782 lines
22 KiB
Python
782 lines
22 KiB
Python
# encoding=utf8
|
||
"""芋道系统数据库迁移工具
|
||
|
||
Author: dhb52 (https://gitee.com/dhb52)
|
||
|
||
pip install simple-ddl-parser
|
||
"""
|
||
|
||
import argparse
|
||
import pathlib
|
||
import re
|
||
import time
|
||
from abc import ABC, abstractmethod
|
||
from typing import Dict, Generator, Optional, Tuple, Union
|
||
|
||
from simple_ddl_parser import DDLParser
|
||
|
||
PREAMBLE = """/*
|
||
Yudao Database Transfer Tool
|
||
|
||
Source Server Type : MySQL
|
||
|
||
Target Server Type : {db_type}
|
||
|
||
Date: {date}
|
||
*/
|
||
|
||
"""
|
||
|
||
|
||
def load_and_clean(sql_file: str) -> str:
|
||
"""加载源 SQL 文件,并清理内容方便下一步 ddl 解析
|
||
|
||
Args:
|
||
sql_file (str): sql文件路径
|
||
|
||
Returns:
|
||
str: 清理后的sql文件内容
|
||
"""
|
||
REPLACE_PAIR_LIST = (
|
||
(" CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci ", " "),
|
||
(" KEY `", " INDEX `"),
|
||
("UNIQUE INDEX", "UNIQUE KEY"),
|
||
("b'0'", "'0'"),
|
||
("b'1'", "'1'"),
|
||
)
|
||
|
||
content = open(sql_file).read()
|
||
for replace_pair in REPLACE_PAIR_LIST:
|
||
content = content.replace(*replace_pair)
|
||
content = re.sub(r"ENGINE.*COMMENT", "COMMENT", content)
|
||
content = re.sub(r"ENGINE.*;", ";", content)
|
||
return content
|
||
|
||
|
||
class Convertor(ABC):
|
||
def __init__(self, src: str, db_type) -> None:
|
||
self.src = src
|
||
self.db_type = db_type
|
||
self.content = load_and_clean(self.src)
|
||
self.table_script_list = re.findall(r"CREATE TABLE [^;]*;", self.content)
|
||
|
||
@abstractmethod
|
||
def translate_type(self, type: str, size: Optional[Union[int, Tuple[int]]]) -> str:
|
||
"""字段类型转换
|
||
|
||
Args:
|
||
type (str): 字段类型
|
||
size (Optional[Union[int, Tuple[int]]]): 字段长度描述, 如varchar(255), decimal(10,2)
|
||
|
||
Returns:
|
||
str: 类型定义
|
||
"""
|
||
pass
|
||
|
||
@abstractmethod
|
||
def gen_create(self, table_ddl: Dict) -> str:
|
||
"""生成 create 脚本
|
||
|
||
Args:
|
||
table_ddl (Dict): 表DDL
|
||
|
||
Returns:
|
||
str: 生成脚本
|
||
"""
|
||
pass
|
||
|
||
@abstractmethod
|
||
def gen_pk(self, table_name: str) -> str:
|
||
"""生成主键定义
|
||
|
||
Args:
|
||
table_name (str): 表名
|
||
|
||
Returns:
|
||
str: 生成脚本
|
||
"""
|
||
pass
|
||
|
||
@abstractmethod
|
||
def gen_index(self, ddl: Dict) -> str:
|
||
"""生成索引定义
|
||
|
||
Args:
|
||
table_ddl (Dict): 表DDL
|
||
|
||
Returns:
|
||
str: 生成脚本
|
||
"""
|
||
pass
|
||
|
||
@abstractmethod
|
||
def gen_comment(self, table_sql: str, table_name: str) -> str:
|
||
"""生成字段/表注释
|
||
|
||
Args:
|
||
table_sql (str): 原始表SQL
|
||
table_name (str): 表名
|
||
|
||
Returns:
|
||
str: 生成脚本
|
||
"""
|
||
pass
|
||
|
||
@abstractmethod
|
||
def gen_insert(self, table_name: str) -> str:
|
||
"""生成 insert 语句块
|
||
|
||
Args:
|
||
table_name (str): 表名
|
||
|
||
Returns:
|
||
str: 生成脚本
|
||
"""
|
||
pass
|
||
|
||
def gen_dual(self) -> str:
|
||
"""生成虚拟 dual 表
|
||
|
||
Returns:
|
||
str: 生成脚本, 默认返回空脚本, 表示当前数据库无需手工创建
|
||
"""
|
||
return ""
|
||
|
||
@staticmethod
|
||
def inserts(table_name: str, script_content: str) -> Generator:
