Repository navigation
ResultSetMetaData.getColumnClassName is broken for jsonb #3926
Description
Describe the issue
ResultSetMetaData.getColumnClassName does not return the correct class name for columns of type jsonb.
Specifically, the javadoc says it should return "the fully-qualified name of the class in the Java programming language that would be used by the method ResultSet.getObject to retrieve the value in the specified column"
The method returns "java.lang.String" in recent driver versions, whereas ResultSet.getObject returns instances of org.postgresql.util.PGobject, and that class does not extend java.lang.String.
This is not a problem for the "json" pg type, which is handled perfectly. Only "jsonb" is broken.
Driver Version?
42.7.9 (but I have tested older versions too)
Java Version?
17.0.13
OS Version?
Linux
PostgreSQL Version?
17.7
To Reproduce
Here is my table:
CREATE TABLE public.jsontest (
id int4 GENERATED ALWAYS AS IDENTITY NOT NULL,
fld_json json NOT NULL,
fld_jsonb jsonb NOT NULL,
CONSTRAINT jsontest_pk PRIMARY KEY (id)
);
I inserted one row with some dummy data.
kotlin test code:
import java.sql.DriverManager
import java.util.Properties
fun main() {
val url = "jdbc:postgresql://localhost/test"
val props = Properties().apply {
setProperty("user", "postgres")
setProperty("password", "")
}
DriverManager.getConnection(url, props).use { conn ->
val sql = "SELECT id, fld_json, fld_jsonb FROM jsontest"
val version = conn.metaData.driverVersion
println("driver version: $version")
conn.createStatement().use { statement ->
statement.executeQuery(sql).use { rs ->
val metaData = rs.metaData
while (rs.next()) {
for (i in 2..metaData.columnCount) {
val name = metaData.getColumnName(i)
val type = metaData.getColumnTypeName(i)
val cl = metaData.getColumnClassName(i)
val value = rs.getObject(i)
println("$name - type $type, metadata class $cl, value ${value.javaClass}")
println("instance test: " + Class.forName(cl).isInstance(value))
}
}
}
}
}
}
Expected behaviour
I expected the metadata class match the class of the returned value, or at least to be a supertype.
More specifically, I expected getColumnClassName to return "org.postgresql.util.PGobject" like it does for json.
It actually returns "java.lang.String" in recent driver versions, which would still be ok (though very weird and inconsistent with the json type) IF the value was actually returned as a String from getObject, but it is not - the actual value is a PGobject.
Actual results:
I tested this with several different driver versions, and got mainly 2 different outputs:
driver version: 42.2.0
fld_json - type json, metadata class org.postgresql.util.PGobject, value class org.postgresql.util.PGobject
instance test: true
fld_jsonb - type jsonb, metadata class java.lang.Object, value class org.postgresql.util.PGobject
instance test: true
Old versions:
type json - perfect match: PGobject to PGobject
type jsonb - inconsistent but not broken: Object to PGobject
driver version: 42.7.9
fld_json - type json, metadata class org.postgresql.util.PGobject, value class org.postgresql.util.PGobject
instance test: true
fld_jsonb - type jsonb, metadata class java.lang.String, value class org.postgresql.util.PGobject
instance test: false
New versions:
type json - perfect match: PGobject to PGobject
type jsonb - actually broken: String to PGobject
Proposed solution (preferred): getColumnClassName should return "org.postgresql.util.PGobject" like it does for json.
Alternative solution (inferior): getColumnClassName should return "java.lang.String" for both json and jsonb, but at the same time getObject should be modified to return a String value.
See also: bug #2153 - I don't really understand what happened there, but it seems to me that it was broken on purpose?! Very bizarre.
Activity
dkozlov-de commentedon Mar 10, 2026
I vote for the fix to this problem. This is clearly a bug. Metadata must return correct class data, otherwise, it is absolutely useless. This issue breaks any libraries that provide dynamic data type resolution based on the column metadata. At some point jsonb started being treated as a String. Well, that was a developers' call (even though it is still not clear to me what the reason for that change was). But if that (changing PGobject to String) was done in metadata, retrieving the column data would have to be changed accordingly to return a String instead of PGobject.
davecramer commentedon Mar 10, 2026
Well the reason is that it actually does return a string.
Can you help me out, what would your code do if it found jsonb ? How would you use this data ?
dkozlov-de commentedon Mar 10, 2026
I would like to clarify my point first. I do not have any problems with treating JSONB column data as a String. It is your call and I respect it. The problem is with resultset still returning PGobject.
We use org.apache.commons.beanutils.RowSetDynaClass and related classes to dynamically retrieve typed data from a resultset. This library uses column metadata to figure out what the type of a column is. Then when we ask for the column data, it tries to cast the data in resultset to the type provided by the metadata. At this point it fails because metadata says it is a String, but what the resultset returns is still PGobject.
My point is that if you change the type reported by metadata to String, you should be consistent and change the real type the resultset returns to String as well. Otherwise, the metadata column class information is wrong/misleading and prevents from proper data evaluation.
davecramer commentedon Mar 10, 2026
ah, thanks for the clarification.
aditsu commentedon Mar 10, 2026
@dkozlov-de I fully agree, I would only add that I think json and jsonb should be handled in a similar fashion by the driver. Either both PGobject (my preference) or both String (also acceptable).
davecramer commentedon Mar 11, 2026
Interestingly, this has been this way for 10 years see #640
given the comments on PR#640 I'm inclined to return the jsonb type as PGObject
Return jsonb as PGObject fixes Issue pgjdbc#3926
dkozlov-de commentedon Mar 11, 2026
@davecramer - thank you very much for the quick response. Yes, this could be a very old issue. We were still using pretty old version of the driver and decided to upgrade it. That's when we got this problem exposed.
aditsu commentedon Mar 11, 2026
I've used json for a long time and was familiar with how the driver returns PGobject, but only started using jsonb very recently and was shocked to find the discrepancy.
davecramer commentedon Mar 12, 2026
@vlsi what do you think of committing this?
I'm thinking we need to do a Major version bump as this changes the current behaviour
Return jsonb as PGObject fixes Issue pgjdbc#3926
Return jsonb as PGObject fixes Issue #3926 (#3956)
fix: align with post-rebase master changes
fix: align with post-rebase master changes
vlsi commentedon Jun 20, 2026
Fixed by #3956 and released in 42.7.11.
ResultSetMetaData.getColumnClassNamenow returnsorg.postgresql.util.PGobjectfor bothjsonandjsonb, matching whatResultSet.getObjectreturns. Regression teststestJsonbResultType/testJsonResultTypecover it.Thanks for the detailed report. Closing as resolved — please reopen if you still see the mismatch on 42.7.11 or later.