What makes a StableInterface?
please add -- PeterSommerlad
- it is supported by a GreatAbstraction
- it incorporates DesignForSlippage
- it is easy to use
- it avoids too strong typing (TypeSafeInterfacesConsideredHarmful)
- it provides good default behavior, even if it is misused
- it doesn't leak implementation details, such as data structures
- it can serve implementations and clients far beyond its initial intention
- everybody understands it and its limitations
- it was invented by its users and used by its inventors
- it is usually a result of rigorous refactoring or designed by very great practical thinkers that are thinking practitioners
- it lacks any really bad features, so no-one feels driven to fix them
- it lacks any really obvious gaps or missing features, so no-one feels driven to add them
- it follows the LawOfDemeter
Tongue in cheek answers:
- It's too hard to change
- Nobody's using it, so no need to change
- It is not permitted to change, everyone codes work arounds to use it
What else makes a StableInterface an AbsoluteJoyToUse? -- DavidBennett
- it can be queried for additional services or extensions. (This PatLet is used in Unix file I/O, X11, OpenGL, COM, CORBA etc.) In PoSaTwo this is called ExtensionInterface (aka ExtensionObject).
- it can be queried for information about itself (eg COM/TLB, .NET)
Can we make a pattern from this stuff?
I think that a StableInterface is a synergistic result of several patterns --NatPryce.
But is a stable interface always worth the cost?
see LeastFlexibleProtocolWins, ImplementationIndependenceLimits