|
||
PREFIX = f"INSERT INTO `{table_name}`"
|
||
|
||
# 收集 `table_name` 对应的 insert 语句
|
||
for line in script_content.split("\n"):
|
||
if line.startswith(PREFIX):
|
||
head, tail = line.replace(PREFIX, "").split(" VALUES ", maxsplit=1)
|
||
head = head.strip().replace("`", "").lower()
|
||
tail = tail.strip().replace(r"\"", '"')
|
||
# tail = tail.replace("b'0'", "'0'").replace("b'1'", "'1'")
|
||
yield f"INSERT INTO {table_name.lower()} {head} VALUES {tail}"
|
||
|
||
@staticmethod
|
||
def index(ddl: Dict) -> Generator:
|
||
"""生成索引定义
|
||
|
||
Args:
|
||
ddl (Dict): 表DDL
|
||
|
||
Yields:
|
||
Generator[str]: create index 语句
|
||
"""
|
||
|
||
def generate_columns(columns):
|
||
keys = [
|
||
f"{col['name'].lower()}{' ' + col['order'].lower() if col['order'] != 'ASC' else ''}"
|
||
for col in columns[0]
|
||
]
|
||
return ", ".join(keys)
|
||
|
||
for no, index in enumerate(ddl["index"], 1):
|
||
columns = generate_columns(index["columns"])
|
||
table_name = ddl["table_name"].lower()
|
||
yield f"CREATE INDEX idx_{table_name}_{no:02d} ON {table_name} ({columns})"
|
||
|
||
@staticmethod
|
||
def filed_comments(table_sql: str) -> Generator:
|
||
for line in table_sql.split("\n"):
|
||
match = re.match(r"^`([^`]+)`.* COMMENT '([^']+)'", line.strip())
|
||
if match:
|
||
field = match.group(1)
|
||
comment_string = match.group(2).replace("\\n", "\n")
|
||
yield field, comment_string
|
||
|
||
def table_comment(self, table_sql: str) -> str:
|
||
match = re.search(r"COMMENT \= '([^']+)';", table_sql)
|
||
return match.group(1) if match else None
|
||
|
||
def print(self):
|
||
"""打印转换后的sql脚本到终端"""
|
||
print(
|
||
PREAMBLE.format(
|
||
db_type=self.db_type,
|
||
date=time.strftime("%Y-%m-%d %H:%M:%S"),
|
||
)
|
||
)
|
||
|
||
dual = self.gen_dual()
|
||
if dual:
|
||
print(
|
||
f"""-- ----------------------------
|
||
-- Table structure for dual
|
||
-- ----------------------------
|
||
{dual}
|
||
|
||
"""
|
||
)
|
||
|
||
error_scripts = []
|
||
for table_sql in self.table_script_list:
|
||
ddl = DDLParser(table_sql.replace("`", "")).run()
|
||
|
||
# 如果parse失败, 需要跟进
|
||
if len(ddl) == 0:
|
||
error_scripts.append(table_sql)
|
||
continue
|
||
|
||
table_ddl = ddl[0]
|
||
table_name = table_ddl["table_name"]
|
||
|
||
# 忽略 quartz 的内容
|
||
if table_name.lower().startswith("qrtz"):
|
||
continue
|
||
|
||
# 为每个表生成个5个基本部分
|
||
create = self.gen_create(table_ddl)
|
||
pk = self.gen_pk(table_name)
|
||
index = self.gen_index(table_ddl)
|
||
comment = self.gen_comment(table_sql, table_name)
|
||
inserts = self.gen_insert(table_name)
|
||
|
||
# 组合当前表的DDL脚本
|
||
script = f"""{create}
|
||
|
||
{pk}
|
||
|
||
{index}
|
||
|
||
{comment}
|
||
|
||
{inserts}
|
||
"""
|
||
|
||
# 清理
|
||
script = re.sub("\n{3,}", "\n\n", script).strip() + "\n"
|
||
|
||
print(script)
|
||
|
||
# 将parse失败的脚本打印出来
|
||
if error_scripts:
|
||
for script in error_scripts:
|
||
print(script)
|
||
|
||
|
||
class PostgreSQLConvertor(Convertor):
|
||
def __init__(self, src):
|
||
