Re: [PATCH] PR c++/42260

[Jason Merrill <jason at redhat dot com>]

On 12/22/2009 07:44 AM, Dodji Seketeli wrote:
> On Mon, Dec 14, 2009 at 01:08:08PM -0500, Jason Merrill wrote:
> > On 12/14/2009 12:31 PM, Dodji Seketeli wrote:
> > the code that looks for a conversion from A to something* should
> > complain about choosing a template.
>
> I first tried to make lookup_conversions not return any template
> conversion operator when called from within add_builtin_candidates.
> That was breaking things because sometimes the template conversion
> op can be properly instantiated. In thoses cases we need to consider that
> that conversion op.

When does this happen?

> I then tried in the patch below to avoid considering the operator '*'
> as a built-in candidate when it's operand has a dependent type. Having a
> built-in operator operand with dependent type would otherwise mean
> that we picked up a template conversion operator that couldn't be
> properly instantiated and was thus converting - after being instantiated
> by add_template_candidate, called by build_user_type_conversion_1 - to a
> dependent type.

It's not clear to me why this would avoid breaking the same cases that 
would be broken by the above change.

I notice that the code in add_builtin_candidate for handling comparisons has

      if (TYPE_PTR_P (type1)
          && null_ptr_cst_p (args[1])
          && !uses_template_parms (type1))

I guess similarly checking !uses_template_parms in the INDIRECT_REF case 
would be the minimal fix.

Jason