[erlang-questions] Creating records dynamically

Richard O'Keefe ok@REDACTED
Fri Jan 22 02:23:26 CET 2010


I'm trying to think of a good use for a function that takes a
record name as an argument and uses it for construction.
I have had good success imagining square circles, but not this.

Amongst other things,
  - "unwarranted chumminess" with the compiler's representation of  
records
  - how come there are lots of record types with the same field names?
  - what is supposed to happen if you pass an atom that is NOT the name
    of a record, or is the name of a record but not one visible in this
    module, or is the name of a record visible in this module that  
doesn't
    have all those fields
  - how would you write a type specification for the put_vals function?
  - why not pass a constructor *function* instead of a record name,
    so that you aren't crippled by a restriction to constructing  
records?

E.g.,

     put_vals(F) -> F(1, 2).

     ... put_vals(fun (X,Y) -> #sample{a = X, b = Y} end) ...

Of course, if you find yourself passing the same constructor function
more than once, you would do well to give it a name:

     mk_sample_a_b(A, B) -> #sample{a = A, b = B}.

     ... put_vals(fun mk_sample_a_b/2) ...


Convention:
	mk_<record name>{_<field name>}...

The horrible thing about Yau-Hsien Huang's sample code
-module(test).
-record(sample, {a, b}).

test(T) ->
    case put_vals(T) of
        #sample{a=X, b=Y} -> [sample, X, Y];
Any -> Any
    end.

put_vals(T) ->
    {T, 1, 2}.

is that it breaks if the -record declaration is changed,
say to	-record(sample, {b, a}).	% change order
     or  -record(sample, {a, b, c}).	% change size

BOA constructors have no place in Erlang.




More information about the erlang-questions mailing list