• if a reference points at a deleted row or the wrong version, should the graph write fail?

The Rust runtime currently supports nested create/update graph writes, reference-only nodes, explicit remove nodes, keep-missing relation metadata, duplicate child-id rejection, and transaction rollback for SQLite and PostgreSQL SQLx executors.

Schema Bootstrap​

For local development and generated-service tests, TeaQL can bootstrap a schema from entity descriptors:

ctx.ensure_sqlite_schema().await?;

The current scope is intentionally conservative. It creates missing tables and adds missing columns. It does not try to be a destructive migration tool: no column drops, no primary-key rebuilds, and no automatic type rewrites.

That line is important because generated domain models change frequently. The bootstrap path should be safe enough for local and CI use, not pretend to replace a real production migration process.

Checkers and Domain Validation​

A checker is not just a validator that rejects a row. It can inspect an object, add structured check results, and sometimes fix fields before persistence.

The generated Rust checker support lets application code write typed checker logic instead of manually reading from a Record:

impl MerchantCheckerLogic for MerchantNameChecker {
    fn check_and_fix_merchant(
        &self,
        _ctx: &UserContext,
        entity: &mut Merchant,
        status: CheckObjectStatus,
        location: &ObjectLocation,
        results: &mut CheckResults,
    ) {
        if status.is_create() {
            self.required_text(&entity.name(), "name", location, results);
        }

        self.min_string_length(&entity.name(), "name", 3, location, results);

        if entity.name() == "fix" {
            entity.update_name("fixed");
        }
    }
}

The runtime still stores the common checker interface at the record level, but the generated adapter maps records into typed entities before calling the checker. That keeps the public application code close to the domain model while preserving a generic runtime path.

Statistics​

TeaQL queries can carry aggregate projections and relation aggregate metadata. The current runtime supports simple aggregates, grouped aggregates, Decimal results for SQL aggregate output, relation count/statistic attachment, and database-column-to-entity-property mapping for relation aggregate keys.

The generated Q APIs can express both simple statistics and relation statistics. For example, a service can count child rows from a parent query without asking application code to hand-build the join every time.

This area is useful but still evolving. The runtime has working SQL and memory paths for the core cases, while broader Java parity still needs more work around memory subqueries and richer relation aggregate shapes.

Tradeoffs​

The most obvious tradeoff is generated code. TeaQL generates a lot of Rust. That cost shows up as compile time, larger diffs, and the need to keep templates disciplined. The benefit is that application code gets a stable, typed facade over a large domain model.

Another tradeoff is that the runtime is not purely compile-time checked. The generated APIs are typed, but the runtime still has a generic query AST, record model, and descriptor registry. That gives it flexibility for dynamic projections, aggregate rows, JSON-style search, and graph planning, but it means some mistakes are caught by generated crate tests rather than by Rust types alone.

The final tradeoff is scope. The Rust rewrite is not trying to clone every Java TeaQL feature or support every database. PostgreSQL and SQLite are enough for now. Web rendering, GraphQL integration, and broad database dialect support are outside the current Rust scope.

What Works Today​

The current Rust runtime and generated crate tests cover:

• SQLite schema bootstrap and additive column changes;
• PostgreSQL schema bootstrap with SQLx;
• CRUD, optimistic locking, soft delete, and recover;
• typed entity fetch into SmartList<T>;
• nested relation enhancement;
• complex object commit through graph writes;
• transaction rollback for graph writes;
• generated Q APIs against SQLite;
• typed checker adapters from generated crates;
• JSON serialization and JSON-expression search paths;
• simple aggregates, grouped aggregates, and relation aggregate statistics.

The public examples can be run with:

cargo run -p teaql-examples --bin sqlite_schema_crud
cargo run -p teaql-examples --bin sqlite_relations_graph

The first command shows schema bootstrap and CRUD against in-memory SQLite. The second saves an object graph and reloads nested relations.

What Is Not Done​

The biggest gaps are:

• more complete memory repository parity for relation enhancement and subquery execution;
• richer checker semantics, especially nested typed object locations and domain-specific labels;
• richer event payloads with old/new values and typed snapshots;
• more value types such as UUID and bytes;
• a decision on whether Rust needs a higher-level service layer above the repository/runtime APIs.

Those gaps are real. They are better kept visible than hidden behind a larger feature list.

Why This Shape Is Worth Exploring​

Most Rust database libraries are good at one of two layers: explicit SQL, or a database-centric ORM. TeaQL explores a third shape: generated domain APIs over a generic runtime that understands entity graphs, relation enhancement, validation, and statistics.

That shape will not fit every codebase. It is most useful when the model is large enough that the generated API becomes an asset, and when the team wants the same domain semantics to appear in queries, graph writes, checkers, and schema bootstrap.

TeaQL generates query APIs such as Q::platforms() and Q::merchants(). That immediately raises a fair question: why not just write SQL?

