We need to kill C syntax for declaring function types

[Don]

bearophile wrote:
> Don:
> 
> Thank you for bringing up this topic, Don :-)
> 
> In D, beside the C syntax for function types, there also is the C syntax for function pointers. I presume it gives less problems than the first one.

Yes, I included that in the patch as well. It makes them deprecated, so 
they compile if and only if the -d command line switch is used.

> See also:
> http://d.puremagic.com/issues/show_bug.cgi?id=4530
> 
> 
>> In the past, Walter has mentioned a weak argument for retaining C-style
>> array declaration syntax, although I personally find it very
>> unconvincing. But C's hideous function pointer syntax is on a whole
>> other level. It's really hurting us. I believe it should be deprecated
>> immediately. But the 'function type' syntax shouldn't be allowed even as
>> a deprecated syntax. It's horrible.
> 
> One of the faults of C++, that increases its useless complexity a lot, is to have in many cases two ways (the C way and the C++ way) to do something. Generally the idea of duplicated syntax is bad.
> 
> But I have translated lot of C code to D1 and I've seen that the possibility that D1 gives me to leave matrix definitions as they are helps me reduce the amount of changes to the C code needed to perform the translation (and I think this may help avoid some translation bugs, but I have no proof of this).
> 
> So I leave declarations like:
> int mat[5][16];
> And only later, when the D1 programs works correctly, I replace them by code like:
> int[16][5] mat;
> 
> To merge such two needs, I have suggested a compilation switch, it may be named "-cstyle". It's designed to help translation of C code to D. If a module is compiled with -cstyle then DMD accepts C-style array declarations and C-style functions pointers, and it also gives warnings against passing by value large fixed-sized arrays and against the usage of global floating floating values before their initialization (those are two things that may trip a person that translates C code to D, because D passes fixed-sized arrays by value and doesn't initialize global floats to 0 but to NaN).
> 
> The idea is, when you need to translate C code to D you use the -cstyle switch, that allows some C syntax and gives those two extra warnings. When your code works, you may finish the translation to D and stop using the -cstyle switch.
> 
> See:
> http://d.puremagic.com/issues/show_bug.cgi?id=4580
> 
> What kind of error message does your patch gives for 4962?

bug.d(6): function declaration without return type. (Note that 
constructors are always named 'this')

That is, it doesn't work out if it was intended to be a constructor, it 
gives the same error message for any function with no return type.
Making a dedicated error message for constructors would require 
longer-range changes to the parser. But I believe that that the 
C++/Java/C# constructor syntax is by far the most common syntax where it 
will happen.