fix: global naming strategy not applied to Java records (#7656) by 1919chichi · Pull Request #7657 · alibaba/fastjson2

1919chichi · 2026-06-13T18:12:27Z

Summary

When serializing Java records, ObjectWriterCreator was calling method.getName() directly to get the field name, bypassing BeanUtils.getterName(method, kotlin, namingStrategy). This meant that both the global naming strategy (set via JSONFactory.getDefaultObjectWriterProvider().setNamingStrategy(...)) and the per-type @JSONType(naming = ...) annotation were silently ignored for record components.

Root cause (ObjectWriterCreator.java):

// Before — records bypassed the naming strategy entirely
if (record) {
    fieldName = method.getName();
} else {
    fieldName = BeanUtils.getterName(method, beanInfo.kotlin, beanInfo.namingStrategy);
    ...
}

// After — records go through the same naming pipeline as regular classes
fieldName = BeanUtils.getterName(method, beanInfo.kotlin, beanInfo.namingStrategy);
if (!record) {
    // NAME_COMPATIBLE_WITH_FILED logic (not applicable to records)
    ...
}

Test plan

JSONTypeNamingSnake#testRecord — verifies @JSONType(naming = SnakeCase) applies to records
JSONTypeNamingSnake#testRecordGlobalNaming — verifies global setNamingStrategy(SnakeCase) applies to records
./mvnw -pl core clean test — all tests pass
./mvnw -pl core -Dfastjson2.creator=reflect clean test — reflect mode tests pass
./mvnw -pl core validate — Checkstyle passes

CLAassistant · 2026-06-13T18:12:34Z

All committers have signed the CLA.

wenshao

[Critical] Deserialization round-trip is broken for records with naming strategies

This PR fixes only the writer side. The reader side (ObjectReaderCreator line ~194) assigns fieldName = paramName from the canonical constructor parameter names with no naming-strategy transformation. After this PR, JSON.toJSONString(new RecordDto("840")) produces {"currency_code":"840"}, but JSON.parseObject("{\"currency_code\":\"840\"}", RecordDto.class) fails to match the key currency_code against the raw parameter name currencyCode — the value is silently dropped.

The existing test only asserts serialization, not round-trip. The reader-side createFieldReaders(provider, objectClass, objectType, owner, parameters, paramNames) overload needs to apply beanInfo.namingStrategy to parameter names, similar to how the field-based overload applies it via BeanUtils.fieldName(fieldName, namingStrategy).

— qwen3.7-max via Qwen Code /review

artpaym · 2026-06-16T22:53:55Z

@wenshao @1919chichi As it seems related to the changes, please also ensure code gen code also gets the same fix (if needed).

1919chichi · 2026-06-17T05:46:17Z

Thanks for the thorough review! Here's a summary of the changes made to address all three points:

1. is/get prefix stripping (Writer side)
Restored the if (record) / else branch and replaced BeanUtils.getterName() with the existing BeanUtils.fieldName(method.getName(), namingStrategy) for records. This applies the naming strategy without any prefix stripping, so a field like isActive correctly serializes to is_active instead of active.

2. Deserialization round-trip (Reader side)
Fixed ObjectReaderCreator: after fetching record field names via BeanUtils.getRecordFieldNames(), the naming strategy is now applied to each parameter name so that incoming JSON keys (e.g. currency_code) hash-match correctly during constructor binding. Added round-trip deserialization assertions to the tests to cover this.

3. JDK 11 compilation failure (Tests)
Moved all record-based tests to the test-jdk17 module (compiled with source 17). Replaced the local record and var with a class-level static record and explicit types. Also added a BooleanRecordDto(boolean isActive) test case to explicitly verify the is prefix is preserved.

wenshao · 2026-06-22T23:25:21Z


 import com.alibaba.fastjson2.JSON;
-import com.alibaba.fastjson2.PropertyNamingStrategy;
 import org.junit.jupiter.api.Tag;


[Critical] Missing import causes compilation failure

The import com.alibaba.fastjson2.PropertyNamingStrategy was removed, but the class is still referenced at lines 27, 33, and 55 (PropertyNamingStrategy.SnakeCase). Since this file is in package com.alibaba.fastjson2.annotation and PropertyNamingStrategy lives in com.alibaba.fastjson2, the explicit import is required.

