November 04, 2012
I just thought that I should bring greater attention to

http://d.puremagic.com/issues/show_bug.cgi?id=8838

As it stands, I think that the slicing static arrays being considered @safe is a major hole in SafeD's safety, and I think that it's one that many of us aren't aware of. But there seems to be some resistance to outright making the slicing of static arrays @system within that bug report, and I recall similar resistance in the thread that prompted that bug report (unfortunately, I don't remember what thread it was - somewhere in D.Learn I think).

So, I was wondering what the general consensus on this was. Should slicing static arrays be considered @system? I honestly don't see how we could do otherwise without the compiler being way, way smarter at detecting escaping references than is ever going to happen.

Also, I really think that if we're agreed that this change needs to be made that it should then be made sooner rather than later, because it's going to break code which actually tries to use @safe.

- Jonathan M Davis
November 04, 2012
Jonathan M Davis:

> I honestly don't see how we could do otherwise
> without the compiler being way, way smarter at detecting escaping references than is ever going
> to happen.

One question is how much work does it take to precisely keep track of such memory zones inside the static type system of D? :-)

Bye,
bearophile
November 04, 2012
On Sunday, 4 November 2012 at 05:31:41 UTC, Jonathan M Davis wrote:
> I just thought that I should bring greater attention to
>
> http://d.puremagic.com/issues/show_bug.cgi?id=8838
>
> As it stands, I think that the slicing static arrays being considered @safe is
> a major hole in SafeD's safety, and I think that it's one that many of us
> aren't aware of. But there seems to be some resistance to outright making the
> slicing of static arrays @system within that bug report, and I recall similar
> resistance in the thread that prompted that bug report (unfortunately, I don't
> remember what thread it was - somewhere in D.Learn I think).
>
> So, I was wondering what the general consensus on this was. Should slicing
> static arrays be considered @system? I honestly don't see how we could do
> otherwise without the compiler being way, way smarter at detecting escaping
> references than is ever going to happen.
>
> Also, I really think that if we're agreed that this change needs to be made
> that it should then be made sooner rather than later, because it's going to
> break code which actually tries to use @safe.
>
> - Jonathan M Davis

Just as a note, if slicing static arrays cannot be @safe, then type-safe variadic functions can't be @safe either, as it essentially does the same thing, just implicitly:

    void foo(int[] a...)
    {
        // 'a' is a slice of stack memory when foo *isn't* called with an explicit array.
    }

The compiler does perform its simple escape analysis on such parameters, but as we know, it's not very smart.

I guess another solution to this particular case is to make it generate a dynamic array when foo is @safe, but I don't think anyone is excited about this option...

Then there's the scope parameter storage class. The compiler isn't very smart about scope parameters either.

So what do we do about all these related issues?

November 04, 2012
On Sunday, November 04, 2012 06:37:57 bearophile wrote:
> One question is how much work does it take to precisely keep track of such memory zones inside the static type system of D? :-)

I don't think that the type system has any concept of that whatsoever.

- Jonathan M Davis
November 04, 2012
On Sunday, November 04, 2012 06:48:15 Jakob Ovrum wrote:
> So what do we do about all these related issues?

I think that anything that the compiler can't absolutely gurantee is @safe must be @system. If that's annoying in some places, then that's life, because we can't compromise on SafeD just because a few things that we use a lot can't be @safe. Now, if we can make further improvements that make it so that the compiler _can_ determine that something is actually @safe when before it couldn't, then that's great. So, in at least some cases, compiler improvements could be used to reduce the annoyance factor, but I don't think that we can compromise on this.

- Jonathan M Davis
November 04, 2012
On Sunday, 4 November 2012 at 05:58:20 UTC, Jonathan M Davis wrote:
> I think that anything that the compiler can't absolutely gurantee is @safe
> must be @system. If that's annoying in some places, then that's life, because
> we can't compromise on SafeD just because a few things that we use a lot can't
> be @safe. Now, if we can make further improvements that make it so that the
> compiler _can_ determine that something is actually @safe when before it
> couldn't, then that's great. So, in at least some cases, compiler improvements
> could be used to reduce the annoyance factor, but I don't think that we can
> compromise on this.
>
> - Jonathan M Davis

I completely agree, but I don't think we should decide what should be guaranteed and what cannot by the current state of the compiler. We need a properly specified goal that programmers can rely on, and implementers work towards.

November 04, 2012
On 11/4/12 1:31 AM, Jonathan M Davis wrote:
> I just thought that I should bring greater attention to
>
> http://d.puremagic.com/issues/show_bug.cgi?id=8838
>
> As it stands, I think that the slicing static arrays being considered @safe is
> a major hole in SafeD's safety, and I think that it's one that many of us
> aren't aware of. But there seems to be some resistance to outright making the
> slicing of static arrays @system within that bug report, and I recall similar
> resistance in the thread that prompted that bug report (unfortunately, I don't
> remember what thread it was - somewhere in D.Learn I think).
>
> So, I was wondering what the general consensus on this was. Should slicing
> static arrays be considered @system? I honestly don't see how we could do
> otherwise without the compiler being way, way smarter at detecting escaping
> references than is ever going to happen.
>
> Also, I really think that if we're agreed that this change needs to be made
> that it should then be made sooner rather than later, because it's going to
> break code which actually tries to use @safe.

Slicing of static arrays is unsafe only if they're stack-allocated and the slice is subsequently escaped.

Andrei

November 04, 2012
On Sunday, November 04, 2012 08:59:10 Andrei Alexandrescu wrote:
> Slicing of static arrays is unsafe only if they're stack-allocated and the slice is subsequently escaped.

If we want to make it so that slicing a static array is @safe when the compiler can determine for sure that the slice isn't escaping, then fine, but in general, it can't do that. And taking the address of a local variable is already considered @system regardless of what you do with it. Taking the slice of a static array is really no different from taking the address of a local variable. It's just that it has a length along with the address. So, it would be completely in line with how taking the address of a local variable is treated to make taking the slice of a static array which is a local variable @system.

Regardless, the point is that there is a major hole at this point with how slicing static arrays is always considered @safe. Nothing which the compiler can't guarantee is @safe should be considered @safe. So, even if we treat slicing static arrays as @safe in some cases, they need to be cases where the compiler is able to prove that it's safe.

- Jonathan M Davis
November 07, 2012
"Jonathan M Davis" <jmdavisProg@gmx.com> wrote in message news:mailman.1518.1352007101.5162.digitalmars-d@puremagic.com...
>I just thought that I should bring greater attention to
>
> http://d.puremagic.com/issues/show_bug.cgi?id=8838
>
> As it stands, I think that the slicing static arrays being considered
> @safe is
> a major hole in SafeD's safety, and I think that it's one that many of us
> aren't aware of. But there seems to be some resistance to outright making
> the
> slicing of static arrays @system within that bug report, and I recall
> similar
> resistance in the thread that prompted that bug report (unfortunately, I
> don't
> remember what thread it was - somewhere in D.Learn I think).
>

Slicing static arrays on the stack is equivalent to taking the address of a local variable, which is already illegal in SafeD.


November 07, 2012
On Wednesday, November 07, 2012 12:44:26 Daniel Murphy wrote:
> Slicing static arrays on the stack is equivalent to taking the address of a local variable, which is already illegal in SafeD.

Which is what I'm arguing, but a number of people seem to really not like the idea of making slicing static arrays on the stack @system even though it _is_ the same thing as taking the address of a local variable, which is @system.

- Jonathan M Davis
« First   ‹ Prev
1 2
Top | Discussion index | About this forum | D home