Currently D refuses code like this:


void foo(const int[] a) {
    int bar() pure {
        return a[0];
    }
}
void main() {}


With two repeated error messages:

test1.d(3): Error: pure nested function 'bar' cannot access mutable data 'a'
test1.d(3): Error: pure nested function 'bar' cannot access mutable data 'a'


But is it possible for D to see that bar function as pure?

Bye,
bearophile

On Tuesday, 14 January 2014 at 19:50:10 UTC, bearophile wrote:
> But is it possible for D to see that bar function as pure?

In the general case, no:

---
auto foo(const int[] a) {
    int bar() {
        return a[0];
    }
    return &bar;
}

void main() {
    int[3] a;
    auto dg = foo(a[]);
    assert(dg() == 0);
    a[0] = 1;
    assert(dg() == 1);
}
---

David

On Tuesday, 14 January 2014 at 20:21:25 UTC, David Nadlinger wrote:
> On Tuesday, 14 January 2014 at 19:50:10 UTC, bearophile wrote:
>> But is it possible for D to see that bar function as pure?
>
> In the general case, no:
>
> ---
> auto foo(const int[] a) {
>     int bar() {
>         return a[0];
>     }
>     return &bar;
> }
>
> void main() {
>     int[3] a;
>     auto dg = foo(a[]);
>     assert(dg() == 0);
>     a[0] = 1;
>     assert(dg() == 1);
> }
> ---
>
> David

Isn't this okay in the context of weak purity?

On Tuesday, 14 January 2014 at 20:21:25 UTC, David Nadlinger wrote:
> On Tuesday, 14 January 2014 at 19:50:10 UTC, bearophile wrote:
>> But is it possible for D to see that bar function as pure?
>
> In the general case, no:
>
> ---
> auto foo(const int[] a) {
>     int bar() {
>         return a[0];
>     }
>     return &bar;
> }
>
> void main() {
>     int[3] a;
>     auto dg = foo(a[]);
>     assert(dg() == 0);
>     a[0] = 1;
>     assert(dg() == 1);
> }
> ---
>
> David

It's pure in the sense that it only modifies data passed to it?

On 01/14/2014 08:50 PM, bearophile wrote:
> Currently D refuses code like this:
>
>
> void foo(const int[] a) {
>      int bar() pure {
>          return a[0];
>      }
> }
> void main() {}
>
>
> With two repeated error messages:
>
> test1.d(3): Error: pure nested function 'bar' cannot access mutable data
> 'a'
> test1.d(3): Error: pure nested function 'bar' cannot access mutable data
> 'a'
>
>
> But is it possible for D to see that bar function as pure?
>
> Bye,
> bearophile

Yes, it should.

https://d.puremagic.com/issues/show_bug.cgi?id=9148

On 01/14/2014 09:21 PM, David Nadlinger wrote:
> On Tuesday, 14 January 2014 at 19:50:10 UTC, bearophile wrote:
>> But is it possible for D to see that bar function as pure?
>
> In the general case, no:
>
> ---
> auto foo(const int[] a) {
>      int bar() {
>          return a[0];
>      }
>      return &bar;
> }
>
> void main() {
>      int[3] a;
>      auto dg = foo(a[]);
>      assert(dg() == 0);
>      a[0] = 1;
>      assert(dg() == 1);
> }
> ---
>
> David

int delegate()pure foo(const int[] a)pure{
    struct S{
        const int[] a;
        int bar()pure{
            return a[0];
        }
    }
    auto s=S(a);
    return &s.bar;
}

void main() {
    int[3] a;
    auto dg = foo(a[]);
    assert(dg() == 0);
    a[0] = 1;
    assert(dg() == 1);
}

On Tuesday, 14 January 2014 at 20:36:43 UTC, Timon Gehr wrote:
> Yes, it should.
>
> https://d.puremagic.com/issues/show_bug.cgi?id=9148

And, of course, you are right. ;) I missed the analogy to member functions w.r.t. the implicit context parameter. Shame on me.

David

On 01/14/2014 09:41 PM, David Nadlinger wrote:
> On Tuesday, 14 January 2014 at 20:36:43 UTC, Timon Gehr wrote:
>> Yes, it should.
>>
>> https://d.puremagic.com/issues/show_bug.cgi?id=9148
>
> And, of course, you are right. ;) I missed the analogy to member
> functions w.r.t. the implicit context parameter. Shame on me.
>
> David

Well, if it wasn't easy to fall into this trap (and this is mostly due to terminology), DMD would actually behave correctly. :o)

Timon Gehr:

> Yes, it should.
>
> https://d.puremagic.com/issues/show_bug.cgi?id=9148

Thank you. I have added my (probably redundant) test case to that issue.

Bye,
bearophile