On Wed, Mar 9, 2016 at 10:03 AM, Steven Watanabe wrote:

AMDG

On 03/09/2016 07:49 AM, Zach Laine wrote:

...

However, I *am* concerned by Steven Watanabe's claim that the library is ADL-unsafe. He did not state exactly why that I could tell, but this

needs

...

to be addressed if so.

There are two issues (I pointed them out as I ran across them in the line-by-line part of my review.) - There are a unqualified calls in many places where ADL is not intended. - The detail namespace is an associated namespace of quite a few components.

Ah, I missed that I guess. Thanks, Steven. Zach

On Thursday, March 10, 2016 at 12:34:26 AM UTC-6, Vicente J. Botet Escriba wrote:

Le 10/03/2016 01:12, Paul Fultz II a écrit :

...

On Wednesday, March 9, 2016 at 8:50:01 AM UTC-6, Zach Laine wrote:

...

There are a number of adaptors that I find have obscure names:

by() is a projection; why not call it project()?

I chose `by` because of how it is used linguistically. That is you

write:

...

`sort(v.begin(), v.end(), by(&Employee::Name, _<_))` which I read "sort by employee name". Also, if I write `by(decay, constructstd::tuple())`, I read "construct tuple by decay". In haskell, it uses `on` for this adaptor, which is what I originally called it. `project` is a more straighforward name. `by` is a comparator while `projet` is a projection.

By is a projection in the example of `construct`.

...

...

compress() is defined in terms of "fold", with a link to the Wikipedia page for same; why not call it foldl()? Similarly, reverse_compress()

should

...

...

be foldr().

Not `foldr`. This is because `foldr` is symetrical. For example, `foldl` and `foldr` should produce the same results:

foldl(_+_, std::string())("Hello", "-", "world"); // "Hello-world" foldr(_+_, std::string())("Hello", "-", "world"); // "Hello-world" It would be good to use a binary function that has two different types to see the difference in behavior.

Yes, you could do `flip(_+_)` to get the same effect. However, I want the callback to be consistent with compress reverse_compress, so it always is called with f(state, element). Also, `fold` and `reverse_fold` are commonly used names in C++, so I would like to keep it more consistent with those usage. If the user wants `fold_right`, it can be easily written, like this: `auto fold_right = compose(compress, flip)`.

...

However, compress and reverse_compress work like this:

compress(_+_, std::string())("Hello", "-", "world"); // "Hello-world" reverse_compress(_+_, std::string())("Hello", "-", "world"); // "world-Hello"

I was reluctant to call it `fold` as it seems to imply some data structure to fold over, whereas this is simply an adaptor. I used the word compress as its a another name used for fold. However, it seems more people would prefer to use `fold` and `reverse_fold`. You are folding the parameter pack. I like those. and I would like also fold_left, fold_right.

...

flow() is a super obscure name! Why is this not reverse_compose() or similar?

I actually got the name from underscore.js. The reason why I chose this name instead of `reverse_compose` is because when this is used with pipable functions it seems confusing:

reverse_compose( filter([](int i) { return i < 3; }), transfrom([](int i) { return i*i; }) )(numbers);

With the word 'reverse' there, it almost looks as if it computes the pipeline in reverse order, which it doesn't. So I would prefer a name without reverse in it. Would pipe works better?

pipe could work, I don't know how others feel about such a name.

...

...

indirect() could more easily be understood if it were called deref() or dereference().

Well, `indirect` is how it is commonly called in libraries such as Boost.Range, range-v3, and PStade libraries. So I would like to keep the name consistent with other similar constructs.

Could you tell us what indirect is in range-v3, what is the signature and the semantics. I don't see it is used for the same purposes, but maybe I'm wrong.

It dereferences the values in a range. They both are functors that will dereference the parametrized type.

ref(flv) is a callable that wraps an lv indirect(ptr) is a callable that wraps a ptr and deref it before calling

Well, it can be used with anything that is dereferenceable. So it works with boost::optional as well.

deref/dereference will only relate the dereference action, but not the call.

deref_on_call is too long?

Yes it is.

...

...

I don't feel as strongly about partial(), but I think it might be clearer if it were called partial_apply() or curry().

Hmm, I don't think a name with `apply` in it is good name for an adaptor. Maybe you could explain why curry is not a good name here? What is the difference between partial an currying a function.

...

conditional() should be called invoke_first(), call_first_of(), or similar. I find it too easy to confuse with if_().

I agree conditional is not a good name.

...

I see, I was trying to describe an adaptor where you could put functions with conditions in it. Other people, seem to prefer a name like `linear`, so I might use that instead. This function is related to match. The difference is that one select the best matching overload and those must be exclusive and the other the first matching overload and the match can be inclusive. I would like to see the semantics on the name, but I have not a concrete proposal.

We need to take in account that this is an adaptor, so there is no call, so `invoke_first` or call_first_off will not work. Those functions creates another function object that applies a different algorithm to select the overloaded functions when called.