super().__init__(src, "PostgreSQL")
|
||
|
||
def translate_type(self, type: str, size: Optional[Union[int, Tuple[int]]]):
|
||
"""类型转换"""
|
||
|
||
type = type.lower()
|
||
|
||
if type == "varchar":
|
||
return f"varchar({size})"
|
||
if type == "int":
|
||
return "int4"
|
||
if type == "bigint" or type == "bigint unsigned":
|
||
return "int8"
|
||
if type == "datetime":
|
||
return "timestamp"
|
||
if type == "bit":
|
||
return "bool"
|
||
if type in ("tinyint", "smallint"):
|
||
return "int2"
|
||
if type == "text":
|
||
return "text"
|
||
if type in ("blob", "mediumblob"):
|
||
return "bytea"
|
||
if type == "decimal":
|
||
return (
|
||
f"numeric({','.join(str(s) for s in size)})" if len(size) else "numeric"
|
||
)
|
||
|
||
def gen_create(self, ddl: Dict) -> str:
|
||
"""生成 create"""
|
||
|
||
def _generate_column(col):
|
||
name = col["name"].lower()
|
||
if name == "deleted":
|
||
return "deleted int2 NOT NULL DEFAULT 0"
|
||
|
||
type = col["type"].lower()
|
||
full_type = self.translate_type(type, col["size"])
|
||
nullable = "NULL" if col["nullable"] else "NOT NULL"
|
||
default = f"DEFAULT {col['default']}" if col["default"] is not None else ""
|
||
return f"{name} {full_type} {nullable} {default}"
|
||
|
||
table_name = ddl["table_name"].lower()
|
||
columns = [f"{_generate_column(col).strip()}" for col in ddl["columns"]]
|
||
filed_def_list = ",\n ".join(columns)
|
||
script = f"""-- ----------------------------
|
||
-- Table structure for {table_name}
|
||
-- ----------------------------
|
||
DROP TABLE IF EXISTS {table_name};
|
||
CREATE TABLE {table_name} (
|
||
{filed_def_list}
|
||
);"""
|
||
|
||
return script
|
||
|
||
def gen_index(self, ddl: Dict) -> str:
|
||
return "\n".join(f"{script};" for script in self.index(ddl))
|
||
|
||
def gen_comment(self, table_sql: str, table_name: str) -> str:
|
||
"""生成字段及表的注释"""
|
||
|
||
script = ""
|
||
for field, comment_string in self.filed_comments(table_sql):
|
||
script += (
|
||
f"COMMENT ON COLUMN {table_name}.{field} IS '{comment_string}';" + "\n"
|
||
)
|
||
|
||
table_comment = self.table_comment(table_sql)
|
||
if table_comment:
|
||
script += f"COMMENT ON TABLE {table_name} IS '{table_comment}';\n"
|
||
|
||
return script
|
||
|
||
def gen_pk(self, table_name) -> str:
|
||
"""生成主键定义"""
|
||
return f"ALTER TABLE {table_name} ADD CONSTRAINT pk_{table_name} PRIMARY KEY (id);\n"
|
||
|
||
def gen_insert(self, table_name: str) -> str:
|
||
"""生成 insert 语句,以及根据最后的 insert id+1 生成 Sequence"""
|
||
|
||
inserts = list(Convertor.inserts(table_name, self.content))
|
||
## 生成 insert 脚本
|
||
script = ""
|
||
last_id = 0
|
||
if inserts:
|
||
inserts_lines = "\n".join(inserts)
|
||
script += f"""\n\n-- ----------------------------
|
||
-- Records of {table_name.lower()}
|
||
-- ----------------------------
|
||
-- @formatter:off
|
||
BEGIN;
|
||
{inserts_lines}
|
||
COMMIT;
|
||
-- @formatter:on"""
|
||
match = re.search(r"VALUES \((\d+),", inserts[-1])
|
||
if match:
|
||
last_id = int(match.group(1))
|
||
|
||
# 生成 Sequence
|
||
script += (
