Sealed By Default, Or Not.
The choice between sealed and unsealed classes is not merely a syntactic one; it can have profound implications on the design, performance, and maintainability of a codebase. Most OOP languages, such as C#, provide explicit keywords to define classes as sealed, while in others, such as Swift, the concept is implicit.
Contents
- What Is A Sealed Class
- What Is An Unsealed Class
- Benchmarks
- Performance Tests
- Sealed By Default
- Unsealed By Default
- Conclusion
What Is A Sealed Class
A sealed class is a class that cannot be inherited. It's a final declaration of a class, ensuring that no other class can extend it. Sealed classes provide a level of security and predictability in the code, as they prevent further modifications to the class hierarchy. They can be particularly useful when you want to encapsulate specific behaviour without the risk of unintended alterations.
What Is An Unsealed Class
An unsealed class is open to inheritance, allowing other classes to extend it. This flexibility can lead to more dynamic and extensible code structures, enabling developers to build upon existing functionality.
Benchmarks
These benchmarks attempt to reproduce the results from the dotnet 6 performance improvements blog post
The below classes can be found here
| Method | Mean | Error | StdDev | Median |
|---|---|---|---|---|
| 'Base Benchmark' | 1.2395 ns | 0.0474 ns | 0.0466 ns | 1.2088 ns |
| 'Sealed Benchmark No Inheritance*' | 0.0019 ns | 0.0048 ns | 0.0042 ns | 0.0000 ns |
| 'Unsealed Benchmark No Inheritance*' | 0.0043 ns | 0.0065 ns | 0.0051 ns | 0.0030 ns |
| 'Sealed Benchmark*' | 0.0079 ns | 0.0124 ns | 0.0116 ns | 0.0000 ns |
| 'Unsealed Benchmark*' | 0.0188 ns | 0.0198 ns | 0.0212 ns | 0.0132 ns |
| 'Sealed Virtual Benchmark*' | 0.0143 ns | 0.0185 ns | 0.0173 ns | 0.0076 ns |
| 'Unsealed Virtual Benchmark' | 1.0502 ns | 0.0066 ns | 0.0055 ns | 1.0511 ns |
| 'Sealed Is Type Benchmark' | 0.2032 ns | 0.0043 ns | 0.0034 ns | 0.2022 ns |
| 'Unsealed Is Type Benchmark' | 1.7030 ns | 0.0316 ns | 0.0295 ns | 1.6929 ns |
The results for Sealed Benchmark, Unsealed Benchmark, Sealed Virtual Benchmark, Sealed Benchmark No Inheritance and Unsealed Benchmark No Inheritance (denoted by *) are marked as "ZeroMeasurement", highlighting that the methods being benchmarked are executing so quickly that they are indistinguishable from empty methods. This means that we cannot reliably measure these methods, and the speed can vary between different runs.
How To Run
From within the SealedVsUnsealed.Benchmark folder
dotnet run --configuration Release
Performance Tests
The below performance results compare different ways of calling methods that add two numbers together, one virtual method and one non-virtual method, each then called through mediator.
Source for the controllers can be found here
Source for the performance tests can be found here
K6 Results
| Method | Requests | Requests/s | Difference |
|---|---|---|---|
| 'Sealed' | 1993252 | 66441/s | 0.51% |
| 'Unsealed' | 1983031 | 66101/s | -0.51% |
| 'Sealed Mediator' | 1973171 | 65772/s | 0.43% |
| 'Unsealed Mediator' | 1964774 | 65492/s | -0.43% |
| 'Sealed Virtual' | 1958160 | 65272/s | 1.56% |
| 'Unsealed Virtual' | 1927837 | 64261/s | -1.56% |
| 'Sealed Virtual Mediator' | 1866520 | 62217/s | 6.31% |
| 'Unsealed Virtual Mediator' | 1752260 | 58408/s | -6.31% |
Full results
/\ |‾‾| /‾‾/ /‾‾/
/\ / \ | |/ / / /
/ \/ \ | ( / ‾‾\
/ \ | |\ \ | (‾) |
/ __________ \ |__| \__\ \_____/ .io
execution: local
script: k6/loadTest.js
output: -
scenarios: (100.00%) 8 scenarios, 200 max VUs, 4m7s max duration (incl. graceful stop):
* unsealedVirtualMediator: 200 looping VUs for 30s (exec: unsealedVirtualMediator)
* sealedVirtualMediator: 200 looping VUs for 30s (exec: sealedVirtualMediator, startTime: 31s)
* unsealedVirtual: 200 looping VUs for 30s (exec: unsealedVirtual, startTime: 1m2s)
* sealedVirtual: 200 looping VUs for 30s (exec: sealedVirtual, startTime: 1m33s)
* unsealed: 200 looping VUs for 30s (exec: unsealed, startTime: 2m4s)
* sealed: 200 looping VUs for 30s (exec: sealed, startTime: 2m35s)
* unsealedMediator: 200 looping VUs for 30s (exec: unsealedMediator, startTime: 3m6s)
* sealedMediator: 200 looping VUs for 30s (exec: sealedMediator, startTime: 3m37s)
running (4m07.0s), 000/200 VUs, 14281122 complete and 1600 interrupted iterations
unsealedVirtualMediator ✓ [======================================] 200 VUs 30s
sealedVirtualMediator ✓ [======================================] 200 VUs 30s
unsealedVirtual ✓ [======================================] 200 VUs 30s
sealedVirtual ✓ [======================================] 200 VUs 30s
unsealed ✓ [======================================] 200 VUs 30s
sealed ✓ [======================================] 200 VUs 30s
