Re: [PATCH] Latest version of the -Wunused-but-set-{variable,parameter} patches (C++ part)

[Jason Merrill <jason at redhat dot com>]

On 04/15/2010 07:53 AM, Dodji Seketeli wrote:
> On Fri, Jan 22, 2010 at 03:17:42PM -0500, Jason Merrill wrote:
> > It still bothers me that you need to touch so many places where we
> > actually are using the value of the expression: build_new,
> > build_delete, decay_conversion, perform_integral_promotions,
> > cp_build_unary_op [PREINCREMENT_EXPR et al], convert_like_real,
> > build_x_va_arg, build_over_call, build_new_method_call, rvalue.
> >
> > I suppose we could improve that situation by making the lvalue uses,
> > rvalue uses and type-only/pseudo-uses in this patch call different
> > entry points.
>
> So I could do away with touching build_new, build_delete,
> cp_build_unary_op, build_over_call, rvalue and build_new_method_call.
> This comes at the expense of adding *INCREMENT_EXPR support to
> mark_exp_read and adding calls to it in build_call_a and ocp_convert.
> I couldn't find how to do away with touching decay_conversion and
> perform_integral_promotions.
>
> Is this the kind of changes that you meant?

If you leave the call in rvalue, can you avoid the *INCREMENT_EXPR change?

My second paragraph above was requesting that uses as rvalue, lvalue and 
type call different functions rather than all calling mark_exp_read.

Jason