This causes core module test compilation to fail:

JSONTypeNamingSnake.java:[27,24] cannot find symbol JSONTypeNamingSnake.java:[33,24] cannot find symbol JSONTypeNamingSnake.java:[55,24] cannot find symbol

Suggested change

import org.junit.jupiter.api.Tag;

import com.alibaba.fastjson2.JSON;

import com.alibaba.fastjson2.PropertyNamingStrategy;

import org.junit.jupiter.api.Tag;

— qwen3.7-max via Qwen Code /review

wenshao · 2026-06-22T23:25:21Z

+    }
+
+    @Test
+    public void testRecordGlobalNaming() {


[Suggestion] testRecordGlobalNaming only tests serialization — deserialization with global naming strategy is untested

The reader-side fix in ObjectReaderCreator.java:1282-1286 applies naming strategy to record constructor parameter names when beanInfo.namingStrategy is non-null. For global strategies, this value comes from ObjectReaderProvider via the BeanInfo constructor. This entire code path has zero test coverage.

A regression in reader-side global naming support for records would go undetected.

Suggested change

public void testRecordGlobalNaming() {

@Test

public void testRecordGlobalNaming() {

ObjectWriterProvider writerProvider = JSONFactory.getDefaultObjectWriterProvider();

ObjectReaderProvider readerProvider = JSONFactory.getDefaultObjectReaderProvider();

PropertyNamingStrategy prevWriter = writerProvider.getNamingStrategy();

PropertyNamingStrategy prevReader = readerProvider.getNamingStrategy();

try {

writerProvider.setNamingStrategy(PropertyNamingStrategy.SnakeCase);

readerProvider.setNamingStrategy(PropertyNamingStrategy.SnakeCase);

assertEquals("{\"currency_code\":\"840\"}", JSON.toJSONString(new GlobalRecordDto("840")));

assertEquals(new GlobalRecordDto("840"), JSON.parseObject("{\"currency_code\":\"840\"}", GlobalRecordDto.class));

} finally {

writerProvider.setNamingStrategy(prevWriter);

readerProvider.setNamingStrategy(prevReader);

}

}

— qwen3.7-max via Qwen Code /review

wenshao · 2026-06-22T23:25:21Z

        if (fieldInfo.fieldName == null || fieldInfo.fieldName.isEmpty()) {
            if (record) {
-                fieldName = method.getName();
+                fieldName = BeanUtils.fieldName(method.getName(), beanInfo.namingStrategy);


[Suggestion] Missing null guard for namingStrategy — inconsistent with reader and sibling code

The record branch unconditionally calls BeanUtils.fieldName(method.getName(), beanInfo.namingStrategy). While BeanUtils.fieldName() handles null by defaulting to "CamelCase" (which is a no-op for standard lowercase-starting record names), this is inconsistent with:

Reader (ObjectReaderCreator.java:1282): if (beanInfo.namingStrategy != null && parameterNames != null)

Writer field path (line 243): if (beanInfo.namingStrategy != null)

Adding a null guard makes the code robust against future changes to BeanUtils.fieldName()'s null-handling and keeps the three paths consistent.

