df-dev - SKILL.md Agent Skill

name: df-dev description: Develop dialect features for datafusion-sqlparser-rs from the upstream contribution list or a description.

DataFusion SQLParser Dev

Usage

/df-dev                  # auto-pick next unchecked item from the TODO below
/df-dev <text description>

TODO

MSSQL (2 remaining)

TRY/CATCH — BEGIN TRY...END TRY BEGIN CATCH...END CATCH
WHILE — standalone statement and cursor loop support
CURSOR — DECLARE CURSOR FOR, OPEN, FETCH, CLOSE, DEALLOCATE
THROW statement
WAITFOR statement
OUTPUT clause on INSERT/UPDATE/DELETE
TRAN shorthand for TRANSACTION
Multiple statements without semicolons

ClickHouse (8 remaining)

PARTITION BY / SAMPLE BY / TTL in CREATE TABLE
ARRAY JOIN
Distributed table syntax
Materialized views with ENGINE / POPULATE
CREATE DICTIONARY
ALTER TABLE UPDATE / ALTER TABLE DELETE
SYSTEM commands
ClickHouse-specific WITH clause

Redshift (4 remaining)

DISTKEY / SORTKEY / DISTSTYLE on CREATE TABLE
ENCODE column compression
COPY with Redshift options
ANALYZE statement
CREATE EXTERNAL SCHEMA

Hive (6 remaining)

Complex types: ARRAY<>, MAP<>, STRUCT<>
TRANSFORM clause
ADD COLUMNS with parenthesized syntax
SET LOCATION on ALTER TABLE
Hive-style CREATE INDEX
DESCRIBE DATABASE

Snowflake (3 remaining)

PUT / GET file staging
CREATE TASK / ALTER TASK
CREATE STREAM

SQLite (4 remaining)

GLOB operator
PRAGMA statements
DETACH DATABASE
ANALYZE

DuckDB (2 remaining)

Lambda functions (list_filter(lambda x : x > 4))
SAMPLE clause
ASOF JOIN

PostgreSQL (1 remaining)

ANALYZE with column specification
VACUUM with options

General (2 remaining)

UPDATE ... ORDER BY (MySQL extension)
CREATE TABLE AS with explicit column list

Instructions

Understand the task: If no argument given, pick the next unchecked item from the TODO above. Otherwise, match the request to an item if applicable.
Follow /dev-autodev loop with these project-specific details:
- Implement:
  - Add or modify relevant files under src/
  - Reuse existing parser infrastructure (e.g., parse_begin_exception_end())
  - Add tests in tests/sqlparser_<dialect>.rs
  - For dialect-specific syntax, include a comment with the official doc link
  - AST convention (issue #1204): wrap new statements in a named struct, not inline enum fields.
```
// Good: named struct
pub struct Throw { pub error_number: Option<Expr>, ... }
Statement::Throw(Throw)

// Bad: inline fields
Statement::Throw { error_number: Option<Expr>, ... }
```
  - Parser convention: parse functions should be standalone (parse the full statement including its keyword) and return the struct, not Statement. In the keyword dispatch, rewind the token first:
```
Keyword::THROW => {
    self.prev_token();
    self.parse_throw().map(Into::into)
},
```
  - No early-return for empty input: do not add short-circuit checks for semicolons or EOF at the start of parse functions. Let the function error naturally on incomplete input.
- Verify:
```
cargo test
cargo clippy --all-targets --all-features -- -D warnings
cargo fmt --all -- --check
```
- Update this skill's TODO: Mark completed items with [x] and update the remaining count.
- PR title must include the dialect name prefix (e.g., MSSQL: Add support for ...).