Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Plain View
Avro >> mail # user >> Using optional fields (nullable unions) with the C-API


+
Markus Resch 2011-11-21, 17:26
Copy link to this message
-
Re: Using optional fields (nullable unions) with the C-API
> I'm using a record like this
>
> {
>    "type" : "record",
>    "name" : "TestPacket",
>    "fields" : [
>        { "name" : "Id" , "type" : "int"},
>        { "name" : "Value" , "type" : ["string","null"]}
> ]
> }
>
> But if I just not write anything to the 'Value' filed,
> avro_writer_memory() refuses to validate it. How do you normally use the
> optional fields with the C-API?

Hi Markus

In Avro, an “optional” field is really just a common pattern where you've got a union, with a “null” for one branch, and some other type for the second branch.  And with all union types, you have to explicitly tell the Avro library which branch is active for a particular value.  (Think of the union as a container that must always contain exactly one value, but where the type of that value can change.)

So, if you're using the new value API, you'd use something like:

  avro_value_t  rec = /* given */;
  avro_value_t  field, branch;
  avro_value_get_by_name(rec, "Value", &field, NULL);
  avro_value_set_branch(&field, 1, &branch);
  avro_value_set_null(&branch);

and if you're using the old datum API:

  avro_datum_t  rec = /* given */;
  avro_datum_t  field, branch;
  avro_record_get(rec, "Value", &field);
  avro_union_set_discriminant(field, 1, &branch);
  /* no avro_null_set in the old API */

In set_branch/set_discriminant, the 1 refers to the index of the "null" type in the union.  I tend to use branch 0 for the null type in these kinds of unions, but that's just a matter of aesthetics.

cheers
–doug