Thread overview
[Issue 5125] New: Optional function purity/nothrowness
Mar 05, 2011
Jonathan M Davis
October 27, 2010
http://d.puremagic.com/issues/show_bug.cgi?id=5125

           Summary: Optional function purity/nothrowness
           Product: D
           Version: D2
          Platform: All
        OS/Version: All
            Status: NEW
          Severity: enhancement
          Priority: P2
         Component: DMD
        AssignedTo: nobody@puremagic.com
        ReportedBy: bearophile_hugs@eml.cc


--- Comment #0 from bearophile_hugs@eml.cc 2010-10-27 04:55:57 PDT ---
A higher order function like "map" (or "filter") is meant to be often a pure
function, this means that the mapping delegate is often pure (see also bug 3934
and bug 5006 ).

A possible solution: the "map" function may be pure if and only if the given delegate/function is pure (a similar problem is present regarding the "nothrow" attribute). But currently there is no good way to do this.

This code shows the situation, you need two map functions, and some code duplication:


pure int sqr(int x) { return x * x; }

pure int[] pureMymap(F)(F f, int[] data) {
    int[] res;
    foreach (x; data)
        res ~= f(x);
    return res;
}

int[] mymap(F)(F f, int[] data) {
    int[] res;
    foreach (x; data)
        res ~= f(x);
    return res;
}

void main() {
    int[] data = [1, 2, 3, 4];

    auto r1 = pureMymap(&sqr, data);
    assert(r1 == [1, 4, 9, 16]);

    int acc = 1;
    int impure(int x) { acc += x; return acc; }
    // Error: pure function 'pureMymap' cannot call impure delegate 'f'
    // auto r2 = pureMymap(&impure, data);
    auto r2 = mymap(&impure, data); // OK
    assert(r2 == [2, 4, 7, 11]);
}



With a string mixing that contains the whole code of map you may remove the code duplication, but it's an ugly solution, and in the end you have two functions with different names still:


import std.metastrings: Format;

enum string mapCode = q{
%s int[] %smap(F)(F f, int[] data) {
    int[] res;
    foreach (x; data)
        res ~= f(x);
    return res;
}
};

mixin(Format!(mapCode, "", "")); // defines map()
mixin(Format!(mapCode, "pure", "pure")); // defines pure puremap()

pure int sqr(int x) { return x * x; }

void main() {
    int[] data = [1, 2, 3, 4];

    auto r1 = puremap(&sqr, data);
    assert(r1 == [1, 4, 9, 16]);

    int acc = 1;
    int impure(int x) { acc += x; return acc; }
    // Error: pure function 'puremap' cannot call impure delegate 'f'
    //auto r2 = puremap(&impure, data);
    auto r2 = map(&impure, data); // OK
    assert(r2 == [2, 4, 7, 11]);
}


A better solution is an attribute that allows to use optional tags, like (you may find a better name for this attribute):


@optional_tag(isPure!F, pure) int[] map(F)(F f, int[] data) {
    int[] res;
    foreach (x; data)
        res ~= f(x);
    return res;
}


The first argument of @optional_tag() is a boolean known at compile-time.

Where isPure is just:

import std.traits: FunctionAttribute, functionAttributes;

template isPure(F) {
    enum bool isPure = functionAttributes!(F) & FunctionAttribute.PURE;
}


A more general alternative syntax:
__traits(optional, isPure!F, pure)

Other more general solutions are possible, this is just a first idea.

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
November 09, 2010
http://d.puremagic.com/issues/show_bug.cgi?id=5125



--- Comment #1 from bearophile_hugs@eml.cc 2010-11-09 04:15:27 PST ---
See bug 4974 for another usage

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
November 14, 2010
http://d.puremagic.com/issues/show_bug.cgi?id=5125



--- Comment #2 from bearophile_hugs@eml.cc 2010-11-14 11:27:06 PST ---
Another idea: http://www.digitalmars.com/webnews/newsgroups.php?art_group=digitalmars.D&article_id=122087

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
March 05, 2011
http://d.puremagic.com/issues/show_bug.cgi?id=5125


Jonathan M Davis <jmdavisProg@gmx.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jmdavisProg@gmx.com


--- Comment #3 from Jonathan M Davis <jmdavisProg@gmx.com> 2011-03-04 16:44:23 PST ---
I believe that Andrei's suggestion for pure was to add optional constraints to pure. So, you'd have pure(condition) instead of pure, and the function would be pure if the constraint was true. Presumably, we could do the same for nothrow as well. @safe, @trusted, and @system (or at least @safe and @trusted) are in the same boat. So, maybe it should just be generalized to all function attributes (which would then presumably include stuff like public and private, though that's pretty pointless other than avoiding special casing attributes might be nice).

Regardless, we're going to need a solution for this if we want attributes like pure to really work with templated functions.

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
June 26, 2011
http://d.puremagic.com/issues/show_bug.cgi?id=5125


bearophile_hugs@eml.cc changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED


--- Comment #4 from bearophile_hugs@eml.cc 2011-06-26 05:03:42 PDT ---
Conditional purity of DMD 2.054 makes this enhancement request unnecessary.

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------