jungm opened a new pull request, #268:
URL: https://github.com/apache/bval/pull/268
This is an experiment on improving perfomance. With these patches BVal is
**~2–38× faster than `master`** (median ~7.7×) and ahead of Hibernate Validator
10 on all 25 benchmarks; before, it was 3–14× slower than HV on most validation
paths. I didn't magically understand every little detail of BVal and profile it
to hell and back, this was done with the help of Claude starting out more as an
experiment to see how far it is able to go. It has gone **A LOT** further than
i anticipated.
Happy to split this into multiple PRs to make it easier to digest,
admittedly that diff is large. But wanted to get this out there and maybe some
first feedback?
<details>
<summary>Benchmarks (ops/ms, higher is better)</summary>
JMH 1.37, JDK 25, 1 thread, 2 forks × 5 iterations. Before = `master`
(aab2f373), HV = 10.0.0-SNAPSHOT (Validation 4.0 branch). Non-idle machine, so
ratios are ±20–30%.
| Benchmark | Before | After | HV | **Speed-up** | Before vs HV | After vs
HV |
|---|---|---|---|---|---|---|
| **Original harness** | | | | | | |
| Constraints pass, reused validator | 996 | 13,419 | 8,183 | **13.5×** |
0.12× | **1.64×** |
| Constraints pass, new validator | 938 | 12,608 | 3,971 | **13.4×** | 0.24×
| **3.17×** |
| Constraints fail, reused validator | 251 | 2,037 | 1,735 | **8.1×** |
0.14× | **1.17×** |
| Constraints fail, new validator | 133 | 3,083 | 1,725 | **23.2×** | 0.08×
| **1.79×** |
| No constraints, reused validator | 113,928 | 189,253 | 125,862 | **1.7×**
| 0.91× | **1.50×** |
| No constraints, new validator | 97,445 | 159,841 | 9,584 | **1.6×** |
10.2× | **16.7×** |
| Build factory | 3.7 | 9.9 | 4.2 | **2.7×** | 0.89× | **2.38×** |
| Build factory + validate once | 1.4 | 3.2 | 2.8 | **2.3×** | 0.50× |
**1.15×** |
| **Messages and constraint types** | | | | | | |
| EL expression in message | 43 | 1,651 | 641 | **38.2×** | 0.07× |
**2.58×** |
| 16 properties, all invalid | 11.8 | 252 | 115 | **21.3×** | 0.10× |
**2.19×** |
| 16 properties, all valid | 144 | 781 | 391 | **5.4×** | 0.37× | **2.00×** |
| Composed constraint, failing | 311 | 2,443 | 2,241 | **7.9×** | 0.14× |
**1.09×** |
| Composed constraint, passing | 780 | 4,632 | 2,933 | **5.9×** | 0.27× |
**1.58×** |
| Class-level constraint, failing | 605 | 3,039 | 2,472 | **5.0×** | 0.24× |
**1.23×** |
| Class-level constraint, passing | 918 | 6,905 | 3,777 | **7.5×** | 0.24× |
**1.83×** |
| **API variants** | | | | | | |
| Mixed valid/invalid beans | 471 | 3,642 | 1,335 | **7.7×** | 0.35× |
**2.73×** |
| Explicit group | 440 | 4,816 | 1,269 | **10.9×** | 0.35× | **3.79×** |
| `validateProperty` | 986 | 2,603 | 1,814 | **2.6×** | 0.54× | **1.44×** |
| `validateValue` | 1,918 | 6,513 | 1,722 | **3.4×** | 1.11× | **3.78×** |
| Method parameters | 345 | 2,218 | 1,376 | **6.4×** | 0.25× | **1.61×** |
| Method return value | 334 | 2,178 | 967 | **6.5×** | 0.35× | **2.25×** |
| **Object graphs** | | | | | | |
| Chain of 8 nested beans | 85 | 865 | 468 | **10.2×** | 0.18× | **1.85×** |
| Map keys, list elements, Optional | 1.3 | 10.6 | 3.4 | **7.9×** | 0.39× |
**3.11×** |
| List of 1,000 beans | 0.45 | 4.3 | 2.4 | **9.7×** | 0.19× | **1.79×** |
| 200 beans with back-references | 2.2 | 17.2 | 13.7 | **7.9×** | 0.16× |
**1.26×** |
</details>
## What changed and why
- **Per-bean, per-group plans:** the matching constraints, the Default-group
redefinition and the relevant properties are cached per `GroupStrategy` on the
descriptors. Previously they were recomputed with streams on every call, and
every constrained property was read even when nothing in the requested groups
applied to it. Lookups check a small identity-keyed array before the
`ConcurrentHashMap`, because validation keeps passing the same strategy
instances.
