C++ PATCH: PR6438

[Jakub Jelinek]

On Wed, Apr 24, 2002 at 03:43:43PM -0700, Mark Mitchell wrote:
> 
> The evil statement-expression extension strikes again.
> 
> Normally, we convert expression-statements to "void".  In other
> words, we turn:
> 
>   a = c + 3;
> 
> into:
> 
>   (void)(a = c + 3);
> 
> in the C++ front end.  This is somewhat pointless, but we do it to
> get certain warnings and such at that point.
> 
> However, if we're in a statement-expression it's bad to do this; the
> value might be needed.  Fortunately, we have a flag that indicates
> that the conversion is being done implicitly, in which case we don't
> really do the conversion; we just pretend.
> 
> Except that when the expression was a ?:-expression, we did do the
> conversion even when we weren't supposed to do so.
> 
> Fixed thusly, tested on i686-pc-linux-gnu, applied on the branch
> and the mainline.

This looks very much like PR c++/5373, with an (uncommited) fix
http://gcc.gnu.org/ml/gcc-patches/2002-01/msg01249.html
which is slightly different from what you've done.

> 2002-04-24  Mark Mitchell  <mark@codesourcery.com>
> 
> 	* cvt.c (convert_to_void): Don't unconditionally make COND_EXPRs
> 	void.

	Jakub