I've used overload -> match first_overload -> conditional

I would prefer `match` and `linear`.

...

...

* Documentation

The Quick Start section is good at showing a compelling motivating case for using the library. The result at the end of the section looks like a

very

...

good start on a "Scrap Your Boilerplate" solution for simple, dump-style printing. It's relatively easy to read and write.

The documentation is what needs the most work, from what I've seen. The Quick Start is quite nice, but then there are some things that are underexplained in the subsequent Overview and Basic Concepts sections. For instance, why are these macros used everywhere and what do they do? They should be used for the first time after showing what the expanded code looks like.

I don't think the docs should show the expansion of the macros, that is

I prefer those, but I'm not yet happy with. Naming is difficult. part

...

of the implementation and not interface. I could show an "idea" what is is expanding to, with explanation of what else it is doing beyond the simple explanation. I agree hat it is absolutely needed to show the exact expansion, but it is clear that the user wants to know what is behind the scenes.

I strongly disagree. Documentation is about documenting the interface, not the implementation. Mainly because the implementation could change or vary between platforms while the interface would remain the same. If the user wants to know about the implementation, they can look at the source code or in the case of macros, look at the preprocessed output.

...

...

This will justify their use for those users that want them, and show those that don't

...

...

they can write instead.

I could show an alternative without the macros, but I would prefer to

what put

...

that in the Advance section. In the introduction of the library, I would rather show the constructs that can be easily used without a need to explain a bunch of caveats. I will put all of them in the advanced section. I would no mention them in the introduction as the user can create the HOF locally or use a factory.

Being able to declare the functions as global variables is a key feature of the library. The example in the quick start guide shows how to implement many things without the need for a lot of template boilerplate. For those users that prefer not to use global function objects(and I have yet seen a compelling reason not to use them) can look at the 'Advanced' section. I realize now, that I need to spend more discussing the advantages of using global function objects(more composability), and address some of the misperceived issues with these global function objects(they can be in a namespace to avoid name conflicts and are not mutable so it won't be a problem in mutilthreaded environments).

Best, Vicente

_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost

On Thursday, March 10, 2016 at 5:26:38 PM UTC-6, Edward Diener wrote:

On 3/10/2016 1:46 PM, Paul Fultz II wrote:

...

On Thursday, March 10, 2016 at 12:34:26 AM UTC-6, Vicente J. Botet

...

wrote:

...

Le 10/03/2016 01:12, Paul Fultz II a écrit :

...

On Wednesday, March 9, 2016 at 8:50:01 AM UTC-6, Zach Laine wrote:

...

There are a number of adaptors that I find have obscure names:

by() is a projection; why not call it project()?

I chose `by` because of how it is used linguistically. That is you

write:

...

`sort(v.begin(), v.end(), by(&Employee::Name, _<_))` which I read "sort by employee name". Also, if I write `by(decay, constructstd::tuple())`,

I

...

...

read "construct tuple by decay". In haskell, it uses `on` for this adaptor, which is what I originally called it. `project` is a more straighforward name. `by` is a comparator while `projet` is a projection.

By is a projection in the example of `construct`.

...

...

...

compress() is defined in terms of "fold", with a link to the Wikipedia page for same; why not call it foldl()? Similarly, reverse_compress()

should

...

...

be foldr().

Not `foldr`. This is because `foldr` is symetrical. For example, `foldl` and `foldr` should produce the same results:

foldl(_+_, std::string())("Hello", "-", "world"); // "Hello-world" foldr(_+_, std::string())("Hello", "-", "world"); // "Hello-world" It would be good to use a binary function that has two different types to see the difference in behavior.

Yes, you could do `flip(_+_)` to get the same effect. However, I want the callback to be consistent with compress reverse_compress, so it always is called with f(state, element). Also, `fold` and `reverse_fold` are commonly used names in C++, so I would like to keep it more consistent with those usage. If the user wants `fold_right`, it can be easily written, like

...

`auto fold_right = compose(compress, flip)`.

...

...

However, compress and reverse_compress work like this:

compress(_+_, std::string())("Hello", "-", "world"); // "Hello-world" reverse_compress(_+_, std::string())("Hello", "-", "world"); // "world-Hello"

I was reluctant to call it `fold` as it seems to imply some data structure to fold over, whereas this is simply an adaptor. I used the word compress as its a another name used for fold. However, it seems more people would

...

...

to

...

use `fold` and `reverse_fold`. You are folding the parameter pack. I like those. and I would like also fold_left, fold_right.

...

flow() is a super obscure name! Why is this not reverse_compose() or similar?

I actually got the name from underscore.js. The reason why I chose this name instead of `reverse_compose` is because when this is used with pipable functions it seems confusing:

reverse_compose( filter([](int i) { return i < 3; }), transfrom([](int i) { return i*i; }) )(numbers);

With the word 'reverse' there, it almost looks as if it computes the pipeline in reverse order, which it doesn't. So I would prefer a name without reverse in it. Would pipe works better?

pipe could work, I don't know how others feel about such a name.

