[PATCH] Fix PR27529, a step towards fixing PR27039

[Mark Mitchell]

Andrew Haley wrote:
> Ian Lance Taylor writes:
> 
>  > Can you give more details about when Java needs to generate code
>  > for pointer conversions?
> 
> Whenever we can't prove by static analysis that the conversion is type
> safe.

To be clear, C++ generates code for pointer-conversions too, for 
derived-to-base casts involving multiple inheritance.  This is simpler 
than Java, which has to do something more likely dynamic_cast, which is 
explicitly a function call in C++.

I agree that we should not have language-dependent interpretation in the 
middle end, and that the right thing is for the Java front end to 
generate explicit code, whether it's the actual code to do the 
conversion or a new tree node that explicitly indicates the conversion, 
or whether it's just that we retreat from the position that CONVERT_EXPR 
is a synonym for NOP_EXPR.

However, I also think that it's OK to have a language-dependent hack in 
the middle-end in 4.2 to deal with this Java bug, so I'm comfortable 
with Ian's approval of the patch for that branch.

-- 
Mark Mitchell
CodeSourcery
mark@codesourcery.com
(650) 331-3385 x713