Same changelog on Postgres and MS SQL Server - sql-server

Question

I am trying to make a changelog work on two different databases: MS SQL Server and PostgreSQL. The changelog works fine on SQL Server, but the case of the database and fields make it break on PostgreSQL. I have tried not using quotes around the values (which throws an error) and used the objectQuotingStategy="QUOTE_ALL_OBJECTS", both which didn't work.
<?xml version="1.0" encoding="UTF-8"?>
<databaseChangeLog
xmlns="http://www.liquibase.org/xml/ns/dbchangelog"
xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
xsi:schemaLocation="http://www.liquibase.org/xml/ns/dbchangelog
http://www.liquibase.org/xml/ns/dbchangelog/dbchangelog-3.1.xsd">
<changeSet id="create-PushApplication" author="me">
<createTable tableName="PushApplication">
<column name="ApplicationKey" type="NUMERIC(19,0)" autoIncrement="true" remarks="Unique id of the application">
<constraints primaryKey="true" primaryKeyName="PushApplication_PK" nullable="false" />
</column>
<column name="ApplicationId" type="VARCHAR(100)" remarks="Id of the application in the Push Server">
<constraints unique="true" uniqueConstraintName="PushApplication_U1" nullable="false" />
</column>
<column name="MasterSecret" type="VARCHAR(100)" remarks="Password for the application in the Push Server">
<constraints nullable="false" />
</column>
<column name="Name" type="VARCHAR(100)" remarks="Name of the application">
<constraints nullable="false" />
</column>
<column name="Description" type="VARCHAR(200)" remarks="Description of the application" />
<column name="DateCreated" type="DATETIME" remarks="Date the application was created" />
</createTable>
</changeSet>
<changeSet id="create-PushVariant" author="me">
<createTable tableName="PushVariant">
<column name="VariantKey" type="NUMERIC(19,0)" autoIncrement="true" remarks="Unique id of the variant">
<constraints primaryKey="true" primaryKeyName="PushVariant_PK" nullable="false" />
</column>
<column name="VariantId" type="VARCHAR(100)" remarks="Id of the variant in the Push Server">
<constraints unique="true" uniqueConstraintName="PushVariant_UK1" nullable="false" />
</column>
<column name="Secret" type="VARCHAR(100)" remarks="Password for the variant in the Push Server">
<constraints nullable="false" />
</column>
<column name="Name" type="VARCHAR(100)" remarks="Name of the variant">
<constraints nullable="false" />
</column>
<column name="Description" type="VARCHAR(200)" remarks="Description of the variant" />
<column name="ApplicationKey" type="NUMERIC(19,0)" remarks="Id of the application the variant belogns to">
<constraints nullable="false" foreignKeyName="PushVariant_FK1" references="PushApplication(ApplicationKey)" />
</column>
</createTable>
</changeSet>
</databaseChangeLog>
ERROR: relation "pushapplication" does not exist
The table "PushApplication" is created in PostgreSQL but when it tries to create "PushVariant", this error is thrown for the foreign key.
This will work if I change all the database names and column names to lower case. However, this then makes the SQL Server have the incorrect case.
GOAL
The goal is to have "PushVariant" in SQL Server and "pushvariant" in PostgreSQL.
Is there a way that the case in the changelog can remain in SQL Server but have it lower case in PostgreSQL?

Answer 1

Supporting multiple DBMS with a common source code is always about compromises.
Even though there is a SQL standard that clearly defines how non-quoted identifiers have to be stored (all UPPERCASE), the two DBMS you are targeting both ignore this. Postgres stores un-quoted identifiers in lowercase, SQL Server is "case-preserving" (although not always case in-sensitive depending on the collation of the database).
In my personal experience when doing cross DBMS work, (always!) using un-quoted lower case identifiers with underscores is the least problematic way. And if you throw Oracle into the mix at some point in time you wind up all uppercase names.
Having said that: I think Liquibase's quoting strategy actually has a bug. According to the documentation, objectQuotingStrategy="QUOTE_ONLY_RESERVED_WORDS" should only quote reserved words, so in your example nothing should be quoted. But Liquibase still quotes any name that is using mixed case - this is still the case with the current 3.4.1 version.
I think the best thing was if Liquibase supported the choice objectQuotingStrategy="NEVER" (which it doesn't)
Another option would be to overwrite the default implementation on how the "need" for quoting is detected for Postgres and then use that implementation when running against the Postgres database.
The implementation is actually quite short and would only need to check for non-standard names (starting with a number, containing spaces or other illegal characters) and then quote only those.
I just tried this with the following implementation:
public class NonQuotingPostgresDatabase
extends PostgresDatabase {
#Override
public String correctObjectName(String objectName, Class<? extends DatabaseObject> objectType) {
return quoteIfNecessary(objectName);
}
#Override
public String escapeObjectName(String objectName, Class<? extends DatabaseObject> objectType) {
return quoteIfNecessary(objectName);
}
private String quoteIfNecessary(String objectName) {
if (requiresQuoting(objectName)) {
return "\"" + objectName + "\"";
}
return objectName;
}
protected boolean requiresQuoting(String identifier) {
if (identifier == null) {
return false;
}
if (isQuoted(identifier)) {
return false;
}
return (identifier.contains("-") || startsWithNumeric(identifier) || isReservedWord(identifier));
}
protected boolean isQuoted(String identifier) {
if (identifier == null) {
return false;
}
return (identifier.startsWith("\"") && identifier.endsWith("\""));
}
}
That class essentially leaves quoted identifiers alone and quotes only identifiers that really require it. The big difference to the built-in class is that it doesn't check for mixed case identifiers.
If this is put into a .jar file and the put into the lib subfolder of the Liquibase distribution it will be picked up automatically. All you need to do is add the parameter --databaseClass=NonQuotingPostgresDatabase when running Liquibase against Postgres