digitalmars.D.learn - alias parameters, what for?
- XavierAP (18/18) May 09 2019 What are the benefits of alias parameters, compared to specifying
- Simen =?UTF-8?B?S2rDpnLDpXM=?= (35/53) May 09 2019 Genericity, for one. This code will work with functions,
- XavierAP (6/6) May 09 2019 Thanks, I get your points. I do think they make more sense for
- Dennis (9/12) May 09 2019 With function pointers, there is extra indirection and the
What are the benefits of alias parameters, compared to specifying the template parameters fully? https://dlang.org/spec/template.html#aliasparameters In most examples, at places in Phobos, and in Andrei's and Ali’s books, alias parameters are used for functions (in the general sense). Why this instead of specifying and typing the parameter functions or delegates? This brings another question, why is it so widespread in Phobos etc. to template these function parameters instead of declaring them as run-time parameters? Is this really always considered beneficial, why? For one it looks like it saves a lot of typing and makes the whole declaration more readable (not to mention possible attribute soups); but the type-checking code just moves to additional template constraints. And finally what’s the advantage of alias parameters in general, besides function/delegates? But I don’t see other uses in practice, although they’re possible.
May 09 2019
On Thursday, 9 May 2019 at 09:52:21 UTC, XavierAP wrote:What are the benefits of alias parameters, compared to specifying the template parameters fully? https://dlang.org/spec/template.html#aliasparameters In most examples, at places in Phobos, and in Andrei's and Ali’s books, alias parameters are used for functions (in the general sense). Why this instead of specifying and typing the parameter functions or delegates?Genericity, for one. This code will work with functions, delegates, templated functions, lambdas, and functors: void call(alias fun)() { fun(); }This brings another question, why is it so widespread in Phobos etc. to template these function parameters instead of declaring them as run-time parameters? Is this really always considered beneficial, why?The compiler can optimize calls to compile-time parameters, but not run-time parameters. Also, as said above, alias parameters support a greater variety of types. If instead run-time parameters were used there'd have to be multiple overloads, or a templated function that accepts all the different callables.For one it looks like it saves a lot of typing and makes the whole declaration more readable (not to mention possible attribute soups); but the type-checking code just moves to additional template constraints.The type checking tends to be a lot looser though - more along the lines of 'Can I call f with these parameters?', rather than 'Is f a function taking two ints and a string, returning a bool, and adorned with safe and nogc?'. There are tradeoffs there, of course, and you should use whichever is more correct for your use case.And finally what’s the advantage of alias parameters in general, besides function/delegates? But I don’t see other uses in practice, although they’re possible.Templates can be passed in alias parameters. This is used quite extensively in std.meta. In fact, alias parameters accept anything* with a name. This makes it easy to build templates that operate on any type or scope (e.g. a module) with the exact same code, and passing them from one template to the other without having to rewrite them to operate on specific types of objects. Another use case is to pass a value of an unknown type. Without alias parameters, you'd have to pass the type as well, something like foo!(int, 3) instead of foo!3, or use a template sequence parameter and a template constraint: template foo(Value...) if (Value.length == 1) {} This is clearly untenable for more than a single such parameter. In my own code, I've almost exclusively used alias parameters for templates, with functions being the second most common, and one or two examples of the module/types mentioned above. * except built-in primitive types -- Simen
May 09 2019
Thanks, I get your points. I do think they make more sense for the standard library, than in every general case (packages for specific uses). Namely, alias parameters provide absolute genericity (instead of overloading every possible use case, or else constraining the API by design), and ultimate runtime performance (always at the expense of compile time).
May 09 2019
On Thursday, 9 May 2019 at 09:52:21 UTC, XavierAP wrote:alias parameters are used for functions (in the general sense). Why this instead of specifying and typing the parameter functions or delegates?With function pointers, there is extra indirection and the function can not be inlined. With alias parameters, something like map!(x => x*2) can be as fast as a hand-written for-loop. A function pointer would make it significantly slower since every multiplication adds a new function call. You can not have a variable map function to the alias parameter, but most often you don't need that. If you do, you can simply do map!(x => dynamicFunc(x)).
May 09 2019