[erlang-questions] simplified attribute declaration

Kostis Sagonas kostis@REDACTED
Fri Mar 19 13:23:55 CET 2010


Kenneth Lundin wrote:
> Hi,
> 
> The possibility to specify attributes without parentheses is a mistake
> that we will corrected as soon as
> possible. It was introduced together with the -spec and -type notation
> which does not need parentheses.
> 
> I don't understand why it is cool.
> Don't even try to use this syntax in your code you will be punished
> with syntax error as soon as we fix this bug.
> 
> /Kenneth Erlang/OTP, Ericsson
> 
> 
> On Thu, Mar 18, 2010 at 10:14 AM, Vlad Dumitrescu <vladdu55@REDACTED> wrote:
>> Hi,
>>
>> I don't remember anyone mentioning this before, but I think it's cool
>> and should be promoted as the recommended syntax:
>>
>> All attributes can now (R13) be specified without parantheses!
>>
>> -module hello_world.
>> -export [init/1, stop/2].
>> -define DEBUG(X), io:format("debug: ~p~n", [X]).


Let me try to clarify the issue of having the possibility to write 
attributes without parentheses because I see a lot of confusion (perhaps 
naturally so, as this feature is not documented), first in Vlad's mail, 
then in the various replies that his mail has generated -- which has 
provoked Kenneth's "threatening" mail to the mailing list.

What is really happening is that attributes which take just one argument 
do not need to be enclosed in parentheses. On the other hand, attributes 
with more than one argument (read: whenever there is a ',' involved, as 
in the case of -define above) need to be enclosed in parentheses. This 
was introduced in order to allow types and specs to be written without 
parentheses but the mechanism is more general and applies to all 1-arity 
attributes. This is fine, I think, and does not need to be changed.

It is true that users who have spent years getting used to the "Erlang 
way of doing things" might find this possibility confusing at first. On 
the other hand, those that have some exposure to ML or Haskell will 
recognize something familiar to them: think of all attributes as taking 
only _one_ parameter and the parentheses are needed for grouping what's 
inside this parameter.  In other words, read the three attributes above 
as written:

-module hello_world .
-export [init/1, stop/2] .

-define (DEBUG(X), io:format("debug: ~p~n", [X])) .

So, in the last case, it's really the ',' that forces the parentheses.


Personally, I do not see any compelling reason why one should be 
REQUIRED to put parentheses around the single argument of unary 
attributes. Of course, note that even today this is entirely optional: 
one CAN put parentheses around this argument if one really wishes to, in 
the same sense that one can write 40 + 2, (40 + 2), (40 + (2)), etc.

Kostis


More information about the erlang-questions mailing list