peg-parser-new-statement

Files (1)

• SKILL.md

SKILL.md (first lines)

# Add a New Statement to the PEG Parser

Add `$ARGUMENTS` to DuckDB's PEG parser. The PEG parser lives in `extension/autocomplete/` and runs as a parser override extension before the legacy Postgres parser.

**Choose your path** based on the statement form:
- **CREATE-form** (e.g. `CREATE TRIGGER`, `CREATE MATERIALIZED VIEW`) → follow Shared Steps, then Part A
- **DROP-form** (e.g. `DROP TRIGGER`) → follow Shared Steps, then Part B

---

## Shared Step 1 — Add the catalog type

In `src/include/duckdb/common/enums/catalog_type.hpp`, add:

```cpp
XXX_ENTRY = <next available number>,
```

In `src/common/enums/catalog_type.cpp`, add to both `CatalogTypeToString` and `CatalogTypeFromString`.

---

## Shared Step 2 — Add enums (if needed)

Create `src/include/duckdb/common/enums/xxx_type.hpp` for any new enums specific to this statement, then regenerate:

```bash
python3 scripts/generate_enum_util.py
```

---

## Part A: CREATE-form statements

### A1 — Define the grammar

Create `extension/autocomplete/grammar/statements/create_xxx.gram`.

Wire the new statement into `CreateStatementVariation` in `create_table.gram`:

```
CreateStatementVariation <- ... / CreateXxxStmt
```

After any `.gram` edit, regenerate (never edit `inlined_grammar.gram` or `inlined_grammar.hpp` directly):

```bash
python3 extension/autocomplete/inline_grammar.py --grammar-file
python3 extension/autocomplete/inline_grammar.py
```

See the **Grammar reference** section below for syntax rules.

---

### A2 — Create the parsed data struct

**Header**: `src/include/duckdb/parser/parsed_data/create_xxx_info.hpp`

```cpp
#pragma once
#include "duckdb/parser/parsed_data/create_info.hpp"

namespace duckdb {

struct CreateXxxInfo : public CreateInfo {
    CreateXxxInfo();

    string name;
    // ... your fields

    unique_ptr<CreateInfo> Copy() const override;
    string ToString() const override;
    DUCKDB_API void Serialize(Serializer &serializer) const override;
    DUCKDB_API static unique_ptr<CreateInfo> Deserialize(Deserializer &deserializer);
};

} // namespace duckdb
```

**Implementation**: `src/parser/parsed_data/create_xxx_info.cpp`
- Constructor: `CreateXxxInfo() : CreateInfo(CatalogType::XXX_ENTRY, INVALID_SCHEMA) {}`
- `Copy()`: call `CopyProperties(*result)` then copy each field
- `ToString()`: reconstruct the SQL string

If a field is a `unique_ptr<SQLStatement>` (not `SelectStatement`), it cannot be auto-serialized. Store the SQL text alongside it:

```cpp
string sql_body_text;          // serialized
unique_ptr<SQLStatement> sql_body; // runtime only
```

Add to `src/parser/parsed_data/CMakeLists.txt`:
```cmake
create_xxx_info.cpp
```

---

### A3 — Add serialization config

In `src/include/duckdb/storage/serialization/create_info.json`, add:

```json
{
  "class": "CreateXxxInfo",
  "base": "CreateInfo",
  "enum": "XXX_ENTRY",
  "includes": ["duckdb/parser/parsed_data/create_xxx_info.hpp"],
  "members": [
    { "id": 200, "name": "field_name", "type": "string" },
    { "id": 201, "name": "my_enum",    "type": "MyEnumType" },
    { "id": 202, "name": "my_bool",    "type": "bool" }
  ]
}
```

Supported types: `string`, `bool`, `int64_t`, `uint64_t`, `vector<string>`, `unique_ptr<SelectStatement>`, any registered enum. `SQLStatement*` is **not** supported — use `string` instead.

From mondaycom/duckdb-claude, MIT.

More from mondaycom/duckdb-claude