unsealedMediator ✓ [======================================] 200 VUs 30s
sealedMediator ✓ [======================================] 200 VUs 30s
✓ status code should be 200
checks...............................: 100.00% ✓ 15418988 ✗ 0
data_received........................: 2.5 GB 10 MB/s
data_sent............................: 1.5 GB 6.3 MB/s
http_req_blocked.....................: avg=1.49µs min=0s med=0s max=66.44ms p(90)=0s p(95)=0s
http_req_connecting..................: avg=84ns min=0s med=0s max=25.52ms p(90)=0s p(95)=0s
http_req_duration....................: avg=2.93ms min=0s med=2.99ms max=134.38ms p(90)=4ms p(95)=4.34ms
{ expected_response:true }.........: avg=2.93ms min=0s med=2.99ms max=134.38ms p(90)=4ms p(95)=4.34ms
http_req_failed......................: 0.00% ✓ 0 ✗ 15419018
http_req_receiving...................: avg=24.75µs min=0s med=0s max=116.68ms p(90)=0s p(95)=0s
http_req_sending.....................: avg=7.95µs min=0s med=0s max=102.26ms p(90)=0s p(95)=0s
http_req_tls_handshaking.............: avg=0s min=0s med=0s max=0s p(90)=0s p(95)=0s
http_req_waiting.....................: avg=2.9ms min=0s med=2.99ms max=134.38ms p(90)=4ms p(95)=4.29ms
http_reqs............................: 15419018 62423.760886/s
iteration_duration...................: avg=3.1ms min=0s med=3ms max=231.56ms p(90)=4ms p(95)=4.7ms
iterations...........................: 15418985 62423.627286/s
Sealed Counter.......................: 1993252 8069.663466/s
Sealed Duration......................: avg=2.856529 min=0 med=2.9996 max=95.3019 p(90)=3.9936 p(95)=4.1124
Sealed Mediator Counter..............: 1973171 7988.365711/s
Sealed Mediator Duration.............: avg=2.8589 min=0 med=2.9997 max=119.1916 p(90)=3.9953 p(95)=4.0759
Sealed Virtual Counter...............: 1958160 7927.593808/s
Sealed Virtual Duration..............: avg=2.904552 min=0 med=2.9998 max=104.7132 p(90)=3.9999 p(95)=4.2312
Sealed Virtual Mediator Counter......: 1866520 7556.590061/s
Sealed Virtual Mediator Duration.....: avg=3.022377 min=0 med=2.9999 max=91.5601 p(90)=4.0012 p(95)=4.5607
Unsealed Counter.....................: 1983031 8028.283836/s
Unsealed Duration....................: avg=2.869557 min=0 med=2.9997 max=92.4925 p(90)=3.9995 p(95)=4.1528
Unsealed Mediator Counter............: 1964774 7954.370529/s
Unsealed Mediator Duration...........: avg=2.894612 min=0 med=2.9997 max=134.3892 p(90)=3.9996 p(95)=4.1571
Unsealed Virtual Counter.............: 1927837 7804.831405/s
Unsealed Virtual Duration............: avg=2.93878 min=0 med=2.9998 max=94.942 p(90)=4.0002 p(95)=4.3455
Unsealed Virtual Mediator Counter....: 1752260 7094.00944/s
Unsealed Virtual Mediator Duration...: avg=3.191408 min=0 med=3.0001 max=94.7954 p(90)=4.301 p(95)=5.001
vus..................................: 200 min=0 max=200
vus_max..............................: 200 min=200 max=200
How To Run
From within the SealedVsUnsealed.Performance folder
To start the server
dotnet run --configuration Release
To start the performance tests
k6 run k6/run.js
Sealed By Default
Benefits
- Sealed classes can offer some performance benefits, especially when a class has overwritten virtual methods.
- Encourages developers to think more about the design of classes and how they are intended to be consumed.
- Sealed classes can easily be unsealed if design requirements change.
Quotes
I know many people (including myself) who believe that classes should indeed be sealed by default.
– Jon Skeet, 2008
Which is why sealed should be the default for classes, of course.
– Jon Skeet, 2018
Please yes!
– Random HN Reader, 2019
Additional Resources
Pendulum swing: sealed by default By Mark Seemann
Why all your classes should be sealed by default in C# By Nick Chapsas
Unsealed By Default
Benefits
- Unsealed classes can be extended, allowing developers to build upon existing functionality.
- Using the sealed keyword selectively, emphasizes clear intent that a class should not be extended, making the keyword more meaningful and impactful.
Quotes
DO NOT seal classes without having a good reason to do so.
– MSDN, 2022
#final #java I use it for constants, but nothing else. I don't declare classes are arguments final.
– Uncle Bob, 2011
Please, no! This shouldn't be the default - it's a constant bugbear of mine where I want to extend a class from a library, and I can't because it's been sealed for no good reason.
– Random HN reader, 2019
Additional Resources
Framework design guidelines - Sealing By Microsoft
Seal By Martin Fowler
Conclusion
After going through all this, I still believe that classes should be sealed by default. While the performance benefits are often minimal, sealing a class provides clear guidance on its intended use. This approach promotes a more modular and decoupled codebase by favouring composition over inheritance.