[Issue 5117] [CTFE] Member function call with rather complex this: side effects ignored

[d-bugmail at puremagic.com]

http://d.puremagic.com/issues/show_bug.cgi?id=5117


Shin Fujishiro <rsinfu at gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[CTFE] Member function call |[CTFE] Member function call
                   |with chained dots: side     |with rather complex this:
                   |effects ignored             |side effects ignored


--- Comment #1 from Shin Fujishiro <rsinfu at gmail.com> 2010-10-25 14:14:06 PDT ---
The problem lies in FuncDeclralation::interpret(), around line 223:
--------------------
    // Don't restore the value of 'this' upon function return
    if (needThis() && thisarg->op == TOKvar && istate)
    {
        VarDeclaration *thisvar = ((VarExp
*)(thisarg))->var->isVarDeclaration();
        for (size_t i = 0; i < istate->vars.dim; i++)
        {   VarDeclaration *v = (VarDeclaration *)istate->vars.data[i];
            if (v == thisvar)
            {   istate->vars.data[i] = NULL;
                break;
            }
        }
    }
--------------------

In the repro code in comment #1, thisarg is 'r.s' (TOKdotvar) and the local
variable 'r' is not removed from the "restore list" istate->vars.  Then, the
interpretor wrongly restores 'r' to init.

Just dealing with TOKdotvar fixes the specific reported problem, but it's not a
general fix.  Still the interpretor should deal with references.  For example:
--------------------
enum dummy = test();

int test()
{
    S s;
    getRef(s).change();
    assert(s.value == 1);     // fails, value == 0
    return 0;
}
ref S getRef(ref S s) { return s; }

struct S
{
    int value;
    void change() { value = 1; }
}
--------------------

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