...

...

...

indirect() could more easily be understood if it were called deref()

or

...

...

dereference().

Well, `indirect` is how it is commonly called in libraries such as Boost.Range, range-v3, and PStade libraries. So I would like to keep

Escriba this: prefer the

...

...

...

name consistent with other similar constructs.

Could you tell us what indirect is in range-v3, what is the signature and the semantics. I don't see it is used for the same purposes, but maybe I'm wrong.

It dereferences the values in a range. They both are functors that will dereference the parametrized type.

...

ref(flv) is a callable that wraps an lv indirect(ptr) is a callable that wraps a ptr and deref it before calling

Well, it can be used with anything that is dereferenceable. So it works with boost::optional as well.

...

deref/dereference will only relate the dereference action, but not the call.

deref_on_call is too long?

Yes it is.

...

...

...

I don't feel as strongly about partial(), but I think it might be clearer if it were called partial_apply() or curry().

Hmm, I don't think a name with `apply` in it is good name for an adaptor. Maybe you could explain why curry is not a good name here? What is the difference between partial an currying a function.

...

conditional() should be called invoke_first(), call_first_of(), or similar. I find it too easy to confuse with if_().

I agree conditional is not a good name.

...

I see, I was trying to describe an adaptor where you could put functions with conditions in it. Other people, seem to prefer a name like `linear`, so I might use that instead. This function is related to match. The difference is that one select the best matching overload and those must be exclusive and the other the first matching overload and the match can be inclusive. I would like to see the semantics on the name, but I have not a concrete proposal.

We need to take in account that this is an adaptor, so there is no call, so `invoke_first` or call_first_off will not work. Those functions creates another function object that applies a different algorithm to select the overloaded functions when called.

I've used overload -> match first_overload -> conditional

I would prefer `match` and `linear`.

...

I prefer those, but I'm not yet happy with. Naming is difficult.

...

...

* Documentation

The Quick Start section is good at showing a compelling motivating

case

...

...

...

for using the library. The result at the end of the section looks like a very good start on a "Scrap Your Boilerplate" solution for simple, dump-style printing. It's relatively easy to read and write.

The documentation is what needs the most work, from what I've seen. The Quick Start is quite nice, but then there are some things that are underexplained in the subsequent Overview and Basic Concepts sections. For instance, why are these macros used everywhere and what do they do? They should be used for the first time after showing what the expanded code looks like.

I don't think the docs should show the expansion of the macros, that is part of the implementation and not interface. I could show an "idea" what is is expanding to, with explanation of what else it is doing beyond the simple explanation. I agree hat it is absolutely needed to show the exact expansion, but it is clear that the user wants to know what is behind the scenes.

I strongly disagree. Documentation is about documenting the interface, not the implementation. Mainly because the implementation could change or vary between platforms while the interface would remain the same. If the user wants to know about the implementation, they can look at the source code or in the case of macros, look at the preprocessed output.

...

...

...

This will justify their use for those users that want them, and show those that don't

...

...

they can write instead.

I could show an alternative without the macros, but I would prefer to

what put

...

that in the Advance section. In the introduction of the library, I would rather show the constructs that can be easily used without a need to explain a bunch of caveats. I will put all of them in the advanced section. I would no mention them in the introduction as the user can create the HOF locally or use a factory.

Being able to declare the functions as global variables is a key feature of the library. The example in the quick start guide shows how to implement many things without the need for a lot of template boilerplate. For those users that prefer not to use global function objects(and I have yet seen a compelling reason not to use them) can look at the 'Advanced' section.

I realize now, that I need to spend more discussing the advantages of using global function objects(more composability), and address some of the misperceived issues with these global function objects(they can be in a namespace to avoid name conflicts and are not mutable so it won't be a problem in mutilthreaded environments).

You do need to explain why "global variables" is a key feature of the library. I cannot understand from your docs why there is any disadvantage instantiating Callables locally as opposed to globally other than the usual fact that a local object can go out of scope.

We can define functions by simply composing other functions together and we don't need to write awkward syntax nor template boilerplate for it. For example, we can simply write `make_tuple` like this: BOOST_FIT_STATIC_FUNCTION(make_tuple) = by(decay, constructstd::tuple()); Alternatively, you could write a factory function for this instead: constexpr auto make_tuple() { return by(decay, constructstd::tuple()); } Of course this requires the user to write `make_tuple()(xs...)`, which I don't think any user would expect this, especially since `std::make_tuple` doesn't work like that. So instead, it can be wrapped in a template function: template constexpr auto make_tuple(Ts&&... xs) -> declype(by(decay, constructstd::tuple())(std::forward<Ts>(xs)...)) { return by(decay, constructstd::tuple())(std::forward<Ts>(xs)...); } Which is a lot of boilerplate to compose some simple functions.

