[1.12.3 Bug] `statically_parse_unrendered_config` stringifies list/dict config values, causing false positives in `upgrade_manifest_json_dbt_version` / `state:modified`

[TomWalsh11]

Is this a new bug in dbt-core?

• I believe this is a new bug in dbt-core
• I have searched the existing issues, and I could not find an existing issue for this bug

Current Behavior

When loading a --state manifest whose metadata.dbt_version differs from the
running dbt's version, upgrade_manifest_json_dbt_version
(core/dbt/artifacts/schemas/upgrades/upgrade_manifest_dbt_version.py) re-derives each
node's unrendered_config by re-parsing its raw SQL's {{ config(...) }} call via
statically_parse_unrendered_config (core/dbt/clients/jinja_static.py) — gated by
state_modified_compare_more_unrendered_values (default true since 1.12).

That parser's construct_static_kwarg_value is typed -> str and stringifies
list/dict values unconditionally. {{ config(cluster_by=['id']) }} ends up with
unrendered_config['cluster_by'] == "['id']" instead of the real list ['id'].

Since same_config/same_contents compare unrendered_config with plain ==, this
type mismatch makes state:modified falsely flag the model as changed — even though
same_body/checksum (the actual SQL) is identical.

Affects any model with a list/dict-valued config key declared inline (cluster_by,
partition_by, grants, labels, meta, unique_key as a list, etc.) whenever
state:modified compares manifests from two different dbt-core versions/builds.

Expected Behavior

The static parser should preserve the real type of list/dict config values (or at
least produce something that round-trips via ast.literal_eval), so the upgrade merge
doesn't introduce a type mismatch against a freshly-parsed, unchanged model.

Steps To Reproduce

1. {{ config(cluster_by=['id']) }} in a model's SQL.
2. dbt parse, keep the manifest (the "new" side).
3. Copy it, edit metadata.dbt_version to a different version string (the "old" side).
4. dbt ls --select state:modified --state <old-manifest-dir>

Expected: model not selected. Actual: selected — unrendered_config['cluster_by']
was rewritten from ['id'] to "['id']" by the upgrade step.

Relevant log output

$ dbt ls --select state:modified --state ./old-manifest/
model.my_project.my_model

(despite identical `raw_code`/checksum on both sides)

Environment

- OS: macOS
- Python: 3.13.0
- dbt: 1.12.3 (also present in 1.12.0)

Which database adapter are you using with dbt?

bigquery

Additional Context

Root-caused against a real ~450-model project: calling same_body, same_config,
same_persisted_description, same_fqn, same_database_representation,
same_contract, same_ref_representation directly (via dbt.contracts.state.PreviousState)
reproduced the flagged/not-flagged status of every model exactly (366/366).

[Aastha204]

Hi, I'd like to take a look at this one. I traced it to construct_static_kwarg_value in jinja_static.py, it always returns a string, so list/dict configs get stringified and cause the false positives in state:modified.

I'm thinking the fix is to have it preserve the original type for lists and dicts instead of casting to str. Would be happy to put up a PR if this is up for grabs!

[TomWalsh11]

Hi, I'd like to take a look at this one. I traced it to construct_static_kwarg_value in jinja_static.py, it always returns a string, so list/dict configs get stringified and cause the false positives in state:modified.

I'm thinking the fix is to have it preserve the original type for lists and dicts instead of casting to str. Would be happy to put up a PR if this is up for grabs!

Hi yes that would be great!

[Aastha204]

Hi, I'd like to take a look at this one. I traced it to construct_static_kwarg_value in jinja_static.py, it always returns a string, so list/dict configs get stringified and cause the false positives in state:modified.
I'm thinking the fix is to have it preserve the original type for lists and dicts instead of casting to str. Would be happy to put up a PR if this is up for grabs!

Hi yes that would be great!

Hi,
PR is up: #16280

Changes:

• Added _try_evaluate_to_python helper that preserves native Python types (list, dict, int, bool, etc.) for literal config values
• construct_static_kwarg_value now returns Any instead of str
• Complex Jinja expressions (calls, comparisons, etc.) still fall back to string representation
• Updated existing tests + added edge-case tests for mixed lists
• All unit tests passing, code-quality checks clean

[Aastha204]

/assign-me