Re: [patch 1/3] Template Lookup

[sami wagiaalla <swagiaal at redhat dot com>]

> Sami>  Same patch but using prepare_for_testing
>
> Thanks.

Thanks for the review.

> I suspect this doesn't interact properly with psymtabs, because from
> what I can tell the template-argument-less symbol is only entered for
> full symbols, not partial symbols.  This means that a reference to some
> function will not work properly until the psymtab is expanded for some
> other reason.

Hmm.. can you give me an example of this ? I tried to construct an 
example with an extern function but I learned that cplus does not allow 
extern template functions.

> Sami>  +gdb_test_no_output "set language c++" ""
>
> There are a couple instances of this that have to be updated in light of
> Doug's recent change to how this is done.

What is the change that Doug made ?

> Sami>  @@ -2458,7 +2458,7 @@ find_overload_match (struct type **arg_types, int nargs,
> Sami>         if (fsym)
> Sami>           {
> Sami>  -          qualified_name = SYMBOL_NATURAL_NAME (fsym);
> Sami>  +          qualified_name = SYMBOL_SEARCH_NAME (fsym);
>
> It took me a while to understand why this could be work.
>
> It seems to me that it is not completely right though.  Consider this case:
>
>      template<int X>  int f (int x) { return x + X; }
>      int compute (int x) { return f<5>  (x) + f<7>  (x); }
>
> In this situation you can't just look at overloads of "f" -- it is
> ambiguous.  You have to do more analysis.

This can be blamed on find_overload_match's inability to deal with 
ambiguous resolution. Removing the template parameter actually makes it 
easier for find_overload_match to handle this situation if/when it 
supports the ambiguous case. It enables find_overload_match to print a 
message about ambiguous overload resolution instead of just "symbol not 
found". Of course p f<5>() and p f<7>() still works correctly.