If there really is some other reason it is completely lost to me. If you would like to point to me in your doc where you explain the use of "global variables" as being a key feature of your library or as being necessary to use the functionality of your library, I would be glad to read about it and ask further about it here.

It is not a necessary feature, but it is an important feature nonetheless, and I need to spend more time explaining its usefulness.

If, OTOH, it is just your preference to use global objects as opposed to the various forms of local instantiation, I really wish you would just say that rather than acting like your library does not work correctly somehow other than with global variables.

The composability of the adaptors and functions applies to both global and local functions. The adaptors provide a simple way to create functions to be passed locally, such as creating a comparator for std::sort(ie `std::sort(first, last, by(&employee::name, _<_))`). However, they can also be used to define global/free functions in a much simpler way as well(such as writing `std::make_tuple`), which I think is equally important. Furthermore, in a quick start guide or introduction, I want to be able to demonstrate the capabilities of the library in a way to show its usefulness, and why someone would choose to use this library. Perhaps, you are ok with awkward syntax or template boilerplate in writing your functions, and would prefer to only use the adaptors for local functions. However, there are others who will find writing global function objects with the adaptors very useful, so I want to be able to show those capabilities in the introduction of the library. I do, however, need to discuss some of the misperceived issues with using global function objects in the documentation. As the issues raised in the review were: 1) Namespacing - This is not a problem because all global function objects can be placed in a namespace. 2) Global state - This is not a problem either, because `BOOST_FIT_STATIC_FUNCTION` declares the function object const with constexpr. So there can be no mutable or changing state internally in the object, and it must be constructed without side effects. I think addressing these concerns in the documentation will help put users at ease when using global function objects, so they have no problem taking advantage of the adaptors to build function objects globally.

Maybe I have missed something but I have the intuition that many others have missed it likewise, from the responses of others about this issue.

I agree with your intuition, and I believe the documentation should spend more time discussing these issues. I hope my email has made it somewhat clearer to you.

...

We can define functions by simply composing other functions together and

we

...

don't need to write awkward syntax nor template boilerplate for it. For example, we can simply write `make_tuple` like this:

BOOST_FIT_STATIC_FUNCTION(make_tuple) = by(decay, constructstd::tuple());

What is the type of by(decay,constructstd::tuple()) ?

It is something like `by_adaptorstd::tuple>`.

I thought I would be able to write locally:

auto make_tuple(by(decay,constructstd::tuple()));

but evidently you are saying that does not work.

That is not what I am saying at all. It can work, the functions can be used locally, like in the first example. However, for `make_tuple` this is not useful at all. If `std::make_tuple` was assigned to a local variable then no one could ever call the function. So you use global function objects when you want to make the function available for consumption by others. And the advantage of defining `make_tuple` this way over writing it as it is written currently in the standard library, is that we can avoid all the template boilerplate involved with this.

...

Alternatively, you could write a factory function for this instead:

constexpr auto make_tuple() { return by(decay, constructstd::tuple()); }

Of course this requires the user to write `make_tuple()(xs...)`, which I don't think any user would expect this, especially since `std::make_tuple`

...

work like that. So instead, it can be wrapped in a template function:

template constexpr auto make_tuple(Ts&&... xs) -> declype(by(decay, constructstd::tuple())(std::forward<Ts>(xs)...)) { return by(decay, constructstd::tuple())(std::forward<Ts>(xs)...); }

Which is a lot of boilerplate to compose some simple functions.

...

If there really is some other reason it is completely lost to me. If you would like to point to me in your doc where you explain the use of "global variables" as being a key feature of your library or as being necessary to use the functionality of your library, I would be glad to read about it and ask further about it here.

It is not a necessary feature, but it is an important feature nonetheless, and I need to spend more time explaining its usefulness.

...

If, OTOH, it is just your preference to use global objects as opposed to the various forms of local instantiation, I really wish you would just say that rather than acting like your library does not work correctly somehow other than with global variables.

The composability of the adaptors and functions applies to both global and local functions. The adaptors provide a simple way to create functions to be passed locally, such as creating a comparator for std::sort(ie `std::sort(first, last, by(&employee::name, _<_))`). However, they can also be used to define global/free functions in a much simpler way as well(such as writing `std::make_tuple`), which I think is equally important.

Furthermore, in a quick start guide or introduction, I want to be able to demonstrate the capabilities of the library in a way to show its usefulness, and why someone would choose to use this library. Perhaps, you are ok with awkward syntax or template boilerplate in writing your functions, and would prefer to only use the adaptors for local functions. However, there are others who will find writing global function objects with the adaptors very useful, so I want to be able to show those capabilities in the introduction of

doesn't the

...

library. I do, however, need to discuss some of the misperceived issues with using global function objects in the documentation. As the issues raised in the review were:

1) Namespacing - This is not a problem because all global function objects can be placed in a namespace.

2) Global state - This is not a problem either, because `BOOST_FIT_STATIC_FUNCTION` declares the function object const with constexpr. So there can be no mutable or changing state internally in the object, and it must be constructed without side effects.

