Unsafe pointer from string

[iskiselev]

Looks like JSIL doesn't support resolving pointer from string. Here is test case:

using System;

static class Program {
    public static void Main () {
        string str = "qww";
        unsafe
        {
            fixed (char* sPtr = str)
            {
                *sPtr = 'a';
            }
        }

        Console.WriteLine(str);
    }
}

Translation result:

    var sPtr = JSIL.UntranslatableInstruction("Implicit conversion to pointer: Untranslatable Expression Conversion of expression '@<var System.String str>' to pointer");
    sPtr.set(("a").charCodeAt(0));
    $T00().WriteLine("qww");

[iskiselev]

It would be very good, if at least read operations will work.

[iskiselev]

Looks like it will be pretty easy to implement for read operations. Will do it.
In any case, mutate string through a pointer is not best idea even in .Net.

[kg]

Yeah, PinAndGetPointer just needs to copy the string's chars into a temporary utf-16 buffer.

[iskiselev]

Will it be OK if we create Pointer with null as MemoryRange and source string as view (I've already done it)? Or there will be problems if we'd like to cast pointer afterward?

[kg]

No, that's no good.

[iskiselev]

What should we do on write through such pointer attempts? Right now I throw NotImplementedException. Or should we allow write to such temporary array (waithout writing it back to source string)?

[kg]

I would allow the write and not represent it in the source string. It's not possible to mutate a string instance in JS anyway.

While mutating a pinned string in C# might work it's a terrible idea so I'm okay with not implementing it. If it's needed, we could make it work by implementing the unpin portion of fixed blocks and using that to flush the change to the string somehow... but the string's object reference would change, so it's still not gonna work.

[iskiselev]

Additional notes: get pointer to char[] element doesn't work also.
One more thing to test - cast pointer to byte* or int* (as sizeof(char)=2).

using System;

static class Program {
    public unsafe static void Main(string[] args)
    {
        string str = "qww";
        unsafe
        {
            fixed (char* sPtr = str.ToCharArray())
            {
                RunMe(sPtr, str.Length);
            }
        }

        Console.WriteLine(str);
    }

    private unsafe static void RunMe(char* sPtr, int length)
    {
        char* endPtr = sPtr + length;
        byte* bPtr = (byte*)sPtr;
        while (sPtr < endPtr)
        {
            Console.WriteLine(*sPtr);
            Console.WriteLine(*bPtr);
            Console.WriteLine(*(bPtr + 1));

            sPtr++;
            bPtr += 2;
        }
    }
}

[kg]

char is tricky since it's a string instance right now. at some point we
should switch them over to ints but the lack of boxing prevents it
currently. same goes for bool.

On 5 October 2015 at 11:34, Igor Kiselev notifications@github.com wrote:

Additional notes: get pointer to char[] element doesn't work also.
One more thing to test - cast pointer to byte* or int* (as sizeof(char)=2).

using System;

static class Program {
public unsafe static void Main(string[] args)
{
string str = "qww";
unsafe
{
fixed (char* sPtr = str.ToCharArray())
{
RunMe(sPtr, str.Length);
}
}

    Console.WriteLine(str);
}

private unsafe static void RunMe(char* sPtr, int length)
{
    char* endPtr = sPtr + length;
    byte* bPtr = (byte*)sPtr;
    while (sPtr < endPtr)
    {
        Console.WriteLine(*sPtr);
        Console.WriteLine(*bPtr);
        Console.WriteLine(*(bPtr + 1));

        sPtr++;
        bPtr += 2;
    }
}

}

—
Reply to this email directly or view it on GitHub
#860 (comment).

[iskiselev]

Could you explain why does lack of boxing prevents us from moving chars to Numbers? Right now I don't see why it is so.

[kg]

char and boolean would lose all type information when cast if unboxed, which poses some difficulties. there's also some other stuff that needs to happen when indexing into strings, etc. i think once boxing/wrapping works the rest can be sorted out.

[markusjohnsson]

That is what I do* in cil.js and it works great. The type is implicitly known after the unboxing and always provided when boxing occurs.

*Edit: using Number for char and bool that is. I've also considered using Number(0) to represent null, as that is closer to MSIL, but not sure that is a good idea.