Suggested change

fieldName = BeanUtils.fieldName(method.getName(), beanInfo.namingStrategy);

fieldName = beanInfo.namingStrategy != null

? BeanUtils.fieldName(method.getName(), beanInfo.namingStrategy)

: method.getName();

— qwen3.7-max via Qwen Code /review

1919chichi · 2026-06-23T07:12:16Z

Updated the PR to address the latest review comments:

Restored the missing PropertyNamingStrategy import in JSONTypeNamingSnake.
Kept record writer field-name handling on the raw accessor name and added an explicit namingStrategy != null guard, so record components such as isActive are not treated as JavaBean getters.
Extended RecordNamingStrategyTest#testRecordGlobalNaming to cover global naming strategy deserialization via ObjectReaderProvider as well as serialization via ObjectWriterProvider.
Restored the accidentally removed lombok.var imports in the issue3601 tests, which caused the same kind of test compilation failure.

Validation run locally:

./mvnw -pl core -Dtest=JSONTypeNamingSnake test -q
../mvnw test -Dtest=RecordNamingStrategyTest -q from test-jdk17/
git diff --check

wenshao

No review findings. Downgraded from Approve to Comment: CI still running.

The implementation correctly integrates naming strategy support for Java records with proper reader/writer symmetry. @JSONField annotations are correctly resolved via constructor parameter annotations with proper precedence over naming strategy. Test coverage is adequate — serialization, deserialization, boolean prefix handling, and global naming strategy are all verified.

— qwen3.7-max via Qwen Code /review

artpaym · 2026-07-08T22:50:36Z

Hi. What's the status of this PR?

wenshao · 2026-07-10T07:47:29Z

CI is all green, but the following issues need to be addressed before merging:

1. Reader-side missing naming strategy support (round-trip broken)

ObjectReaderCreator ~L194 does not apply naming strategy transformation to record canonical constructor parameter names. Serialization produces currency_code, but deserialization fails to match against currencyCode — the value is silently dropped.

Apply beanInfo.namingStrategy to parameter names (same as the field-based overload), and add a deserialization assertion in testRecordGlobalNaming to verify round-trip.

2. `is`/`get` prefix incorrectly stripped

Routing records through BeanUtils.getterName() strips isActive to active. Records need a separate path: apply naming strategy only, skip JavaBean prefix detection.

3. `namingStrategy` null guard

The writer record branch calls BeanUtils.fieldName() without a null check on namingStrategy, inconsistent with the reader side and neighboring code paths.

Once these three are fixed, this is good to merge. Thanks for the contribution!

中文版本

CI 已经全绿，但以下几个问题需要解决后才能合入：

1. Reader 侧缺少 naming strategy 支持（round-trip 断掉）

ObjectReaderCreator ~L194 对 record canonical constructor 参数名没有做 naming strategy 转换。序列化产出 currency_code，反序列化时无法匹配 currencyCode，值被静默丢弃。

需要像 field-based overload 那样对参数名应用 beanInfo.namingStrategy，并在 testRecordGlobalNaming 中补充反序列化断言验证 round-trip。

2. `is`/`get` 前缀误剥离

Record 直接走 BeanUtils.getterName() 会把 isActive 剥成 active。Record 应走独立路径：只做 naming strategy 转换，跳过 JavaBean 前缀检测。

3. `namingStrategy` null guard

Writer record 分支调用 BeanUtils.fieldName() 前缺少 null 判断，与 reader 侧及相邻代码路径不一致。

修好这三点就可以合了，感谢贡献！

@jsontype

When a record's field name was resolved, `method.getName()` was called directly instead of `BeanUtils.getterName(…, namingStrategy)`, causing the global (and @jsontype) naming strategy to be silently ignored. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

- Writer: use BeanUtils.fieldName() for records to avoid stripping is/get prefixes that are part of the field name in Java records - Reader: apply namingStrategy to record parameterNames so deserialized JSON keys (e.g. currency_code) match the strategy-applied names - Tests: move record tests to test-jdk17 module (Java 16+ required), replace local record and var with class-level record and explicit types, add round-trip deserialization assertions and boolean prefix test Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

1919chichi · 2026-07-10T16:13:40Z

CI is all green, but the following issues need to be addressed before merging:

1. Reader-side missing naming strategy support (round-trip broken)

ObjectReaderCreator ~L194 does not apply naming strategy transformation to record canonical constructor parameter names. Serialization produces currency_code, but deserialization fails to match against currencyCode — the value is silently dropped.

