[CT-1823] [Bug] [1.4.0rc1] --favor-state not working as intended

[jtcohen6]

Not working

In situations of "deferral," the --favor-state flag should tell dbt to always use the upstream resources defined in the --state manifest, even in cases where that resource has already been materialized in the current (dev/CI) namespace.

While #5859 added the flag appropriately, and the favor_state argument to Manifest.merge_from_artifact, the flag is not actually being piped through the task and into that method. So it never takes effect!

I think this wasn't caught because the old-style integration tests for this functionality were really tricky to read & understand. I only caught this while converting those tests to the new framework (#6616).

Repro case

-- models/model_a.sql
select 1 as id

-- models/model_b.sql
select * from {{ ref('model_a') }}

$ dbt run --target prod
$ mkdir state
$ mv target/manifest.json state/
$ dbt run --target dev
$ dbt run --target dev --select model_b --defer --favor-state --state state/

Because the --favor-state flag was passed, model_b should select from the "prod" version of model_a, even though it exists in the dev schema. Instead, it still selects from the dev version:

  create view "jerco"."dbt_jcohen"."model_b__dbt_tmp" as (
    select * from "jerco"."dbt_jcohen"."model_a"
  );

Missing from some relevant tasks

Also in v1.4, we're adding support for --defer to dbt compile + dbt docs generate. (Makes sense!) Let's make sure --favor-state is available on all the same commands as --defer.

[Tech debt] Inconsistent implementation

Also, the way --favor-state was implemented is parallel to --defer, but different from most of our other flags since v1.0:

dbt-core/core/dbt/flags.py

Lines 117 to 118 in 20c95a4

	DEFER_MODE = env_set_truthy("DBT_DEFER_TO_STATE")
	FAVOR_STATE_MODE = env_set_truthy("DBT_FAVOR_STATE")

(Namely, the argparse default is set based on the flag, rather than the other way around!)

dbt-core/core/dbt/main.py

Lines 504 to 515 in 20c95a4

	def _add_favor_state_argument(*subparsers):
	for sub in subparsers:
	sub.add_optional_argument_inverse(
	"--favor-state",
	enable_help="""
	If set, defer to the state variable for resolving unselected nodes, even if node exist as a database object in the current environment.
	""",
	disable_help="""
	If defer is set, expect standard defer behaviour.
	""",
	default=flags.FAVOR_STATE_MODE,
	)

Let's just solve for this when we move these flags/etc into the new CLI (#6585). We can always add back-compat for the older env vars + flag names — since, in the meantime, some users might add code to their projects like {{ flags.FAVOR_STATE_MODE }}.

[jtcohen6]

The old integration tests pass with or without the fix for this in ccbdcba — not very helpful!

Whereas, the new converted tests include one (test_defer_state.py::TestRunDeferStateIFFNotExists::test_run_defer_iff_not_exists) that really does fail if I un-commit the changes from ccbdcba.

I'm inclined to keep the fix + the test conversion together, to merge into main + backport to 1.4.latest. That way, we'll actually have a way to know that this new functionality is still working for all v1.4.x releases.