The short answer is that TeaQL is not trying to replace handwritten SQL everywhere. Generated query APIs are useful when the repeated work is not the SQL text itself, but the domain semantics around the SQL: field names, relation names, relation loading, statistics, validation, JSON search, and graph-shaped results.

The Handwritten SQL Case​

There are many cases where handwritten SQL is the right tool:

• a query is performance-critical and carefully tuned;
• the shape is reporting-specific rather than domain-object-specific;
• the team wants exact control over joins, hints, indexes, and execution plans;
• the service is small enough that generated APIs would add more weight than value.

TeaQL should not get in the way of those cases. The Rust runtime keeps a generic query AST and SQL compiler, but the design does not require every query to be expressed through generated methods.

If explicit SQL is the clearest abstraction, use explicit SQL.

The Generated API Case​

Generated APIs become useful when a large domain model produces many ordinary queries with the same repeated structure:

let merchants = Q::merchants()
    .select_name()
    .select_platform_with(Q::platforms().select_name())
    .which_names_contain("tea")
    .order_by_create_time_desc()
    .page(1, 20)
    .execute_for_list(&ctx)
    .await?;

This is not shorter than SQL in every case. The value is that the query is tied to the generated domain model:

• fields are exposed as methods;
• object relations are selected by relation name, not storage column name;
• readable filters follow the model vocabulary;
• subqueries can be constructed from the target entity's own generated API;
• execution returns typed entities or typed SmartList<T> collections.

That matters when the model is large and the same entity relationships appear across many services, pages, and business workflows.

Relation Semantics​

A generated API can encode relation semantics that plain SQL does not name.

For example:

let platforms = Q::platforms()
    .select_merchant_list_with(
        Q::merchants()
            .select_name()
            .select_platform(),
    )
    .execute_for_list(&ctx)
    .await?;

The application is not just asking for a join. It is asking for a platform list where each platform contains a merchant_list, and each merchant can carry its reverse platform relation when selected. That object shape is a domain result, not only a SQL result set.

The runtime still compiles to SQL and fetches records. The generated API gives the caller a domain-oriented way to request the shape.

Subqueries Without Stringly-Typed Entity Names​

TeaQL still has SelectQuery internally. Application code can use it directly, but generated crates should prefer the entity-specific Q entrypoints:

let platforms = Q::platforms()
    .select_merchant_list_with(
        Q::merchants()
            .which_names_are("TeaQL Merchant"),
    )
    .execute_for_list(&ctx)
    .await?;

This keeps the application out of string-based construction such as SelectQuery::new("Merchant") for ordinary code. The generic query type remains available for runtime and escape-hatch scenarios.

Statistics​

Large business systems rarely stop at list queries. They also need counts, sums, grouped aggregates, relation counts, and derived summary rows.

A generated API can give those operations domain names:

let platforms = Q::platforms()
    .select_name()
    .count_merchant_list_as("merchant_count")
    .execute_for_list(&ctx)
    .await?;

The runtime can attach relation aggregate results back to parent rows, while the generated request builder keeps the caller focused on domain concepts.

This does not remove the need for handwritten reporting SQL. It gives ordinary domain statistics a consistent path.

Why Keep a Generic Runtime?​

If generated APIs are useful, why not make the whole runtime typed?

Because TeaQL still needs dynamic behavior:

• generated service crates vary by domain model;
• aggregate rows can contain dynamic projection fields;
• JSON-expression search and raw SQL escape hatches need flexible query metadata;
• graph planning operates across entity types;
• relation enhancement starts with records and then maps into typed entities.

The generated layer is typed for application ergonomics. The runtime layer is generic so it can plan, compile, enhance, validate, and execute across many generated models.

Where Generated APIs Hurt​

Generated code is not free:

• it increases compile time;
• it creates large diffs when templates change;
• it can hide SQL details if developers treat it as magic;
• it needs strong generated-crate tests to catch template regressions.

TeaQL tries to offset that with explicit generated code, local runtime crates, and generated service tests that run real SQLite scenarios. The code is not hidden behind a server or reflection system; it is Rust source that can be inspected.

Still, generated APIs should be used where they buy something. A small service with ten queries may be better served by direct sqlx.

A Practical Boundary​

The boundary I use is this:

• if the query expresses a domain object shape, relation graph, validation path, or repeated business filter, prefer the generated Q API;
• if the query expresses a one-off report, hand-tuned performance path, or database-specific operation, prefer handwritten SQL;
• if generated code starts hiding a critical execution detail, drop down a level.

TeaQL's design only works if the lower level remains available.

The Real Goal​

The goal is not to win an argument against SQL. SQL remains the execution language and often the best authoring language.

The goal is to avoid making every application workflow manually rediscover the same domain relationships. In a large model, the generated API becomes a shared vocabulary: Q::platforms(), select_merchant_list_with(...), have_platform(), count_merchant_list_as(...), and so on.