I think addressing these concerns in the documentation will help put users at ease when using global function objects, so they have no problem taking advantage of the adaptors to build function objects globally.

...

Maybe I have missed something but I have the intuition that many others have missed it likewise, from the responses of others about this issue.

I agree with your intuition, and I believe the documentation should spend more time discussing these issues. I hope my email has made it somewhat clearer to you.

_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost

Le 11/03/2016 07:18, Edward Diener a écrit :

On 3/11/2016 12:10 AM, Paul Fultz II wrote:

...

...

...

We can define functions by simply composing other functions together and

we

...

don't need to write awkward syntax nor template boilerplate for it. For example, we can simply write `make_tuple` like this:

BOOST_FIT_STATIC_FUNCTION(make_tuple) = by(decay, constructstd::tuple());

What is the type of by(decay,constructstd::tuple()) ?

It is something like `by_adaptorstd::tuple>`.

...

I thought I would be able to write locally:

auto make_tuple(by(decay,constructstd::tuple()));

but evidently you are saying that does not work.

That is not what I am saying at all. It can work, the functions can be used locally, like in the first example. However, for `make_tuple` this is not useful at all. If `std::make_tuple` was assigned to a local variable then no one could ever call the function.

I don't understand this. I have a 'make_tuple' callable above and if I want to use it somewhere else I pass it to whatever functionality needs it. You, instead want things to be global so that they can be used by other functionality without passing anything around. Isn't this just an argument about programming style ? I don't mind if you like your programming style but I doubt I am the first or last programmer who prefers variables created and used when needed. In fact I have a very strong suspicion that "my" programming style is that of the majority of C++ programmers and that only a minority of programmers like global variables even if they can be global within a namespace scope. So what I am essentially saying is that while I understand you like your style and your macros for creating "global" callables, inferring that this is central to using your library, when it is not AFAICS, is going to lose you end-users as well as misrepresent your library. I see much more plusses to your library if the result of using your adapters, decorators, functions, and utilities can be instantiated anywhere and used in ways that any variable in C++ can be used. if however there are real technical limitations, as opposed to "I like this style because I it is easier for me to use with Fit", of using Fit functionality to create "local" objects I think you need to spell out what those limitations are. Eduard, I don't know if you are aware of the Eric's paper [1] about customization points. In this papers Eric presents the technique use by Fit to define global function objects. These global variables have no modifiable state and are just like functions, but are high order functions. Eric received a very good feedback from C++ committee and AFAIK the range-v3 library [2] and the Range TS [3] are based on this technique.

I believe that it will be better if Paul show what is behind those macros by writing what the user could do without. E.g. the static_const ODR helper lets the user just write |constexpr auto const & begin = boost::static_const<__detail::__begin_fn>;| Paul goes one step far, by defining some macros that help to do that. Best, Vicente [1] N4381 - Suggested Design for Customization Points http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2015/n4381.html [2] https://github.com/ericniebler/range-v3 [3] N4569 - Proposed Ranges TS working draft http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2016/n4569.pdf

On Friday, March 11, 2016 at 12:19:15 AM UTC-6, Edward Diener wrote:

On 3/11/2016 12:10 AM, Paul Fultz II wrote:

...

...

...

We can define functions by simply composing other functions together

...

...

we

...

don't need to write awkward syntax nor template boilerplate for it. For example, we can simply write `make_tuple` like this:

BOOST_FIT_STATIC_FUNCTION(make_tuple) = by(decay, constructstd::tuple());

What is the type of by(decay,constructstd::tuple()) ?

It is something like `by_adaptorstd::tuple>`.

...

I thought I would be able to write locally:

auto make_tuple(by(decay,constructstd::tuple()));

but evidently you are saying that does not work.

That is not what I am saying at all. It can work, the functions can be used locally, like in the first example. However, for `make_tuple` this is not useful at all. If `std::make_tuple` was assigned to a local variable

and then no

...

one could ever call the function.

I don't understand this. I have a 'make_tuple' callable above and if I want to use it somewhere else I pass it to whatever functionality needs it. You, instead want things to be global so that they can be used by other functionality without passing anything around. Isn't this just an argument about programming style ?

Its not about programming style. Its about showing what the library is fully capable of doing. For example, say you are library writer and have just written a tuple class: template struct tuple { ... }; Now you would like to provide a `make_tuple` to make it easier for users to construct the tuple. Traditionally, you would write a function like this: template constexpr auto make_tuple(Ts&&... xs) -> decltype(tuple(std::forward<Ts>(xs)...)) { return tuple(std::forward<Ts>(xs)...); } However, there is a lot of boilerplate involved with this. Plus there a lot important things that need to happen with this function, such as using perfect forwardind, the `std::decay_t` type trait, and trailing decltype so it is SFINAE-friendly. So its not completely straightforward. So instead, the Fit library simplifies all of this, and you can just write this to provide the function to the user: BOOST_FIT_STATIC_FUNCTION(make_tuple) = by(decay, construct<tuple>()); Perhaps, you prefer to write the more complicated version, and that's fine, but I feel it does a disservice to the library, not showing how it can simplify a lot of code.