Apply beanInfo.namingStrategy to parameter names (same as the field-based overload), and add a deserialization assertion in testRecordGlobalNaming to verify round-trip.

2. is/get prefix incorrectly stripped

Routing records through BeanUtils.getterName() strips isActive to active. Records need a separate path: apply naming strategy only, skip JavaBean prefix detection.

3. namingStrategy null guard

The writer record branch calls BeanUtils.fieldName() without a null check on namingStrategy, inconsistent with the reader side and neighboring code paths.

Once these three are fixed, this is good to merge. Thanks for the contribution!

中文版本

Hi All three issues have been addressed in db6666405 ("fix: address record naming review comments"), on top of the reader-side fix already landed in dc90fd47a. Details below.

1. Reader-side missing naming strategy support (round-trip broken)

Fixed in ObjectReaderCreator#createObjectReader (core/src/main/java/com/alibaba/fastjson2/reader/ObjectReaderCreator.java:1280-1287):

if (record && parameterNames == null) {
    parameterNames = BeanUtils.getRecordFieldNames(objectClass);
    if (beanInfo.namingStrategy != null && parameterNames != null) {
        for (int i = 0; i < parameterNames.length; i++) {
            parameterNames[i] = BeanUtils.fieldName(parameterNames[i], beanInfo.namingStrategy);
        }
    }
}

The record canonical-constructor parameter names now go through the same naming-strategy transform as the writer side, so a value serialized to currency_code deserializes correctly.

Covered by testRecordGlobalNaming in test-jdk17/src/test/java/com/alibaba/fastjson2/RecordNamingStrategyTest.java:31-46, which asserts both directions:

assertEquals("{\"currency_code\":\"840\"}", JSON.toJSONString(new GlobalRecordDto("840")));
assertEquals(new GlobalRecordDto("840"), JSON.parseObject("{\"currency_code\":\"840\"}", GlobalRecordDto.class));

2. `is`/`get` prefix incorrectly stripped

Fixed in ObjectWriterCreator#getFieldName (core/src/main/java/com/alibaba/fastjson2/writer/ObjectWriterCreator.java:756-759). Records no longer route through BeanUtils.getterName() (which does JavaBean prefix detection); they apply the naming strategy directly to the raw accessor name:

if (record) {
    fieldName = beanInfo.namingStrategy != null
            ? BeanUtils.fieldName(method.getName(), beanInfo.namingStrategy)
            : method.getName();
} else {
    fieldName = BeanUtils.getterName(method, beanInfo.kotlin, beanInfo.namingStrategy);
    ...

Added regression test testBooleanFieldNotStripped (RecordNamingStrategyTest.java:25-29) covering exactly the isActive case from the review:

public record BooleanRecordDto(boolean isActive) {}
...
assertEquals("{\"is_active\":true}", JSON.toJSONString(new BooleanRecordDto(true)));
assertEquals(new BooleanRecordDto(true), JSON.parseObject("{\"is_active\":true}", BooleanRecordDto.class));

3. `namingStrategy` null guard

Fixed at the same location (ObjectWriterCreator.java:757), shown in the snippet above — the record branch now short-circuits to method.getName() when namingStrategy is null, matching the reader side (ObjectReaderCreator.java:1282) and the writer field-based path (ObjectWriterCreator.java:243).

