On Friday, 24 June 2022 at 14:30:54 UTC, Zoadian wrote:

The Simula class concept is to a large extent meant to support changing the model later. So it is sensible for simulation where you experiment (programming cost being higher than hardware costs). This is basically where inheritance comes in. You have a working model, then someobody tells you: we actually need to model not only Doctors, but also Nurses, AmbulanceDrivers etc. How quickly can you accomodate that wish? This is also a strength of relation-based databases, just add more tables.

I don't think existing languages have good support for hardware. C++/C requires the compiler to use the same layout for fields in a record as the source. That is a bummer for inheritance in terms of cache-lines, but that is not an OO weakness, it is a language weakness.

If we talk high level OO, then there should be no hard relationship between the code and the layout (e.g. it could broken down to structs of arrays or arrays of structs or components etc by the compiler based on hints or analysis).

Main stream languages are very primitive still… Outside of Rust, many languages seems to be satisfied by adding a thin layer over LLVM…

Sure, you can implement an OO-model in C or assembly. Most games of the 80s probably had some ad hoc OO-model implemented in clever ways in assembly to pack it into 32/64K RAM :-)

On Friday, 24 June 2022 at 15:14:44 UTC, Ola Fosheim Grøstad wrote:

On a personal note: I also use have a default preference for OO when that makes it possible to avoid documentation because the implementation has a 1-1 correspondence with the model. Then you don't have to deal with out-of-date documentation and the boring task of writing documentation.

This is also desirable when you evolve software rather than plan it (experimental prototyping or simulation). If you expect a lot of successive changes then documentation becomes a real burden.

On Friday, 24 June 2022 at 15:14:44 UTC, Ola Fosheim Grøstad wrote:

Please show me an alternative that allows to respect the substitution principle.

On Friday, 24 June 2022 at 18:08:15 UTC, user1234 wrote:

The simple approach is to reserve space for the fields you want to be on the same cache line. The root cause for the C++ model is separate compilation. If the compiler knows the full class hierarchy then it can choose a different layout. That is obvious.

On Saturday, 25 June 2022 at 18:05:31 UTC, Paul Backus wrote:

What debate? You can easily represent objects in a relational database. One table for the superclass, another one for the subclass extension. The main discussion has really been about whether you want relational or hierarchical structure. The latter is better for performance. It is fairly easy to distribute objects or hierarchical, but relational is harder, yet very flexible. Keep in mind that XML databases often are implemented as relational tables, and you can represent objects as XML with ease…

The real clash is really with lower level languages adding OOP, as OOP is very much a high level concept. Java and C# don't really improve a whole lot on the runtime organization over Simula. Which is sad. Still primitive in other words.

On Saturday, 25 June 2022 at 18:23:20 UTC, Ola Fosheim Grøstad wrote:

In other words, C++ messed up OOP by exposing the implementation and thereby depriving the compiler from optimization opportunities. OOP does not dictate a particular organization in memory.