I don't mind if you like your programming style but I doubt I am the first or last programmer who prefers variables created and used when needed.

Have you ever used `fopen`? That is an object that gets defined at compile- time, not when you use it. That is when you write: void foo() { std::cout << "foo"; } You have just created a global function object, it occupies memory, you can take its address: auto* address_of_foo = &foo; Futhermore, this object wasn't initialized when you used it, it was initialized at compile-time. The `BOOST_FIT_STATIC_FUNCTION` works exactly the same way: struct foo_fn { void operator()() const { std::cout << "foo"; } }; BOOST_FIT_STATIC_FUNCTION(foo) = foo_fn(); So, `foo` gets initialized at compile-time, and you can take its address: auto* address_of_foo = &foo; So I dont see how you could say that `void foo()` is acceptable and `BOOST_FIT_STATIC_FUNCTION(foo)` since they work in the same way.

In fact I have a very strong suspicion that "my" programming style is that of the majority of C++ programmers and that only a minority of programmers like global variables even if they can be global within a namespace scope.

Perhaps, I should use different terminolgy to describe `BOOST_FIT_STATIC_FUNCTION`. Maybe it might be better to say that it creates a function(which it does) rather than say it create a global variable, since some developers may associate global variable with other things that are not related to `BOOST_FIT_STATIC_FUNCTION`.

So what I am essentially saying is that while I understand you like your style and your macros for creating "global" callables, inferring that this is central to using your library, when it is not AFAICS, is going to lose you end-users as well as misrepresent your library. I see much more plusses to your library if the result of using your adapters, decorators, functions, and utilities can be instantiated anywhere and used in ways that any variable in C++ can be used. if however there are real technical limitations, as opposed to "I like this style because I it is easier for me to use with Fit", of using Fit functionality to create "local" objects I think you need to spell out what those limitations are.

Its not a technical limitation, but I want to make sure to show the full depth of the capability of what the library can do. If you do not want to use the full capability, then thats ok, but there is no there is nothing problematic or dangerous defining functions using `BOOST_FIT_STATIC_FUNCTION`.

...

So you use global function objects when you want to make the function available for consumption by others. And the advantage of defining `make_tuple` this way over writing it as it is written currently in the standard library, is that we can avoid all the template boilerplate involved with this.

...

...

Alternatively, you could write a factory function for this instead:

constexpr auto make_tuple() { return by(decay, constructstd::tuple()); }

Of course this requires the user to write `make_tuple()(xs...)`, which

...

don't think any user would expect this, especially since `std::make_tuple` doesn't work like that. So instead, it can be wrapped in a template function:

template constexpr auto make_tuple(Ts&&... xs) -> declype(by(decay, constructstd::tuple())(std::forward<Ts>(xs)...)) { return by(decay, constructstd::tuple())(std::forward<Ts>(xs)...); }

Which is a lot of boilerplate to compose some simple functions.

...

If there really is some other reason it is completely lost to me. If you would like to point to me in your doc where you explain the use of "global variables" as being a key feature of your library or as being necessary to use the functionality of your library, I would be glad to read about it and ask further about it here.

It is not a necessary feature, but it is an important feature nonetheless, and I need to spend more time explaining its usefulness.

...

If, OTOH, it is just your preference to use global objects as opposed to the various forms of local instantiation, I really wish you would just say that rather

...

...

acting like your library does not work correctly somehow other than with global variables.

The composability of the adaptors and functions applies to both global and local functions. The adaptors provide a simple way to create functions to be passed locally, such as creating a comparator for std::sort(ie `std::sort(first, last, by(&employee::name, _<_))`). However, they can also be used to define global/free functions in a much simpler way as well(such as writing `std::make_tuple`), which I think is equally important.

Furthermore, in a quick start guide or introduction, I want to be able to demonstrate the capabilities of the library in a way to show its usefulness, and why someone would choose to use this library. Perhaps, you are ok with awkward syntax or template boilerplate in writing your functions, and would prefer to only use the adaptors for local functions. However, there are others who will find writing global function objects with the adaptors very useful, so I want to be able to show those capabilities in the introduction of

I than the

...

library. I do, however, need to discuss some of the misperceived issues with using global function objects in the documentation. As the issues raised in the review were:

1) Namespacing - This is not a problem because all global function objects can be placed in a namespace.

2) Global state - This is not a problem either, because `BOOST_FIT_STATIC_FUNCTION` declares the function object const with constexpr. So there can be no mutable or changing state internally in the object, and it must be constructed without side effects.

I think addressing these concerns in the documentation will help put users at ease when using global function objects, so they have no problem taking advantage of the adaptors to build function objects globally.

...