|
||
"\n\n"
|
||
+ f"""DROP SEQUENCE IF EXISTS {table_name}_seq;
|
||
CREATE SEQUENCE {table_name}_seq
|
||
START {last_id + 1};"""
|
||
)
|
||
|
||
return script
|
||
|
||
def gen_dual(self) -> str:
|
||
return """DROP TABLE IF EXISTS dual;
|
||
CREATE TABLE dual
|
||
(
|
||
);"""
|
||
|
||
|
||
class OracleConvertor(Convertor):
|
||
def __init__(self, src):
|
||
super().__init__(src, "Oracle")
|
||
|
||
def translate_type(self, type: str, size: Optional[Union[int, Tuple[int]]]):
|
||
"""类型转换"""
|
||
type = type.lower()
|
||
|
||
if type == "varchar":
|
||
return f"varchar2({size if size < 4000 else 4000})"
|
||
if type == "int":
|
||
return "number"
|
||
if type == "bigint" or type == "bigint unsigned":
|
||
return "number"
|
||
if type == "datetime":
|
||
return "date"
|
||
if type == "bit":
|
||
return "number(1,0)"
|
||
if type in ("tinyint", "smallint"):
|
||
return "smallint"
|
||
if type == "text":
|
||
return "clob"
|
||
if type in ("blob", "mediumblob"):
|
||
return "blob"
|
||
if type == "decimal":
|
||
return (
|
||
f"number({','.join(str(s) for s in size)})" if len(size) else "number"
|
||
)
|
||
|
||
def gen_create(self, ddl) -> str:
|
||
"""生成 CREATE 语句"""
|
||
|
||
def generate_column(col):
|
||
name = col["name"].lower()
|
||
if name == "deleted":
|
||
return "deleted number(1,0) DEFAULT 0 NOT NULL"
|
||
|
||
type = col["type"].lower()
|
||
full_type = self.translate_type(type, col["size"])
|
||
nullable = "NULL" if col["nullable"] else "NOT NULL"
|
||
default = f"DEFAULT {col['default']}" if col["default"] is not None else ""
|
||
# Oracle 中 size 不能作为字段名
|
||
field_name = '"size"' if name == "size" else name
|
||
# Oracle DEFAULT 定义在 NULLABLE 之前
|
||
return f"{field_name} {full_type} {default} {nullable}"
|
||
|
||
table_name = ddl["table_name"].lower()
|
||
columns = [f"{generate_column(col).strip()}" for col in ddl["columns"]]
|
||
field_def_list = ",\n ".join(columns)
|
||
script = f"""-- ----------------------------
|
||
-- Table structure for {table_name}
|
||
-- ----------------------------
|
||
CREATE TABLE {table_name} (
|
||
{field_def_list}
|
||
);"""
|
||
|
||
# oracle INSERT '' 不能通过 NOT NULL 校验
|
||
script = script.replace("DEFAULT '' NOT NULL", "DEFAULT '' NULL")
|
||
|
||
return script
|
||
|
||
def gen_index(self, ddl: Dict) -> str:
|
||
return "\n".join(f"{script};" for script in self.index(ddl))
|
||
|
||
def gen_comment(self, table_sql: str, table_name: str) -> str:
|
||
script = ""
|
||
for field, comment_string in self.filed_comments(table_sql):
|
||
script += (
|
||
f"COMMENT ON COLUMN {table_name}.{field} IS '{comment_string}';" + "\n"
|
||
)
|
||
|
||
table_comment = self.table_comment(table_sql)
|
||
if table_comment:
|
||
script += f"COMMENT ON TABLE {table_name} IS '{table_comment}';\n"
|
||
|
||
return script
|
||
|
||
def gen_pk(self, table_name: str) -> str:
|
||
"""生成主键定义"""
|
||
return f"ALTER TABLE {table_name} ADD CONSTRAINT pk_{table_name} PRIMARY KEY (id);\n"
|
||
|
||
def gen_index(self, ddl: Dict) -> str:
|
||
