rzo1 commented on code in PR #304:
URL: https://github.com/apache/openjpa/pull/304#discussion_r4550724786


##########
openjpa-jdbc/src/main/java/org/apache/openjpa/jdbc/identifier/Normalizer.java:
##########
@@ -71,6 +71,24 @@ public static String normalizeString(String name) {
         return name;
     }
 
+    /**
+     * Normalizes a single part name with the given rule rather than the
+     * default rule, such as the rule of a dictionary that delimits its
+     * reserved words.
+     * @param name
+     * @param rule
+     * @since 4.2.0
+     */
+    public static String normalizeString(String name, IdentifierRule rule) {
+        if (StringUtil.isEmpty(name)) {
+            return name;
+        }
+        if (!normalizer.isDelimited(rule, name)) {

Review Comment:
   Yes, that split is the intended design:
   
   - The static `Normalizer` now does only the database-independent part: it 
delimits what has to be delimited syntactically, the same for every database. 
Before this PR it was "default rule + whatever reserved words the last 
configured MySQL/MariaDB dictionary had put into it", which is exactly the bug: 
the dictionary configured last decided the names of every persistence unit in 
the JVM.
   - The reserved words of a database only matter where a name is generated for 
that database or reaches it, and those paths now go through the dictionary 
(`DBDictionary.normalize` / `toDBName` / `fromDBName`): default names in the 
mapping defaults and `makeNameValid`, configured schemas 
(`openjpa.jdbc.Schema`, the persistence-unit default schema) and given table 
names, `XMLSchemaParser`, DDL/SQL, and names read back from the database 
metadata. A `DBIdentifier` created statically before that is an intermediate 
value; the dictionary normalizes it when it matters.
   - To make sure nothing relied on the old shared state, 
`TestNamesPerDictionary` records the generated names (tables, columns, join 
tables/columns, sequences, indexes, constraints, delimited and mixed-case 
names, configured schemas, metadata lookups) per dictionary. For 17 setups the 
names of each dictionary alone are identical before and after this change, and 
with several dictionaries configured one after another, interleaved or on 
different threads at the same time, each gets the names it gets alone. Plus the 
full suite on Derby, the dictionary/schema tests on MySQL, MariaDB and 
PostgreSQL, and the TCK on Derby, PostgreSQL, MySQL and MariaDB.
   
   So the static instance being "unequipped" is what makes it safe: it can no 
longer carry one database's words into another persistence unit.
   



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to