Version `0.10.0` breaks `dbt compile`

[colombian-cannonball]

Bug: dbt compile crashes when prefer_recent is enabled

Environment

Component	Version
upstream_prod	0.10.0
dbt-core	1.11.6
dbt-snowflake	1.11.2
Python	3.10.18
OS	Linux

Steps to reproduce

1. Set upstream_prod_prefer_recent: true in dbt_project.yml
2. Run dbt compile -s <any_model>

Error

Runtime Error
  Compilation Error in model <model> (models/marts/<folder>/<model>.sql)
    'dict object' has no attribute '_upstream_prod_cache'

    > in macro ref (macros/ref.sql)
    > called by macro ref (macros/ref.sql)
    > called by model <model> (models/marts/<folder>/<model>.sql)

Root cause

populate_cache() runs via on-run-start, which doesn't execute during dbt compile. However, dbt compile sets execute = true, so default__ref enters the prefer_recent branch and accesses graph["_upstream_prod_cache"] with bracket notation — which fails because the key was never set.

The execute is false guard at the top of default__ref does not protect against this because execute is true during compile (it's only false during parse).

Affected commands

• dbt compile
• Likely dbt show and any command that compiles without running hooks

Unaffected commands

• dbt run, dbt build, dbt test (all execute hooks)

Suggested fix

Either guard the cache access in default__ref:

{% if prefer_recent is true and dev_exists is true and "_upstream_prod_cache" in graph %}

Or ensure populate_cache always sets the key (even as an empty dict) and use .get() for the inner lookup.

Workaround

In a custom ref.sql wrapper, disable prefer_recent when the cache is absent:

{% if prefer_recent and "_upstream_prod_cache" not in graph %}
  {% set prefer_recent = false %}
{% endif %}

[LewisDavies]

Thanks for raising an issue. I appreciate the detail but it looks like it was AI-generated and doesn't include the specific command you used. I am able to run dbt compile -s <model-name> without any errors.

I'm not sure if this applies to you but I was getting this error when running preview queries through the dbt Power User VS Code extension. Adding and "_upstream_prod_cache" in graph should address that problem but I want to make sure it will fix your issue too. Can you tell me exactly what did before getting the error?