penkovsky_sa/cherrypick
1
0
Fork 0
mirror of https://github.com/pese-git/cherrypick.git synced 2026-01-25 05:59:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

docs(report): update comparative DI benchmark results and conclusions for cherrypick, get_it, riverpod (eng, ru)

Browse Source
This commit is contained in:
Sergey Penkovsky
2025-08-07 16:46:53 +03:00
parent 44a8a3fcb2
commit 41d49e98d0
2 changed files with 107 additions and 113 deletions
Download Patch File Download Diff File Expand all files Collapse all files

105
benchmark_di/REPORT.md
View File

@@ -1,79 +1,76 @@
# DI Benchmark Results: cherrypick vs get_it
# Comparative DI Benchmark Report: cherrypick vs get_it vs riverpod
## Benchmark parameters
## Benchmark Parameters
| Parameter | Value |
|------------------|-----------------------|
| --benchmark | all |
| --chainCount (-c)| 10, 100 |
| --nestingDepth (-d)| 10, 100 |
| --repeat (-r) | 2 |
| --warmup (-w) | 1 (default) |
| --repeat (-r) | 5 |
| --warmup (-w) | 2 |
| --format (-f) | markdown |
| --di | cherrypick, get_it |
| --di | cherrypick, get_it, riverpod |
---
## Benchmark scenarios
## Benchmark Scenarios
**(1) RegisterSingleton**
Registers and resolves a singleton. Baseline DI speed.
**(2) ChainSingleton**
A dependency chain A → B → ... → N (singleton). Measures how fast DI resolves deep singleton chains by name.
**(3) ChainFactory**
Same as ChainSingleton, but every chain element is a factory. Shows DI speed for stateless 'creation chain'.
**(4) AsyncChain**
Async chain (async factory). Measures DI performance for async graphs.
**(5) Named**
Registers two bindings with names ("impl1", "impl2"), resolves by name. Tests named lookup.
**(6) Override**
Registers a chain/alias in a child scope and resolves UniversalService without a name in that scope. Simulates override and modular/test architecture.
1. **RegisterSingleton** — Registers and resolves a singleton. Baseline DI speed.
2. **ChainSingleton** — A dependency chain A → B → ... → N (singleton). Deep singleton chain resolution.
3. **ChainFactory** — All chain elements are factories. Stateless creation chain.
4. **AsyncChain** — Async chain (async factory). Performance on async graphs.
5. **Named** — Registers two bindings with names, resolves by name. Named lookup test.
6. **Override** — Registers a chain/alias in a child scope. Tests scope overrides.
---
## Comparative Table (Mean, ΔRSS), chainCount=10, nestingDepth=10
## Comparative Table: chainCount=10, nestingDepth=10 (Mean, PeakRSS)
| Scenario | cherrypick Mean (us) | cherrypick ΔRSS | get_it Mean (us) | get_it ΔRSS |
|--------------------|---------------------:|----------------:|-----------------:|------------:|
| RegisterSingleton | 21.0 | 320 | 24.5 | 80 |
| ChainSingleton | 112.5 | -3008 | 2.0 | 304 |
| ChainFactory | 8.0 | 0 | 4.0 | 0 |
| AsyncChain | 36.5 | 0 | 13.5 | 0 |
| Named | 1.5 | 0 | 0.5 | 0 |
| Override | 27.5 | 0 | 0.0 | 0 |
| Scenario | cherrypick Mean (us) | cherrypick PeakRSS | get_it Mean (us) | get_it PeakRSS | riverpod Mean (us) | riverpod PeakRSS |
|--------------------|---------------------:|-------------------:|-----------------:|---------------:|-------------------:|-----------------:|
| RegisterSingleton | 10.00 | 273104 | 15.20 | 261872 | 13.00 | 268512 |
| ChainSingleton | 10.20 | 271072 | 1.00 | 262000 | 41.20 | 268784 |
| ChainFactory | 5.00 | 299216 | 5.00 | 297136 | 43.80 | 271296 |
| AsyncChain | 43.40 | 290640 | 23.40 | 342976 | 105.20 | 285920 |
| Named | 1.00 | 297008 | 1.00 | 449824 | 2.20 | 281136 |
| Override | 5.40 | 297024 | 0.00 | 449824 | 30.20 | 281152 |
## Maximum load: chainCount=100, nestingDepth=100
## Maximum Load: chainCount=100, nestingDepth=100 (Mean, PeakRSS)
| Scenario | cherrypick Mean (us) | cherrypick ΔRSS | get_it Mean (us) | get_it ΔRSS |