- **Lazy property frames:** a property is read the first time a group needs
it, so a failing group-sequence step never reads properties that only later
steps use.
- **Property reads through cached `MethodHandle`s:**
`Field#get`/`Method#invoke` take a slow path when the reflective object isn't a
JIT constant (~48% of simple validation). Reflection remains the fallback, with
the same exception wrapping.
- **EL:** `ELFacade` ran a ~126-alternation look-behind regex over every
`${` message to disarm `#{`, and several built-in default messages contain EL.
A direct scan produces identical output (38× on the EL benchmark).
- **Messages:** fully interpolated messages that need no EL are cached,
keyed on the constraint's attribute map *identity*, since descriptors return a
stable instance. The message template is memoized instead of re-read
reflectively on every violation.
- **Traversable resolver:** skipped when it is exactly
`DefaultTraversableResolver` with no JPA, since it can only say yes and calling
it forced path copies. The exact-class check keeps subclasses honored.
- **`@ReportAsSingleViolation`:** violations of the composing constraints
are discarded anyway, so they are no longer built (path, interpolation,
hashing).
- **Container elements:** the runtime container key is cached per class (it
was resolved reflectively per container per call), the ancestor lookup no
longer copies paths, and the lambdas and synchronized `Lazy` on the per-element
path are gone.
- **Executables:** descriptors are cached per `Executable` instead of
hashing a new `Signature` twice per call, and default parameter names are
cached.
- **`@Email`:** a hand-written matcher replaces `DEFAULT_EMAIL_PATTERN`
(fuzzed against it on 4M strings), and the default `.*` regexp is checked by
scanning for line terminators.
- **Groups:** resolved groups are cached for a single explicit group.
- **Smaller fixes:**
- `DescriptorManager#getBeanDescriptor` wrote to a CHM on every cache
*hit*; it no longer does.
- Descriptors are looked up once per validation, keyed by runtime class.
- The validator is memoized on `ConstraintD`.
- `PathImpl` is backed by an `ArrayList`.
- `NodeImpl`/`ConstraintViolationImpl` hash codes are computed without
varargs arrays.
## Worth knowing
- **Traversable resolver calls:** `isReachable` is now only called for
properties that are actually validated, as HV does.
- **Violation order:** it is now deterministic; frames were previously in an
identity `HashSet`.
- **`DefaultParameterNameProvider`:** it returns cached unmodifiable lists.
- **Serialization:** the serialized form of `PathImpl` changed.
- **API:** `DescriptorManager#getCachedBeanConstrained` was removed;
`ValidateParameters.parameterNames` is now private; there are some new public
methods on internal classes.
- **Harness (`bval-perf`):**
- New `ScenarioBenchmark` with 17 scenarios for both providers; it asserts
equal violation counts before measuring.
- `<proc>full</proc>`, because JDK 23+ skips annotation processing and no
benchmark list was being generated.
- Registers Tomcat's `ExpressionFactory` ahead of the test-jar
`DelegateExpressionFactory`, which does a `ServiceLoader` lookup per call.
- **Open question:** `ELFacade` replaces unescaped `#{` with a literal `$0`.
That is kept as-is here, but it was probably meant to be `\#{`.
--
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]