return "\n".join(f"{script};" for script in self.index(ddl))
|
||
|
||
def gen_insert(self, table_name: str) -> str:
|
||
"""拷贝 INSERT 语句"""
|
||
inserts = []
|
||
for insert_script in Convertor.inserts(table_name, self.content):
|
||
# 对日期数据添加 TO_DATE 转换
|
||
insert_script = re.sub(
|
||
r"('\d{4}-\d{2}-\d{2} \d{2}:\d{2}:\d{2}')",
|
||
r"to_date(\g<1>, 'SYYYY-MM-DD HH24:MI:SS')",
|
||
insert_script,
|
||
)
|
||
inserts.append(insert_script)
|
||
|
||
## 生成 insert 脚本
|
||
script = ""
|
||
last_id = 0
|
||
if inserts:
|
||
inserts_lines = "\n".join(inserts)
|
||
script += f"""\n\n-- ----------------------------
|
||
-- Records of {table_name.lower()}
|
||
-- ----------------------------
|
||
-- @formatter:off
|
||
{inserts_lines}
|
||
COMMIT;
|
||
-- @formatter:on"""
|
||
match = re.search(r"VALUES \((\d+),", inserts[-1])
|
||
if match:
|
||
last_id = int(match.group(1))
|
||
|
||
# 生成 Sequence
|
||
script += f"""
|
||
|
||
CREATE SEQUENCE {table_name}_seq
|
||
START WITH {last_id + 1};"""
|
||
|
||
return script
|
||
|
||
|
||
class SQLServerConvertor(Convertor):
|
||
"""_summary_
|
||
|
||
Args:
|
||
Convertor (_type_): _description_
|
||
"""
|
||
|
||
def __init__(self, src):
|
||
super().__init__(src, "Microsoft SQL Server")
|
||
|
||
def translate_type(self, type: str, size: Optional[Union[int, Tuple[int]]]):
|
||
"""类型转换"""
|
||
|
||
type = type.lower()
|
||
|
||
if type == "varchar":
|
||
return f"nvarchar({size if size < 4000 else 4000})"
|
||
if type == "int":
|
||
return "int"
|
||
if type == "bigint" or type == "bigint unsigned":
|
||
return "bigint"
|
||
if type == "datetime":
|
||
return "datetime2"
|
||
if type == "bit":
|
||
return "varchar(1)"
|
||
if type in ("tinyint", "smallint"):
|
||
return "tinyint"
|
||
if type == "text":
|
||
return "nvarchar(max)"
|
||
if type in ("blob", "mediumblob"):
|
||
return "varbinary(max)"
|
||
if type == "decimal":
|
||
return (
|
||
f"numeric({','.join(str(s) for s in size)})" if len(size) else "numeric"
|
||
)
|
||
|
||
def gen_create(self, ddl: Dict) -> str:
|
||
"""生成 create"""
|
||
|
||
def _generate_column(col):
|
||
name = col["name"].lower()
|
||
if name == "id":
|
||
return "id bigint NOT NULL PRIMARY KEY IDENTITY"
|
||
if name == "deleted":
|
||
return "deleted bit DEFAULT 0 NOT NULL"
|
||
|
||
type = col["type"].lower()
|
||
full_type = self.translate_type(type, col["size"])
|
||
nullable = "NULL" if col["nullable"] else "NOT NULL"
|
||
default = f"DEFAULT {col['default']}" if col["default"] is not None else ""
|
||
return f"{name} {full_type} {default} {nullable}"
|
||
|
||
table_name = ddl["table_name"].lower()
|
||
columns = [f"{_generate_column(col).strip()}" for col in ddl["columns"]]
|
||
filed_def_list = ",\n ".join(columns)
|
||
script = f"""-- ----------------------------
|
||
-- Table structure for {table_name}
|
||
-- ----------------------------
|
||
DROP TABLE IF EXISTS {table_name};
|
||
CREATE TABLE {table_name} (
|
||
{filed_def_list}
|
||
)
|
||
GO"""
|
||
|
||
return script
|
||
|
||
def gen_comment(self, table_sql: str, table_name: str) -> str:
