Format result-set of join (split parent/child rows)

Question

Hi,
I have the need to do a simple join however instead of returning all the columns from both the joined tables in 1 row, we need to have them split into multiple rows (i.e. parent record info on 1 row and child record info on the next)
e.g. If we have the following data:
table_1
id    parent_id
101    null
102    null
103    null
104    101
105    101
106    102
I need the data as follows:
id    parent_id    record_type
101    null        'parent'
104    101        'child'
105    101        'child'
102    null        'parent'
106    102        'child'
(note: since 103 did not have any children, it was not returned (as expected & as desired) - also, dont worry about order of results for now)
This can be achieved this by using the following:
;with temp
as
select [id],parent_id,[header_1],'parent' as type
from table1
temp2 as
select [id],parent_id,[header_1],'child' as type
from table1
select distinct A.*
from temp A
join temp2 B on B.parent_id = A.id
union
select distinct B.*
from temp A
join temp2 B on B.parent_id = A.id
However, is there any better/more efficient way to get this?
Appreciate any help on this.

Answer 1

For future reference, I'll point out that you ask 2 very different questions.  When you ask about "better", you need to define how you measure that quality.  "Better", like "pretty", has no universal measure.  
Next, you ask about "more efficient".  Well, your example is very simplistic - perhaps to the point of being unreasonably so.  You measure efficiency by examining the query plan - and that requires an actual populated table, with indexes,
constraints, and data that accurately reflects your data  in both size and distribution of values. If you simplify things too much, you may find that your attempts to improve things lead you down the wrong path.  
Also, be careful what you assume about your data.  You have conveniently created a set of rows that have a rather natural order.  If you care about the order of your resultset - and generally someone somewhere does - then you need to develop a
solution that works for all situations and does not implicitly rely on assumptions.  And that requires that you put some thought into choosing your sample/test data.  With that said, below is yet another alternative:
if object_id('table_1') is not null
drop table table_1;
go
create table table_1 (id int not null constraint pkx primary key, parent_id int null);
alter table table_1 add constraint fkx foreign key (parent_id) references table_1 (id);
go
insert table_1(id, parent_id)
values (101, null), (102, null), (103, null), (104, 101), (105, 101), (106, 102), (5, 102);
with cte as (select * from table_1 as src
where parent_id is not null or
(parent_id is null and exists (select * from table_1 as chld where chld.parent_id = src.id)))
select *, case when parent_id is null then 'parent' else 'child' end as record_type from cte
order by isnull(parent_id, id), case when parent_id is null then 0 else 1 end, id;
go
Notice the additional row of data I added where id = 5.   I will also add that you should use "union all" and not "union".  There is no need to force the engine to do work that is not needed.