Maybe I have missed something but I have the intuition that many others have missed it likewise, from the responses of others about this issue.

I agree with your intuition, and I believe the documentation should spend more time discussing these issues. I hope my email has made it somewhat clearer to you.

_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost

On 3/11/2016 4:52 AM, Peter Dimov wrote:

Edward Diener wrote:

...

I don't understand this. I have a 'make_tuple' callable above and if I want to use it somewhere else I pass it to whatever functionality needs it. You, instead want things to be global so that they can be used by other functionality without passing anything around. Isn't this just an argument about programming style ?

No.

Your assignment is to define a make_tuple function in a header, say "make_tuple.hpp", so that other people can use it. This is supposedly a part from a library that allows the user to manipulate tuples. I don't understand your confusion and your reference to programming styles.

My confusion comes from the fact that I did not realize that Fit was actually defining a function. I thought instead that Fit was taking a Callable and ghenerating another Callable when using Fit's functionality, in much the same way that boost::bind does. I know that the macro in Fit is called BOOST_FIT_STATIC_FUNCTION ( or BOOST_FIT_STATIC_LAMBDA ) but the term 'function' in the Fit documentation is so often used when Callable is the actual correction designation that I thought that this was one more instance of this usage. Thanks for straightening me out with your explanation.

Le 11/03/2016 01:50, Paul Fultz II a écrit :

On Thursday, March 10, 2016 at 5:26:38 PM UTC-6, Edward Diener wrote:

...

On 3/10/2016 1:46 PM, Paul Fultz II wrote:

...

On Thursday, March 10, 2016 at 12:34:26 AM UTC-6, Vicente J. Botet

...

wrote:

...

I will put all of them in the advanced section. I would no mention them in the introduction as the user can create the HOF locally or use a factory.

Being able to declare the functions as global variables is a key feature of the library. The example in the quick start guide shows how to implement many things without the need for a lot of template boilerplate. For those users that prefer not to use global function objects(and I have yet seen a compelling reason not to use them) can look at the 'Advanced' section.

I realize now, that I need to spend more discussing the advantages of using global function objects(more composability), and address some of the misperceived issues with these global function objects(they can be in a namespace to avoid name conflicts and are not mutable so it won't be a problem in mutilthreaded environments). You do need to explain why "global variables" is a key feature of the

Escriba library. I cannot understand from your docs why there is any disadvantage instantiating Callables locally as opposed to globally other than the usual fact that a local object can go out of scope.

We can define functions by simply composing other functions together and we don't need to write awkward syntax nor template boilerplate for it. For example, we can simply write `make_tuple` like this:

BOOST_FIT_STATIC_FUNCTION(make_tuple) = by(decay, constructstd::tuple());

Alternatively, you could write a factory function for this instead:

constexpr auto make_tuple() { return by(decay, constructstd::tuple()); }

This function should be names maker_tuple at it builds an adaptor :)

Of course this requires the user to write `make_tuple()(xs...)`, which I don't think any user would expect this, especially since `std::make_tuple` doesn't work like that. So instead, it can be wrapped in a template function:

template constexpr auto make_tuple(Ts&&... xs) -> declype(by(decay, constructstd::tuple())(std::forward<Ts>(xs)...)) { return by(decay, constructstd::tuple())(std::forward<Ts>(xs)...); }

Which is a lot of boilerplate to compose some simple functions.

These are the kind of things that can be described in an advanced section?

...

If there really is some other reason it is completely lost to me. If you would like to point to me in your doc where you explain the use of "global variables" as being a key feature of your library or as being necessary to use the functionality of your library, I would be glad to read about it and ask further about it here.

It is not a necessary feature, but it is an important feature nonetheless, and I need to spend more time explaining its usefulness.

I believe all agree that the feature is useful for those that want to declare global HOFs. Some don't like global variables representing functions other don't like macros. The library is useful for those people too.

...

If, OTOH, it is just your preference to use global objects as opposed to the various forms of local instantiation, I really wish you would just say that rather than acting like your library does not work correctly somehow other than with global variables.

This is not the case. You can declare local variables for these HOF or define functions that return these HOF. What is not simple with the current language is to declare them as global variables, and this is the design rationale for those macros.

...

The composability of the adaptors and functions applies to both global and local functions. The adaptors provide a simple way to create functions to be passed locally, such as creating a comparator for std::sort(ie `std::sort(first, last, by(&employee::name, _<_))`). However, they can also be used to define global/free functions in a much simpler way as well(such as writing `std::make_tuple`), which I think is equally important.

Furthermore, in a quick start guide or introduction, I want to be able to demonstrate the capabilities of the library in a way to show its usefulness, and why someone would choose to use this library. Perhaps, you are ok with awkward syntax or template boilerplate in writing your functions, and would prefer to only use the adaptors for local functions. However, there are others who will find writing global function objects with the adaptors very useful, so I want to be able to show those capabilities in the introduction of the library. I do, however, need to discuss some of the misperceived issues with using global function objects in the documentation. As the issues raised in the review were:

1) Namespacing - This is not a problem because all global function objects can be placed in a namespace. This is obvious. There is no issue here.

