[client-v2] JSON typed paths containing a space or comma break when the JSON value is nested in a Dynamic column

[claude]

Description

ClickHouse allows a JSON typed path to contain characters that require backtick quoting (a space, a comma). When such a JSON value is nested inside a Dynamic column, BinaryStreamReader.readDynamicData() reconstructs the type as a string from the binary type encoding and re-parses it with ClickHouseColumn.of(...). Path names are appended unquoted, so a path containing a space or comma produces a malformed type string and the whole query fails.

Top-level JSON columns are not affected: there the type comes from the RowBinaryWithNamesAndTypes header, where the server emits backticks, and ClickHouseUtils.readNameOrQuotedString handles them correctly. Same for JSON inside Map/Tuple/Array (header-derived). Only the binary Dynamic type-encoding path is broken.

Observed:

Query	Result
SELECT '{"a b": 1}'::JSON(`a b` Int64)	✅ {a b=1}
SELECT '{"a,b": 1}'::JSON(`a,b` Int64)	✅ {a,b=1}
SELECT map('k', '{"a b": 1}'::JSON(`a b` Int64))	✅ {k={a b=1}}
SELECT '{"a": 1}'::JSON(`a` Int64)::Dynamic	✅ {a=1}
SELECT '{"a b": 1}'::JSON(`a b` Int64)::Dynamic	❌ IllegalArgumentException: Unknown data type: b Int64
SELECT '{"a,b": 1}'::JSON(`a,b` Int64)::Dynamic	❌ IllegalArgumentException: Unknown data type: b Int64

The failure happens while decoding the column type, so the entire query fails — not just the offending path.

ClickHouse server version

26.7.1.1315 (local HTTP endpoint), verified against main (385f9b4-era working tree, client-v2 0.10.0-rc1-SNAPSHOT), Java 17.

Reproduction

client-v2/src/test/java/com/clickhouse/client/datatypes/JsonSpacePathTest.java:

package com.clickhouse.client.datatypes;

import com.clickhouse.client.api.Client;
import com.clickhouse.client.api.query.GenericRecord;
import org.testng.annotations.Test;

import java.util.List;
import java.util.Map;

import static org.testng.Assert.assertEquals;

public class JsonSpacePathTest {

    @Test
    public void testJsonPathWithSpaceInsideDynamic() {
        try (Client client = new Client.Builder()
                .addEndpoint("http://localhost:8123")
                .setUsername("default").setPassword("")
                .compressServerResponse(false)
                .build()) {

            // works
            List<GenericRecord> ok = client.queryAll("SELECT '{\"a b\": 1}'::JSON(`a b` Int64)");
            assertEquals(((Map<?, ?>) ok.get(0).getObject(1)).get("a b"), 1L);

            // throws: IllegalArgumentException: Unknown data type: b Int64
            List<GenericRecord> dyn = client.queryAll("SELECT '{\"a b\": 1}'::JSON(`a b` Int64)::Dynamic");
            assertEquals(((Map<?, ?>) dyn.get(0).getObject(1)).get("a b"), 1L);
        }
    }
}

Expected: both selects return {a b=1}.

Actual: the second one throws

com.clickhouse.client.api.ClientException: Failed to get query response
Caused by: java.lang.IllegalArgumentException: Unknown data type: b Int64
	at com.clickhouse.data.ClickHouseColumn.readColumn(ClickHouseColumn.java:...)
	at com.clickhouse.data.ClickHouseColumn.parseJSONColumn(ClickHouseColumn.java:846)
	at com.clickhouse.client.api.data_formats.internal.BinaryStreamReader.readDynamicData(BinaryStreamReader.java:1530)

Suggested fix

client-v2/src/main/java/com/clickhouse/client/api/data_formats/internal/BinaryStreamReader.java, readDynamicData() case JSON: (~lines 1508–1530): each path string read from the wire is appended verbatim:

for (int i = 0; i < numberOfTypedPaths; i++) {
    typeDef.append(readString(input)).append(' '); // path
    ClickHouseColumn column = readDynamicData();
    typeDef.append(column.getOriginalTypeName()).append(',');
}

Paths (and likewise the skip-path and path-regexp strings that follow) should be backtick-quoted with inner backticks escaped before being appended, e.g. typeDef.append('`').append(path.replace("`", "")).append('') ``. ClickHouseUtils.readNameOrQuotedString` already understands backticks and doubled-backtick escapes, so the round-trip works once the quoting is emitted.

The Nested branch a few lines below builds its type string the same way and likely has the same weakness for names needing quoting.

Link

Analogous issue reported against the .NET client: ClickHouse/clickhouse-cs#502 (there the top-level case is broken too; in clickhouse-java only the Dynamic-nested case is).