|
||
"""生成字段及表的注释"""
|
||
|
||
script = ""
|
||
|
||
for field, comment_string in self.filed_comments(table_sql):
|
||
script += f"""EXEC sp_addextendedproperty
|
||
'MS_Description', N'{comment_string}',
|
||
'SCHEMA', N'dbo',
|
||
'TABLE', N'{table_name}',
|
||
'COLUMN', N'{field}'
|
||
GO
|
||
|
||
"""
|
||
|
||
table_comment = self.table_comment(table_sql)
|
||
if table_comment:
|
||
script += f"""EXEC sp_addextendedproperty
|
||
'MS_Description', N'{table_comment}',
|
||
'SCHEMA', N'dbo',
|
||
'TABLE', N'{table_name}'
|
||
GO
|
||
|
||
"""
|
||
return script
|
||
|
||
def gen_pk(self, table_name: str) -> str:
|
||
"""生成主键定义"""
|
||
return ""
|
||
|
||
def gen_index(self, ddl: Dict) -> str:
|
||
"""生成 index"""
|
||
return "\n".join(f"{script}\nGO" for script in self.index(ddl))
|
||
|
||
def gen_insert(self, table_name: str) -> str:
|
||
"""生成 insert 语句"""
|
||
|
||
# 收集 `table_name` 对应的 insert 语句
|
||
inserts = []
|
||
for insert_script in Convertor.inserts(table_name, self.content):
|
||
# SQLServer: 字符串前加N,hack,是否存在替换字符串内容的风险
|
||
insert_script = insert_script.replace(", '", ", N'").replace(
|
||
"VALUES ('", "VALUES (N')"
|
||
)
|
||
# 删除 insert 的结尾分号
|
||
insert_script = re.sub(";$", r"\nGO", insert_script)
|
||
inserts.append(insert_script)
|
||
|
||
## 生成 insert 脚本
|
||
script = ""
|
||
if inserts:
|
||
inserts_lines = "\n".join(inserts)
|
||
script += f"""\n\n-- ----------------------------
|
||
-- Records of {table_name.lower()}
|
||
-- ----------------------------
|
||
-- @formatter:off
|
||
BEGIN TRANSACTION
|
||
GO
|
||
SET IDENTITY_INSERT {table_name.lower()} ON
|
||
GO
|
||
{inserts_lines}
|
||
SET IDENTITY_INSERT {table_name.lower()} OFF
|
||
GO
|
||
COMMIT
|
||
GO
|
||
-- @formatter:on"""
|
||
|
||
return script
|
||
|
||
def gen_dual(self) -> str:
|
||
return """DROP TABLE IF EXISTS dual
|
||
GO
|
||
|
||
CREATE TABLE dual
|
||
(
|
||
id int NULL
|
||
)
|
||
GO
|
||
|
||
EXEC sp_addextendedproperty
|
||
'MS_Description', N'数据库连接的表',
|
||
'SCHEMA', N'dbo',
|
||
'TABLE', N'dual'
|
||
GO"""
|
||
|
||
|
||
class DM8Convertor(Convertor):
|
||
def __init__(self, src):
|
||
super().__init__(src, "DM8")
|
||
|
||
def translate_type(self, type: str, size: Optional[Union[int, Tuple[int]]]):
|
||
"""类型转换"""
|
||
type = type.lower()
|
||
|
||
if type == "varchar":
|
||
return f"varchar({size})"
|
||
if type == "int":
|
||
return "int"
|
||
if type == "bigint" or type == "bigint unsigned":
|
||
return "bigint"
|
||
if type == "datetime":
|
||
return "datetime"
|
||
if type == "bit":
|
||
return "bit"
|
||
if type in ("tinyint", "smallint"):
|
||
return "smallint"
|
||
if type == "text":
|
||
return "text"
|
||
if type == "blob":
|
||
return "blob"
|
||
if type == "mediumblob":
|
||
return "varchar(10240)"
|
||
if type == "decimal":
|
||
return (
|
||
f"decimal({','.join(str(s) for s in size)})" if len(size) else "decimal"
|
||
)
|
||
|
||
def gen_create(self, ddl) -> str:
|
||
"""生成 CREATE 语句"""
|
||
|
||
def generate_column(col):
|
||
name = col["name"].lower()
