using `{{ ref('anything') }}` in a `set_sql_header` call resolves to the current model and not 'anything'

Describe the bug

Here’s a minimal reproducible example (running on BigQuery).

-- test_tmp_1.sql
{{
    config(materialized="table")
}}

select a.* from unnest([
    struct(1 as key, 1 as value)
]) as a

-- test_tmp_2.sql
{{
    config(materialized="table")
}}

{% call set_sql_header(config) %}
    select * from {{ ref('test_tmp_1') }};
{% endcall %}

select * from {{ ref('test_tmp_1') }}

When I dbt run this project, I get an error when test_tmp_2 is run saying something along the lines of “test_tmp_2 does not exist”.

The output SQL for test_tmp_2 after a dbt run shows:

    select * from `project`.`dataset`.`test_tmp_2`;


  create or replace table `project`.`dataset`.`test_tmp_2`
  
  
  OPTIONS()
  as (
    



select * from `project`.`dataset`.`test_tmp_1`
  );
    

Expected behavior

The ref('test_tmp_1') call in the call set_sql_header(config) block should resolve to test_tmp_1.

Instead it resolves to test_tmp_2.

System information

Which database are you using dbt with?

• postgres
• redshift
• bigquery
• snowflake
• other (specify: ____________)

The output of dbt --version:

installed version: 0.18.0
   latest version: 0.18.0

Up to date!

Plugins:
  - bigquery: 0.18.0
  - snowflake: 0.18.0
  - redshift: 0.18.0
  - postgres: 0.18.0

The operating system you’re using:

macOS 10.15.6

The output of python --version:

Python 3.8.1

Additional context

N/A

[gmailMichaelSha]

I am seeing this issue when I am trying to create a temporary table using set_sql_header with a source. Using snowflake database

{{
    config(
        materialized='table',
    )
}}
{%- call set_sql_header(config) -%}
    CREATE OR REPLACE TEMPORARY TABLE {{ target.database }}.{{target.schema}}.temp_tbl_1 AS
    SELECT * FROM {{ source('source_1', 'world_cities') }};
{%- endcall -%}

SELECT *
FROM {{ target.database ~ "." ~ target.schema ~ ".temp_tbl_1" }}

[elyobo]

Another use case: setting a variable to the result of part of a query, so that it can be referenced in multiple places. BigQuery reevaluates CTEs if they’re queried more than once (I assume there’s some rational behaviour for this, I guess due to parallelisation?) and so the recommended approach is to set a variable to the result and reference it instead.