sqlglot-maxcompute

A SQLGlot dialect plugin for Alibaba Cloud MaxCompute (formerly ODPS).

Registers the maxcompute dialect via Python entry points so that SQLGlot can parse and generate MaxCompute SQL.

Installation

pip install sqlglot-maxcompute

Requires Python ≥ 3.9 and SQLGlot ≥ 30.1.

Usage

import sqlglot

# Parse MaxCompute SQL
ast = sqlglot.parse_one("SELECT DATEADD(dt, 1, 'DAY')", read="maxcompute")

# Transpile from another dialect to MaxCompute
sqlglot.transpile("SELECT DATE_ADD(dt, 1)", read="spark", write="maxcompute")
# ["SELECT DATEADD(dt, 1, 'DAY')"]

# Transpile from MaxCompute to another dialect
sqlglot.transpile("SELECT DATETRUNC(dt, 'MONTH')", read="maxcompute", write="spark")
# ["SELECT TRUNC(dt, 'MONTH')"]

# TO_DATE return type depends on args:
#   without format → DATE    (exp.TsOrDsToDate)
#   with format    → DATETIME (exp.StrToTime)
sqlglot.transpile("TO_DATE('20240101', 'yyyymmdd')", read="maxcompute", write="spark")
# ["TO_TIMESTAMP('20240101', 'yyyymmdd')"]

# Round-trip MaxCompute DDL
sqlglot.transpile(
    "CREATE TABLE t (id BIGINT) LIFECYCLE 30",
    read="maxcompute",
    write="maxcompute",
)
# ["CREATE TABLE t (id BIGINT) LIFECYCLE 30"]

What's supported

Parser — MaxCompute → canonical AST

Functions not listed are handled via Hive inheritance and work without explicit mapping (e.g. SPLIT, REGEXP_EXTRACT, COLLECT_LIST, PERCENTILE, all math/trig functions, window functions).

Generator — canonical AST → MaxCompute SQL

Explicit transforms on top of Hive:

DDL

• LIFECYCLE n — table retention in days
• RANGE CLUSTERED BY (cols) [SORTED BY (cols)] INTO n BUCKETS
• AUTO PARTITIONED BY (TRUNC_TIME(col, 'unit') [AS alias])
• TBLPROPERTIES ('key'='value') — coexists correctly with LIFECYCLE
• COMMENT on columns and tables

Development

uv sync            # install dependencies
uv run pytest      # run all tests

License

MIT