|
||
if name == "id":
|
||
return "id bigint NOT NULL PRIMARY KEY IDENTITY"
|
||
|
||
type = col["type"].lower()
|
||
full_type = self.translate_type(type, col["size"])
|
||
nullable = "NULL" if col["nullable"] else "NOT NULL"
|
||
default = f"DEFAULT {col['default']}" if col["default"] is not None else ""
|
||
return f"{name} {full_type} {default} {nullable}"
|
||
|
||
table_name = ddl["table_name"].lower()
|
||
columns = [f"{generate_column(col).strip()}" for col in ddl["columns"]]
|
||
field_def_list = ",\n ".join(columns)
|
||
script = f"""-- ----------------------------
|
||
-- Table structure for {table_name}
|
||
-- ----------------------------
|
||
CREATE TABLE {table_name} (
|
||
{field_def_list}
|
||
);"""
|
||
|
||
# oracle INSERT '' 不能通过 NOT NULL 校验
|
||
script = script.replace("DEFAULT '' NOT NULL", "DEFAULT '' NULL")
|
||
|
||
return script
|
||
|
||
def gen_index(self, ddl: Dict) -> str:
|
||
return "\n".join(f"{script};" for script in self.index(ddl))
|
||
|
||
def gen_comment(self, table_sql: str, table_name: str) -> str:
|
||
script = ""
|
||
for field, comment_string in self.filed_comments(table_sql):
|
||
script += (
|
||
f"COMMENT ON COLUMN {table_name}.{field} IS '{comment_string}';" + "\n"
|
||
)
|
||
|
||
table_comment = self.table_comment(table_sql)
|
||
if table_comment:
|
||
script += f"COMMENT ON TABLE {table_name} IS '{table_comment}';\n"
|
||
|
||
return script
|
||
|
||
def gen_pk(self, table_name: str) -> str:
|
||
"""生成主键定义"""
|
||
return ""
|
||
|
||
def gen_index(self, ddl: Dict) -> str:
|
||
return "\n".join(f"{script};" for script in self.index(ddl))
|
||
|
||
def gen_insert(self, table_name: str) -> str:
|
||
"""拷贝 INSERT 语句"""
|
||
inserts = list(Convertor.inserts(table_name, self.content))
|
||
|
||
## 生成 insert 脚本
|
||
script = ""
|
||
if inserts:
|
||
inserts_lines = "\n".join(inserts)
|
||
script += f"""\n\n-- ----------------------------
|
||
-- Records of {table_name.lower()}
|
||
-- ----------------------------
|
||
-- @formatter:off
|
||
SET IDENTITY_INSERT {table_name.lower()} ON;
|
||
{inserts_lines}
|
||
COMMIT;
|
||
SET IDENTITY_INSERT {table_name.lower()} OFF;
|
||
-- @formatter:on"""
|
||
|
||
return script
|
||
|
||
|
||
def main():
|
||
parser = argparse.ArgumentParser(description="芋道系统数据库转换工具")
|
||
parser.add_argument(
|
||
"type",
|
||
type=str,
|
||
help="目标数据库类型",
|
||
choices=["postgres", "oracle", "sqlserver", "dm8"],
|
||
)
|
||
args = parser.parse_args()
|
||
|
||
sql_file = pathlib.Path("../mysql/ruoyi-vue-pro.sql").resolve().as_posix()
|
||
convertor = None
|
||
if args.type == "postgres":
|
||
convertor = PostgreSQLConvertor(sql_file)
|
||
elif args.type == "oracle":
|
||
convertor = OracleConvertor(sql_file)
|
||
elif args.type == "sqlserver":
|
||
convertor = SQLServerConvertor(sql_file)
|
||
elif args.type == "dm8":
|
||
convertor = DM8Convertor(sql_file)
|
||
else:
|
||
raise NotImplementedError(f"不支持目标数据库类型: {args.type}")
|
||
|
||
convertor.print()
|
||
|
||
|
||
if __name__ == "__main__":
|
||
main()
|