That vocabulary is where TeaQL is useful. Not because SQL is bad, but because large domain models need more than SQL strings to stay coherent.

Object graph persistence looks simple until the first update path arrives. A new parent with new children is close to a recursive insert. A real business object graph is not.

TeaQL's save_graph runtime exists because the runtime needs to understand entity identity, relation ownership, reference semantics, missing children, soft delete, and transaction boundaries at the same time.

The Easy Case​

The simple case is a parent object with children:

let order = Order {
    id: 1,
    version: 1,
    name: "graph-order".to_owned(),
    lines: SmartList::from(vec![
        OrderLine {
            id: 10,
            order_id: 0,
            name: "first-line".to_owned(),
            product_id: 100,
            product: Some(Product {
                id: 100,
                name: "keyboard".to_owned(),
            }),
        },
    ]),
};

repo.save_entity_graph(order)?;

The runtime can create the product, create the order, attach the line to the order, and create the line. If that were the whole problem, graph save would be a convenience wrapper over insert.

The harder cases are the reason TeaQL treats graph persistence as a plan.

Identity and State​

When a graph reaches the runtime, each node has to be classified:

• Create: the row does not exist yet and should be inserted;
• Update: the row exists and selected fields should be changed;
• Reference: the row must exist, but the graph should not mutate it;
• Remove: the row is explicitly removed from a relation or graph.

Those states are not interchangeable. A referenced row should fail if it does not exist. A removed row should not silently become an update. A create with a duplicate id should not be treated as a harmless no-op.

TeaQL exposes this through graph operation hints and validates the state against the repository when needed.

Relation Ownership​

An object relation may attach a foreign key, or it may be detached. A one-to-many relation may delete missing children, or it may keep missing children untouched.

That relation metadata matters during updates:

#[teaql(relation(
    target = "OrderLine",
    local_key = "id",
    foreign_key = "order_id",
    many,
    delete_missing = false
))]
lines: SmartList<OrderLine>,

With delete_missing = false, sending an order with one line does not mean all other order lines should be deleted. Without that metadata, a graph update can look correct in tests and still be dangerous in production.

Attach metadata is similar. Some relations express ownership and should write a foreign key. Others are navigational and should not.

Missing Children​

The most common graph update question is what to do with children that were in the database but are missing from the submitted graph.

There are several possible meanings:

• the caller did not load that child and is not trying to change it;
• the caller loaded the full relation and intentionally removed that child;
• the relation is a projection and should not imply ownership;
• the child should be soft-deleted rather than physically deleted.

TeaQL avoids guessing from the shape alone. Relation descriptors carry the policy. The graph planner uses that policy to decide whether to keep missing rows or plan a remove/soft-delete operation.

Planning Before Execution​

The runtime builds a mutation plan before executing:

Order:
  update: [id=1, fields=name]

OrderLine:
  create: [id=12]
  update: [id=10, fields=name]
  delete: [id=11]

Product:
  reference: [id=100]

This plan is useful for three reasons.

First, it lets TeaQL validate relation semantics before making changes.

Second, it allows compatible mutations to be batched by entity type and updated field set.

Third, it gives developers a debugging surface. UserContext::plan_for_save_graph() can be used to inspect what the runtime thinks it is going to do.

Transactions Are Part of the Feature​

A graph write that touches multiple tables has to be transactional. If a child insert fails after the parent update succeeds, the application now has a partial business object.

TeaQL requires graph writes to go through a transactional executor. SQLite and PostgreSQL SQLx paths both have transaction-boundary support. The test suite verifies rollback behavior for graph writes.

That requirement is deliberately strict. A graph write without a transaction is usually worse than no graph-write helper at all.

Typed Entity Graphs​

Generated service crates should not force application code to build raw graph nodes by hand. The current generated API can work with typed entities and turn them into graph nodes internally:

merchant
    .update_name("TeaQL Merchant")
    .update_platform_id(1_u64)
    .save(&ctx)
    .await?;

The runtime still uses a generic graph representation after extraction. That is intentional. The typed layer is for application ergonomics; the generic layer is for planning, batching, validation, and execution.

What This Is Not​

This is not a general object database. The database remains relational. TeaQL does not try to hide SQL or table design completely.

It is also not a replacement for explicit migrations. Graph writes operate against descriptors and repositories. Schema evolution is handled separately by the additive ensure_schema path for local development and tests, or by a proper migration process in production.

Why It Matters​

In small systems, graph save can be overkill. In large business systems, object graphs are often the natural unit of change: an order and its lines, a platform and its merchants, a service contract and its documents.

The hard part is not creating the first version. The hard part is updating the second version without losing data, silently mutating references, or leaving a partial graph after a failure.

That is why TeaQL treats object graph persistence as planning first and execution second.