Re: [C++14/lambda] Experimental polymorphic lambda patches

[Jason Merrill <jason at redhat dot com>]

On 07/01/2013 04:26 PM, Adam Butcher wrote:
> I've finally reworked the generic lambda patches I made back in 2009
> (http://gcc.1065356.n5.nabble.com/lambda-Latest-experimental-polymorphic-lambda-patches-tt713016.html)
> to apply against GCC's mainline.

Great!

> Missing features:
>
>   - Conversion to function pointer of stateless generic lambdas is
>     currently unsupported.  Currently a warning is generated for all
>     such lambdas at their definition.
>
> Other issues: (see XXX, FIXME, TODO)
>
>    a)  Location of implementation.

Let's move all the lambda code into a new lambda.c, semantics.c has 
gotten huge.

>    b)  What to do about name mangling?

I think what you have already should do the right thing; for normal 
function template instantiations we mangle the uninstantiated type of 
the template along with the template arguments, so I think for a generic 
lambda just mangling the uninstantiated parameter types is the way to go.

> Any comments appreciated.  Guidance on implementing the conversion
> operator for stateless generic lambdas would be useful.  I've made a
> few attempts which were lacking!

It seemed to me that N3649's description of the semantics should 
translate pretty easily into an implementation.  What problems are you 
running into?

> +              Couldn't figure out a clean way to test for lambda inside
> +              check_member_template.  */

LAMBDA_TYPE_P (current_class_type) doesn't work?

>    fn = lambda_function (closure);
> +  if (TREE_CODE (fn) == TEMPLATE_DECL)
> +    fn = DECL_TEMPLATE_RESULT (fn);

I think you could use STRIP_TEMPLATE before returning from 
lambda_function so that we don't need to adjust all the callers to deal 
with stripping the template, so instead of

> -    lambda = BASELINK_FUNCTIONS (lambda);
> +    lambda = OVL_CURRENT (BASELINK_FUNCTIONS (lambda));

You could have

  lambda = STRIP_TEMPLATE (get_first_fn (lambda));

> +  if (TREE_CODE (d) == TEMPLATE_DECL)
> +    return d;

Instead of this, let's avoid calling instantiate_decl from 
instantiate_class_template_1.

>   e)  Should collect 'auto' parameters during parse and realloc
>       template parameter list once at the end of the parameter list
>       rather than incrementally.

I agree, I don't think you need to do anything special while parsing the 
parameter list.

Jason