We need to kill C syntax for declaring function types
Don
nospam at nospam.com
Mon Oct 4 05:55:55 PDT 2010
Steven Schveighoffer wrote:
> On Mon, 04 Oct 2010 05:07:07 -0400, Don <nospam at nospam.com> wrote:
>
>> A great example of how C syntax is hurting us.
>> ---
>> I found this bit of code in std.container, inside BinaryHeap:
>>
>> size_t insert(ElementType!Store value)
>> {
>> static if (is(_store.insertBack(value)))
>> {
>> ...
>> }
>> else ...
>>
>> What does the static if do? It's *intended* to check if _store has a
>> member function insertBack(), which accepts type of a 'value'.
>>
>> But instead, it ALWAYS silently does the 'else' clause.
>> Unless _store.insertBack is a valid *type*, (eg, alias int insertBack;).
>> In which case it gives an error "declaration value is already defined".
>
> I'm really confused here. I could have sworn I've seen compile time
> checks like this everywhere. Is the "bug" that's unflagged by the
> compiler that you are missing a typeof(...)?
Yes.
>>
>> Why?
>>
>> This happens because
>> x(y); is valid C syntax for declaring a type 'y', such that &y is of
>> type 'x function()'.
>
> Wait, I thought when declaring a function pointer, you had to have the
> (*) in there? Plus I thought you had to have an extra set of
> parentheses? I've never seen this before.
>
> Trying...
>
> Oh, that's freaking awful. x(y); silently compiles into something
> useless, you need to declare it with a typedef in order to use it (and
> even then, it's horrible). It's like accepting the line
>
> int;
>
> Yes, 100% agree, get rid of this. Does that mean we need to get rid of
> C-style function pointer declarations? Because I recently saw a use for
> them (deciphering C-style function pointer syntax).
I agree, I use it for the same thing. In the patch, I made C-style
function pointers deprecated (so you can still compile it with the -d
switch) and C-style abominable x(y) types illegal.
I would expect that it always remains a deprecated syntax, to aid
conversion of C-style code. But in my experience, you always want to get
rid of them eventually.
More information about the Digitalmars-d
mailing list