All three tests pass locally:

$ mvn -f test-jdk17/pom.xml -am test -Dtest=RecordNamingStrategyTest
Tests run: 3, Failures: 0, Errors: 0, Skipped: 0

Let me know if anything still looks off.

wenshao

No issues found. LGTM! ✅

— qwen3.8-max-preview via Qwen Code /review

1919chichi mentioned this pull request Jun 13, 2026

[BUG] global SnakeCase naming strategy is not applied to Java records #7656

Closed

wenshao requested changes Jun 16, 2026

View reviewed changes

Comment thread core/src/main/java/com/alibaba/fastjson2/writer/ObjectWriterCreator.java Outdated

Comment thread core/src/test/java/com/alibaba/fastjson2/annotation/JSONTypeNamingSnake.java Outdated

wenshao requested changes Jun 22, 2026

View reviewed changes

wenshao reviewed Jun 23, 2026

View reviewed changes

1919chichi and others added 3 commits July 10, 2026 23:52

fix: address record naming review comments

db66664

1919chichi force-pushed the fix/record-snake-case-naming-strategy branch from 33b0511 to db66664 Compare July 10, 2026 15:52

wenshao approved these changes Jul 31, 2026

View reviewed changes

wenshao merged commit d1029c7 into alibaba:main Aug 2, 2026
30 checks passed

-    public void testRecordGlobalNaming() {
+    @Test
+    public void testRecordGlobalNaming() {
+        ObjectWriterProvider writerProvider = JSONFactory.getDefaultObjectWriterProvider();
+        ObjectReaderProvider readerProvider = JSONFactory.getDefaultObjectReaderProvider();
+        PropertyNamingStrategy prevWriter = writerProvider.getNamingStrategy();
+        PropertyNamingStrategy prevReader = readerProvider.getNamingStrategy();
+        try {
+            writerProvider.setNamingStrategy(PropertyNamingStrategy.SnakeCase);
+            readerProvider.setNamingStrategy(PropertyNamingStrategy.SnakeCase);
+            assertEquals("{\"currency_code\":\"840\"}", JSON.toJSONString(new GlobalRecordDto("840")));
+            assertEquals(new GlobalRecordDto("840"), JSON.parseObject("{\"currency_code\":\"840\"}", GlobalRecordDto.class));
+        } finally {
+            writerProvider.setNamingStrategy(prevWriter);
+            readerProvider.setNamingStrategy(prevReader);
+        }
+    }

Uh oh!

Conversation

1919chichi commented Jun 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Test plan

Uh oh!

CLAassistant commented Jun 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

wenshao left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

artpaym commented Jun 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

1919chichi commented Jun 17, 2026

Uh oh!

wenshao Jun 22, 2026

Choose a reason for hiding this comment

Uh oh!

wenshao Jun 22, 2026

Choose a reason for hiding this comment

Uh oh!

wenshao Jun 22, 2026

Choose a reason for hiding this comment

Uh oh!

1919chichi commented Jun 23, 2026

Uh oh!

wenshao left a comment

Choose a reason for hiding this comment

Uh oh!

artpaym commented Jul 8, 2026

Uh oh!

wenshao commented Jul 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

1. Reader-side missing naming strategy support (round-trip broken)

2. is/get prefix incorrectly stripped

3. namingStrategy null guard

1. Reader 侧缺少 naming strategy 支持（round-trip 断掉）

2. is/get 前缀误剥离

3. namingStrategy null guard

Uh oh!

1919chichi commented Jul 10, 2026

1. Reader-side missing naming strategy support (round-trip broken)

2. is/get prefix incorrectly stripped

3. namingStrategy null guard

1. Reader-side missing naming strategy support (round-trip broken)

2. is/get prefix incorrectly stripped

3. namingStrategy null guard

Uh oh!

wenshao left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

1919chichi commented Jun 13, 2026 •

edited

Loading

CLAassistant commented Jun 13, 2026 •

edited

Loading

artpaym commented Jun 16, 2026 •

edited

Loading

wenshao commented Jul 10, 2026 •

edited

Loading

2. `is`/`get` prefix incorrectly stripped

3. `namingStrategy` null guard

2. `is`/`get` 前缀误剥离

3. `namingStrategy` null guard

2. `is`/`get` prefix incorrectly stripped

3. `namingStrategy` null guard

2. `is`/`get` prefix incorrectly stripped

3. `namingStrategy` null guard