Parse model hooks for ref statements

[drewbanin]

Describe the bug

Hooks are not currently "parsed" at parse-time - they are just executed directly at runtime. As a result, a ref in a hook can result in an error message like:

dbt was unable to infer all dependencies for the model "table_a".
  This typically happens when ref() is placed within a conditional block.
  To fix this, add the following hint to the top of the model "table_a":
  -- depends_on: {{ ref('table_b') }} 

This error can be avoided by parsing model hooks at parse-time.

Steps To Reproduce

-- models/model_a.sql

select 1 as id

-- models/model_b.sql

{{ config(post_hook="select * from {{ ref('model_a') }}") }}

select 1 as id

run the models shown above and observe the error message.

Beware

Parsing hooks should be straightforward and a non-issue most of the time, but some users might be (read: are definitely) doing weird things in hooks. All of the adapter methods that invoke SQL queries should be no-op'd at parse-time.... are there valid hooks that would fail if dbt starts parsing them?

Example hook that's no problem to parse

grant select on {{ this }} to db_reader

Example hook that may be problematic (contrived)

{% for node in graph.nodes.values() %}
  -- this wouldn't work anyway - just an example
  grant select on schema.{{ node.identifier }}
{% endfor %}

We should better codify the hook context available at parse-time and run-time and document what kinds of hook queries, if any, are no longer supported (or how to work around this change).

[drewbanin]

closed by #2025