5 ms·

Generic functions like in Java/C# can be implemented efficiently and with zero overhead. There is nothing that prevents a JIT compiler to generate several versi

Generic functions like in Java/C# can be implemented efficiently and with zero overhead. There is nothing that prevents a JIT compiler to generate several versions of the code optimised for different types at runtime, exactly like C++ does at compile-time. Some VMs already do that to some extent in order e.g. to better inline calls for the generic argument. Still, those optimisations are in their infancy, but the situation improves with every year. "Don't get me wrong. I am quite happy with this making my skills very rare and expensive, giving me both job security and decent compensation." Eh? C++ skills are not any more expensive than decent Java/C#/Python/(put whatever mainstream language here) skills. Who told you that? This is supply-demand. There is lower supply of C++ coders, so some may think the prices should be higher, but in fact it is compensated by lower demand. C++ is also still being taught at schools, so those skills are not that rare as, say, Haskell, Erlang or Go.

pandaman 13y ago
How do you figure out the types at run time with zero overhead? I guess we have different concept of "zero". As for the price of my skills - my paycheck told me that. And sure, it could be low demand (there are vacancies in my field, that had been open for years, but I don't really know what kind of demand there is for Python, could be decades for all I care). I am just happy for my compensation and if others have more expensive skills - I am happy for them too :)
pkolaczk 13y ago
Nothing is truly zero overhead. C++ templates are neither, because they create some overhead in code size which is obviously non-free (both space-wise and time-wise). Zero-overhead for me = optimized to the point that noone can notice. There is no problem for a VM to generate 10 versions of a tight loop, every one optimised for particular type, with a switch for the type right before entering the loop, or the whole method, or even a whole call-tree.
pandaman 13y ago
We also mean different things by overhead. For me (and the C++ committee) it means something that is done in addition. So for me templates create zero overhead as the do not produce more code than the same, non-template code. In practice, templates produce less code as there is a higher chance to fold common code from the instantiations of the same template than from many handwritten functions.
- pkolaczk 13y ago
  Comparing to other generics implementations - they do add a lot of code, much more than is really needed to achieve good performance. Sure, some of this code can be reduced at linking stage, but for this to work you need to statically link everything which has its own set of problems (overhead of repeated code in every binary). IMHO the approach is a great example of premature optimisation: generate 2^N versions then try to fold some similar ones to save space. Much better approach is: generate 1 version and specialize small parts when really needed. Technically, assuming sufficiently smart compiler, both would end up with exactly the same code, performing exactly the same, but the latter solution is preferable for other non-performance wise reasons (like better type checking, ability to distribute binary libraries, faster compile times, etc.). BTW: I still often find Scala containers to be often much faster than C++ STL. But this is probably for much different reasons than template/generics implementation.
  - pandaman 13y ago
    I have lost you. Are you still arguing that C++ is internally inconsistent because it's not doing stuff that Java does? Are you arguing that templates add a lot more code than a VM and JIT compiler and languages with generic functions are as efficient as C++? In the first case - I fail to see how it follows. In the second case - our levels of expertise is too far apart to discuss anything. I am not saying you are not qualified, I might be just too out of the loop and missing some dramatic advances in Java, Scala etc.

pandaman 13y ago

How do you figure out the types at run time with zero overhead? I guess we have different concept of "zero". As for the price of my skills - my paycheck told me that. And sure, it could be low demand (there are vacancies in my field, that had been open for years, but I don't really know what kind of demand there is for Python, could be decades for all I care). I am just happy for my compensation and if others have more expensive skills - I am happy for them too :)

pkolaczk 13y ago

Nothing is truly zero overhead. C++ templates are neither, because they create some overhead in code size which is obviously non-free (both space-wise and time-wise). Zero-overhead for me = optimized to the point that noone can notice. There is no problem for a VM to generate 10 versions of a tight loop, every one optimised for particular type, with a switch for the type right before entering the loop, or the whole method, or even a whole call-tree.

pandaman 13y ago

We also mean different things by overhead. For me (and the C++ committee) it means something that is done in addition. So for me templates create zero overhead as the do not produce more code than the same, non-template code. In practice, templates produce less code as there is a higher chance to fold common code from the instantiations of the same template than from many handwritten functions.

pkolaczk 13y ago
Comparing to other generics implementations - they do add a lot of code, much more than is really needed to achieve good performance. Sure, some of this code can be reduced at linking stage, but for this to work you need to statically link everything which has its own set of problems (overhead of repeated code in every binary). IMHO the approach is a great example of premature optimisation: generate 2^N versions then try to fold some similar ones to save space. Much better approach is: generate 1 version and specialize small parts when really needed. Technically, assuming sufficiently smart compiler, both would end up with exactly the same code, performing exactly the same, but the latter solution is preferable for other non-performance wise reasons (like better type checking, ability to distribute binary libraries, faster compile times, etc.). BTW: I still often find Scala containers to be often much faster than C++ STL. But this is probably for much different reasons than template/generics implementation.
- pandaman 13y ago
  I have lost you. Are you still arguing that C++ is internally inconsistent because it's not doing stuff that Java does? Are you arguing that templates add a lot more code than a VM and JIT compiler and languages with generic functions are as efficient as C++? In the first case - I fail to see how it follows. In the second case - our levels of expertise is too far apart to discuss anything. I am not saying you are not qualified, I might be just too out of the loop and missing some dramatic advances in Java, Scala etc.