hibernate-demo

Companion repository for three ankurm.com posts on Hibernate 7's persistence-context APIs: get() vs getReference(), merge() vs refresh(), and batch inserts. Every claim in those posts that comes from this repo traces to a named JUnit test here and a captured transcript in docs/output/ — nothing is asserted that wasn't actually run. Every surprising behavior described in the three posts has a reproducible test backing it; the table below maps each one.

Versions

Component	Version
Hibernate ORM	7.4.1.Final (GA 2026-06-09)
Spring Boot	4.1.0 (GA 2026-06-10)
Java	25 (LTS)
H2	in-memory, managed by Spring Boot

See docs/00-versions.md for how these were verified and a trap worth knowing about if you bump the Spring Boot version.

Quickstart

git clone https://ankurm.com/git.app/asmhatre/hibernate-demo.git
cd hibernate-demo
./mvnw test

Requires JDK 25 and a network connection the first time (Maven needs to fetch plugins online before -o offline mode works for subsequent runs).

Every surprising claim, mapped to a test

Each row is one ./mvnw -Dtest=ClassName test away from reproducing itself. This is the reproduction path referenced throughout all three posts and all three doc chapters below.

Test class	Backs post	Proves
GetVsGetReferenceTest	4859 (get vs getReference)	The 4-calls-4-outcomes table, the same-session matrix, proxy identity vs equals()
MergeRefreshTest	4860 (merge vs refresh)	merge() returns the pre-existing managed instance; merge() initializes a cascaded LAZY collection; refresh() silently discards an unflushed edit
OptimisticLockTest	4860 (merge vs refresh)	Exactly when OptimisticLockException surfaces relative to merge()/commit
IdentityBatchTest	4861 (batch inserts)	GenerationType.IDENTITY disables batching entirely
SequenceBatchTest	4861 (batch inserts)	GenerationType.SEQUENCE allows real batching
AllocationSizeSweepTest	4861 (batch inserts)	allocationSize sweep (1, 10, 25, 50) at fixed batch_size=25
BatchSizeSweepTest	4861 (batch inserts)	batch_size sweep (1, 10, 25, 50) at fixed allocationSize=50

./mvnw -Dtest=GetVsGetReferenceTest test
./mvnw -Dtest=MergeRefreshTest,OptimisticLockTest test
./mvnw -Dtest=IdentityBatchTest,SequenceBatchTest test
./mvnw -Dtest=AllocationSizeSweepTest test
./mvnw -Dtest=BatchSizeSweepTest test
./mvnw test   # the whole suite, 24 tests, 0 failures as of the last commit

CommandLineRunner scenarios

The original narrative scenarios are still here, unchanged, for anyone who wants to read a straight-line script instead of a test class:

Profile	Runs	Chapter
getvsload	session.get() vs session.getReference(), proxies, LazyInitializationException	docs/01-get-vs-load.md
mergerefresh	merge() vs refresh() against a @Version-ed entity	docs/02-merge-vs-refresh.md
insert-identity	Batch insert attempt with GenerationType.IDENTITY	docs/03-inserting-objects.md
insert-sequence	The same insert, with GenerationType.SEQUENCE	docs/03-inserting-objects.md

./scripts/run.sh getvsload
./scripts/run.sh mergerefresh
./scripts/run.sh insert-identity
./scripts/run.sh insert-sequence

Regenerating captured output

./scripts/run-all.sh

Regenerates the CommandLineRunner-scenario files in docs/output/. scripts/clean_output.py strips JVM noise and a harmless duplicate SQL echo line so the committed transcripts stay readable — nothing else is edited by hand. The test-suite transcripts in docs/output/ (the session matrix, the sweeps) were captured the same way, from ./mvnw -Dtest=... test piped through the same script.

Documentation index

Chapter	Covers
00 — Versions	Verified version pins, and the Spring Boot patch that silently changes which Hibernate patch you get
01 — get() vs getReference()	The 4-calls-4-outcomes table, the same-session matrix, proxy identity vs equals()
02 — merge() vs refresh()	Three named experiments: which method fails loudly vs silently, exactly when the optimistic-lock check fires, and what a cascaded LAZY collection does under merge()
03 — Hibernate 7 batch inserts	IDENTITY vs SEQUENCE, the allocationSize and batch_size sweeps, Session vs StatelessSession, with hibernate.generate_statistics as evidence throughout

Captured output index

File	Source
docs/output/get-vs-load.txt	./scripts/run.sh getvsload (CommandLineRunner)
docs/output/get-vs-getreference-tests.txt	./mvnw -Dtest=GetVsGetReferenceTest test
docs/output/merge-vs-refresh.txt	./scripts/run.sh mergerefresh (CommandLineRunner)
docs/output/merge-vs-refresh-tests.txt	./mvnw -Dtest=MergeRefreshTest,OptimisticLockTest test
docs/output/insert-identity.txt	./scripts/run.sh insert-identity (also matches IdentityBatchTest)
docs/output/insert-sequence.txt	./scripts/run.sh insert-sequence (also matches SequenceBatchTest)
docs/output/allocation-and-batch-size-sweeps.txt	./mvnw -Dtest=AllocationSizeSweepTest,BatchSizeSweepTest test

Layout

hibernate-demo/
├── pom.xml
├── LICENSE
├── scripts/
│   ├── run.sh              start one CommandLineRunner profile, run it, exit
│   ├── run-all.sh          regenerate the CommandLineRunner docs/output/ files
│   └── clean_output.py     strip JVM noise + a duplicate SQL echo line from a raw capture
├── src/main/java/com/ankurm/hibernatedemo/
│   ├── HibernateDemoApplication.java
│   ├── model/               Book, Note, WidgetIdentity, WidgetSequence,
│   │                        WidgetAlloc1/10/25/50, WidgetBatchSweep1/10/25/50
│   └── scenario/            one CommandLineRunner per profile
├── src/test/java/com/ankurm/hibernatedemo/
│   ├── GetVsGetReferenceTest.java
│   ├── MergeRefreshTest.java, OptimisticLockTest.java
│   ├── IdentityBatchTest.java, SequenceBatchTest.java
│   └── AllocationSizeSweepTest.java, BatchSizeSweepTest.java
└── docs/
    ├── 00-versions.md .. 03-inserting-objects.md
    └── output/*.txt         captured, unedited console transcripts (both runner and test output)

License

MIT — see LICENSE.