2) Global state - This is not a problem either, because `BOOST_FIT_STATIC_FUNCTION` declares the function object const with constexpr. So there can be no mutable or changing state internally in the object, and it must be constructed without side effects.

I think addressing these concerns in the documentation will help put users at ease when using global function objects, so they have no problem taking advantage of the adaptors to build function objects globally.

...

Maybe I have missed something but I have the intuition that many others have missed it likewise, from the responses of others about this issue.

I agree with your intuition, and I believe the documentation should spend more time discussing these issues. I hope my email has made it somewhat clearer to you.

Best, Vicente

On Friday, March 11, 2016 12:30 PM, Bjorn Reese wrote:

...

On 03/11/2016 01:50 AM, Paul Fultz II wrote:

...

library. I do, however, need to discuss some of the misperceived issues with using global function objects in the documentation. As the issues raised in the review were:

The destruction order of global objects in different translation units is undefined. This means that I cannot use a global fit function in the destructor of one of my own global objects because the former may have

been destroyed before the latter.

This is not a problem, because it is initialized at compile-time, that means it is a literal type: http://en.cppreference.com/w/cpp/concept/LiteralType So besides the constructor not having side-effects, the destructor must be trivial, as well. For example, this cannot compile: struct foo_fn { template auto operator()(Ts&&...) const {} ~foo_fn() { std::cout << "Hello"; } }; BOOST_FIT_STATIC_FUNCTION(foo) = foo_fn(); int main() { foo(); } So with trivial destructors, the order of destruction doesn't matter. Furthermore, on most compilers(that is everything but MSVC), the function is the same across all translation units. So there is only one function in the executable, not one per translation unit. Taking the address of the function will yield the same address across all translation units. This is the same way functions work.

Le 13/03/2016 03:35, Lee Clagett a écrit :

On Fri, 11 Mar 2016 19:41:22 +0000 (UTC) paul Fultz wrote:

...

...

On Friday, March 11, 2016 12:30 PM, Bjorn Reese wrote:

...

On 03/11/2016 01:50 AM, Paul Fultz II wrote:

...

library. I do, however, need to discuss some of the misperceived issues with using global function objects in the documentation. As the issues raised in the review were: The destruction order of global objects in different translation units is undefined. This means that I cannot use a global fit function in the destructor of one of my own global objects because the former may have been destroyed before the latter. This is not a problem, because it is initialized at compile-time, that means it is a literal type:

http://en.cppreference.com/w/cpp/concept/LiteralType

So besides the constructor not having side-effects, the destructor must be trivial, as well. For example, this cannot compile:

struct foo_fn { template auto operator()(Ts&&...) const {}

~foo_fn() { std::cout << "Hello"; } };

BOOST_FIT_STATIC_FUNCTION(foo) = foo_fn();

int main() { foo(); }

So with trivial destructors, the order of destruction doesn't matter. Furthermore, on most compilers(that is everything but MSVC), the function is the same across all translation units. So there is only one function in the executable, not one per translation unit. Taking the address of the function will yield the same address across all translation units. This is the same way functions work.

The following program throws an exception and aborts in Clang 3.4 and Gcc 4.8, both on Ubuntu 14.04

#include <iostream> #include <stdexcept> #include #include

class log_ { public: log_() : active_(true) { std::cout << "constructing log" << std::endl; // or some equally terrible global state changing // code such as a file open }

~log_() { std::cout << "destructing log" << std::endl; active_ = false; }

void operator()(const char* const msg) const { if (active_) { std::cout << msg << std::endl; } else { throw std::runtime_error{"not active"}; } } private: bool active_; };

FIT_STATIC_FUNCTION(log) = fit::static_{};

struct one_ { ~one_() { log("destructing one"); } };

struct two_ { two_() { log("constructing two"); } };

const one_ one{}; const two_ two{};

int main() { return 0; }

Both Gcc and Clang print:

constructing log constructing two destructing log terminate called after throwing an instance of 'std::runtime_error' what(): not active

The generated program conforms to the C++ standard because `log_` was created as a function local static, and the call to `operator()` after destruction can do _anything_ (undefined behavior). Also, `log_` can be constructed before `one_` and `two_` (and therefore destructed after both) as well, and still conform to the C++ specifications. Or at least that is my understanding of static construction/destruction, which is complex in C++.

Lee

Tanks Lee for this example. I believe that static_ needs to require that the the parameter is is_trivially_destructible [1]. In [2] we can read " The |static_| adaptor is a static function adaptor that allows any default-constructible function object to be static-initialized.". It seems the implementation doesn't do the check. Please, could you add an issue on Github? Vicente [1] http://www.cplusplus.com/reference/type_traits/is_trivially_destructible/ [2] http://pfultz2.github.io/Fit/doc/html/static/index.html