[Issue 10966] Leaked destruction in static array postblit

[d-bugmail at puremagic.com]

http://d.puremagic.com/issues/show_bug.cgi?id=10966


Kenji Hara <k.hara.pg at gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
           Severity|normal                      |critical


--- Comment #1 from Kenji Hara <k.hara.pg at gmail.com> 2013-09-05 02:07:08 PDT ---
This is -O switch issue with dmd.
Test following reduced case.

--------
extern(C) int printf(const char*, ...);

int k;

struct S
{
    int a;
    this(this)
    {
        printf("postblitting %d... ", a);
        if (k++ == 2)
        {
            printf("FAILED!\n");
            throw new Exception("BOOM!");
        }
        a += 10;
        printf("OK => %d\n", a);
    }
    ~this()
    {
        printf("destroying %d\n", a);
    }
}

void main()
{
    S s0     = S(0);
    S[4] ss1 = [S(1), S(2), S(3), S(4)];
    S[4] ss2 = s0;  //Case 1 to N
    // call _d_arraysetctor defined in this module
    // instead of the one stored in druntime
}

// copy from druntime/src/rt/arrayassign.d
extern (C) void* _d_arraysetctor(void* p, void* value, int count, TypeInfo ti)
{
    import core.stdc.string : memcpy;

    void* pstart = p;   // with -O, variable pstart is wrongly made an alias of
p
    auto element_size = ti.tsize;

    try
    {
        foreach (i; 0 .. count)
        {
            // Copy construction is defined as bit copy followed by postblit.
            memcpy(p, value, element_size);
            ti.postblit(p);
            p += element_size;
        }
    }
    catch (Throwable o)
    {
        // Destroy, in reverse order, what we've constructed so far
        while (p > pstart)
        {
            p -= element_size;
            ti.destroy(p);
        }
        printf("--end\n");

        throw o;
    }
    return pstart;
}
--------

Without -O switch, _d_arraysetctor works as expected.
With -O switch, it doesn't work correctly.

Today druntime is normally compiled with -O switch, therefore this problem
occurs always.

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