|--------------------|---------------------:|----------------:|-----------------:|------------:|
| RegisterSingleton | 1.0 | 32 | 1.0 | 0 |
| ChainSingleton | 3884.0 | 0 | 1.5 | 34848 |
| ChainFactory | 4088.0 | 0 | 50.0 | 12528 |
| AsyncChain | 4287.0 | 0 | 17.0 | 63120 |
| Named | 1.0 | 0 | 0.0 | 0 |
| Override | 4767.5 | 0 | 1.5 | 14976 |
| Scenario | cherrypick Mean (us) | cherrypick PeakRSS | get_it Mean (us) | get_it PeakRSS | riverpod Mean (us) | riverpod PeakRSS |
|--------------------|---------------------:|-------------------:|-----------------:|---------------:|-------------------:|-----------------:|
| RegisterSingleton | 1.00 | 271072 | 1.20 | 262000 | 2.00 | 268688 |
| ChainSingleton | 49.20 | 303312 | 1.20 | 297136 | 253.20 | 270784 |
| ChainFactory | 45.00 | 293952 | 51.80 | 342720 | 372.80 | 308640 |
| AsyncChain | 261.60 | 297008 | 25.00 | 450640 | 821.80 | 285968 |
| Named | 1.00 | 297008 | 1.00 | 449824 | 2.00 | 281136 |
| Override | 226.60 | 301632 | 1.80 | 477344 | 498.60 | 294752 |
---
## Scenario explanations
## Scenario Explanations
- **RegisterSingleton:** Registers and resolves a singleton dependency, baseline test for cold/hot startup speed.
- **ChainSingleton:** Deep chain of singleton dependencies. Cherrypick is much slower as depth increases; get_it is nearly unaffected.
- **ChainFactory:** Creation chain with new instances per resolve. get_it generally faster on large chains due to ultra-simple factory registration.
- **AsyncChain:** Async factory chain. get_it processes async resolutions much faster; cherrypick is much slower as depth increases due to async handling.
- **Named:** Both DI containers resolve named bindings nearly instantly, even on large graphs.
- **Override:** Child scope override. get_it (thanks to stack-based scopes) resolves immediately; cherrypick supports modular testing with controlled memory use.
- **RegisterSingleton**: Baseline singleton registration and resolution.
- **ChainSingleton**: Deep singleton chains, stress for lookup logic.
- **ChainFactory**: Stateless factory chains.
- **AsyncChain**: Async factories/graphs.
- **Named**: Named binding resolution.
- **Override**: Scope override and modular/test archetypes.
## Summary
---
- **get_it** demonstrates impressive speed and low overhead across all scenarios and loads, but lacks diagnostics, advanced scopes, and cycle detection.
- **cherrypick** is ideal for complex, multi-layered, production or testable architectures where scope, overrides, and diagnostics are critical. Predictably slower on deep/wide graphs, but scales well and provides extra safety.
## Conclusions
**Recommendation:**
- Use cherrypick for enterprise, multi-feature/testable DI needs.
- Use get_it for fast games, scripts, tiny Apps, and hot demos.
- **GetIt** has record-best speed and the lowest memory use in almost every scenario. Especially effective for deep/wide graphs and shows ultra-high stability (lowest jitter).
- **Cherrypick** is fast, especially on simple chains or named resolutions, but is predictably slower as complexity grows. Excels in production, codegen, and testable setups where advanced scopes/diagnostics matter.
- **Riverpod** holds its ground in basic and named scenarios, but time and memory grow much faster under heavy/complex loads, especially for deep async/factory/override chains.
### Recommendations
- Use **GetIt** when maximum performance and low memory are top priorities (games, scripts, simple apps, perf-critical UI).
- Use **Cherrypick** for scalable, multi-package testable apps, where advanced scopes, codegen, and diagnostics are needed.
- Use **Riverpod** when you need reactive state or deep Flutter integration, and the DI graph's depth/width is moderate.
---
_Last updated: August 7, 2025, with full scenario matrix. Developed using open-source benchmark scripts._