Manipulating C++ Containers Functionally

Manipulating C++ Containers Functionally(yapb-soc.blogspot.com)

35 points by splinterofchaos 13 years ago | 10 comments

Aardwolf 13 years ago |

I do agree that C++'s STL is sometimes too verbose.

It's important to be generic, but it'd have been great if the most common conveniences were added as well to make it pleasant to use. E.g.: erase remove idiom, count but no contains, specifying begin() and end() even when you mean the whole container, ... Anything where you need to type the variable name of your container more than once to do one operation on it.

Not just the containers and stl algorithms have that annoyance, but also the string stream operators to convert numbers to strings, this is way too verbose imho.

At least the new initializer lists make some things way more compact and convenient now :)

akaus 13 years ago | |

There is STO library: http://github.com/lvv/sto

> E.g.: erase remove idiom

remove sub-string: string("abcd") - "cd"

> specifying begin() and end()

unary '+', '-' overloaded to return to begin()/end()

> when you mean the whole container, ...

Euler example in just one expression (with boost lambda): range(1000) | (!(_1%3) || !(_1%3)) | add

zainny 13 years ago | |

Fortunately with C++11 we no longer have to do the stringstream dance thanks to built-in stoi/stol string member functions: http://en.cppreference.com/w/cpp/string/basic_string/stol

splinterofchaos 13 years ago | | |

Technically, those are free functions defined on strings, not members functions of the class string.

But also: std::to_string

http://en.cppreference.com/w/cpp/string/basic_string/to_stri...

bcoates 13 years ago |

I think going this way is going to turn out painful. The sort of immutable container logic that people associate with FP is against the nature of C++ and you'll be constantly bumping into other people's code, and parts of the library, that don't like it.

C++ style prefers

  for_each(xs.begin(), xs.end(), f);

  xs = std::transform(xs,f);

Because the first does not copy the container and its elements, which in C++ is a meaningful operation which may not be allowed for xs's elements, or may be remarkably expensive.

I thought the annoying verbosity when using the whole container with <algorithm> was fixed in C++11 like it was for for loops, but I can't find documentation to that effect. Boost's range algorithms library is still useful here:

http://www.boost.org/doc/libs/1_52_0/libs/range/doc/html/ind...

splinterofchaos 13 years ago | |

> The sort of immutable container logic that people associate with FP is against the nature of C++...

My argument is that C++, a multi-paradigm language, benefits from not being bound to purity and that being able to mix imperative and functional code is a good thing. We aren't stuck in pure-land so we can insert IO where it needs to be, not where it propagates, but neither do we need to be stuck in state-land. Functional programming is not against the "nature" of the language, it gives us the freedom to choose.

> Because the first does not copy the container and its elements...

This is a quality of implementation detail. It would have been detrimental to the point of the article to waste time showing optimizations such as "if F is a function X to X, use the in-place version". But I assure you, my optimization testing has shown there is little difference between the code generated by this and vanilla <algorithm>. In fact, it is at times better.

https://github.com/splinterofchaos/Pure/blob/master/samples/...

wonderzombie 13 years ago |

I was a little bit afraid to read this one. I've glanced at some of the functional patterns in the STL — function objects, ugh — and come away shaking my head.

But despite my problems with C++ and/or the STL, I really appreciated this piece. The author seems to know his stuff pretty well, as there are references to Haskell, monoids, laziness, as well as a brief discussion of performance.

It's probably too much to hope that this will win terribly many converts, but hey, what do I know?

ahelwer 13 years ago | |

What's wrong with function objects? They're essentially functions with partial application.

alexchamberlain 13 years ago |

I've recently discovered the delights of <algorithm>... now if I could just figure out how to bind member functions in my build environment (Solaris) I'd be a very happy man.

splinterofchaos 13 years ago | |

You want either mem_fn (C++11) or mem_fun.

http://en.cppreference.com/w/cpp/utility/functional