[erlang-questions] Trouble with init/1 spec

[Jay Nelson]

On Mar 24, 2010, at 9:00 AM, Kostis Sagonas wrote:

>  The type
>
> 	[string() | non_neg_integer()]
>
> constrains the input argument better than list()
>
>>  Declaring it as list(any()) means the fields on the record have  
>> to all be declared any().  As long as my port field is declared  
>> non_neg_integer() in the record definition, I can't get past the  
>> dialyzer warning.
>
> Huh?  Are you sure?  The record fields can be declared whatever  
> type you want.  I am not sure to what dialyzer warning you are  
> referring to, but perhaps we can take this off list.

I didn't state that too clearly.  The list(any()) caused dialyzer to  
deduce that the record assignment was creating an any() value in a  
field, therefore my record declaration each field needs to be any()  
rather than string() or non_neg_integer().

If I make the arglist [string() | non_neg_integer()], then I have to  
declare all my string() fields in the record to be of the same type  
because of the record constructor assignments which is really  
misleading to label a server name or user name as a non_neg_integer()  
and propagate that to any functions that access the field.

Right now the only way to get around it easily is to make my port  
field a string() and do the list_to_integer/1 conversion every time I  
access it.  That way all the fields in the init/1 list just happen to  
become string().

This seems to be a general problem for the init/1 behaviour callbacks  
which are designed to be flexible, but don't map easily to a record  
assignment that dialyzer can reason about.

jay