Fusion Diaries: exposures, strongly-typed test args, (2025-07-30)

[dataders]

EXTRA! EXTRA! Read all about it!

Here’s your last weekly Fusion diary for July

In the past week we’ve seen:

• 51 issues closed across the public and internal repos
• 8 new betas released
• a new change to the default behavior of how static analysis behaves when it encounters introspective macros (see the below-linked docs — big shoutout to the community. a lot of great discussion on this)
• a logging bug introduced and squashed (sorry!)
• a new slew of Databricks SQL expressions that are now “understood” (WINDOW, DECODE, TRY_PARSE_JSON, FROM_XML) thanks @VDFaller!

🆕 What’s been released this week

dbt Fusion engine releases

Run dbt system update to get the latest and greatest, including:

• 2.0.0-beta.54 stop leaking of trace logs, for real this time (shoutout @GClunies for both flagging and being patient with us)
• 2.0.0-beta.53 exposures (#13), strongly typed test args (#191, #250, more context below), stop the leaking of trace logs that was introduced in beta.51 (#483)
• 2.0.0-beta.52 closer conformance b/w Core and Fusion manifests
• 2.0.0-beta.51 support Snowflake ignore_case setting. support Databricks’ try_parse_json and from_xml
• 2.0.0-beta.49 support for base64-encoded DER private keys in Snowflake authentication
• 2.0.0-beta.48 hardening of source freshness (see #450) Databrick’s WINDOW function (#425), don't default-disable static analysis when encountering introspective queries (check out the new docs for details, big shoutout to the community and discussion around this!)
• 2.0.0-beta.47 fix for BigQuery’s special 4-part names for information schema (#432) Redshift list_schema
• 2.0.0-beta.46 support Databricks’ DECODE (#427), add support in BQ for time ingestion partitioning (#399)

VS Code extension + LSP

• Now if previewing a model or CTE fails, you will see an error in the result pane! #424
• when you click “Compile File” on an unsaved model file, it will first save so that you see the latest changes (or failures)
• clearer error messages when the LSP binary fails to download
• FIXED: Windows bug that prevented compilation from happening after saving a model (shoutout @sophieml for help with this!)

strongly typed test args

In last week's diary (dbt-fusion#452), we mentioned that strongly typed test arguments were landing soon. They’ve now shipped and are included in:

• beta.53
• dbt Core 1.10 (as deprecation warnings)
• dbt-autofix v0.11.0 (for auto-remediating deprecated usage)

When @andiedeleon and I encountered #191 in her team’s project almost 2 months ago, I thought the fix would be easy. But it turned out to be much thornier and kicked off a huge effort to lock down for the long-term a sustainable way to pass extra arguments to custom test definitions.

One challenge with this work was how to roll this change out without breaking current Fusion users and the package ecosystem at large. The solution was to ship partial support for the old authoring layer. Within the month we plan to help with migration for both current users as well as adapter and package maintainers.

🚧 Work in progress

UDF support

like we mentioned last week, we continue working on both:

• automatic pulldown of UDF definitions to enable static analysis in Fusion (#69)

• thinking deeply about the experience for UDFs defined and used in a dbt project

  Out-of-the-box support for UDFs dbt-core#11851

Redshift

We are in the home stretch for Redshift support! We’re on track to meet our July 31 commitment with user+pass authentication, and fast follow with more authentication methods soon. Stay tuned!

🐉 Here be dragons

bigquery: require storage read API to be enabled

I mistakenly thought this dragon vanquished but it's back from the dead! We endeavor to make it go away for good soon. See dbt-fusion#236#issuecomment-3136522903 for more context and chime in if you're seeing something similar.

deferral compilation for introspective queries

#381 is a dragon we have yet to completely vanquish since mentioning it in last week’s diary.

For those using deferral, we’re still seeing some reports of Fusion not properly deferring when introspective macros are in the mix. If you’re seeing something unexpected like the below, when trying to compile your project, please reach out!

error: dbt1501: Failed to render SQL invalid operation:
  Compilation Error for model.ga_analytics.rpt_self_service_deferred_revenue_waterfalls
      from models/marts/accounting/rpt_self_service_deferred_revenue_waterfalls.sql
  In get_column_values():
     relation ANALYTICS_DEV.dbt_dataders.rpt_self_service_accounting_activity
     does not exist and no default value was provided.
(in dbt_packages/dbt_utils/macros/sql/get_column_values.sql:21)
  --> dbt_packages/dbt_utils/macros/sql/get_column_values.sql:21:1

🤔 Looking for feedback

Have strong opinions on logs and stack traces?

Let us know! We’re focusing over the next month on the UX of the CLI.

@joellabes does a great job enumerating some of the work to be done in this Issue comment.

Reach out on Slack or open harangues as GitHub issues describing what you do or don’t like.

👓 Stuff you should read

UDF support

check out @graciegoheen’s discussion about UDF support if you haven’t yet! There’s some big squishy ideas in there that only the dbt community can help us iron out. For example:

Should a UDF in dbt be more like a macro? or a model materialization? 🤔

dbt-labs/dbt-core#11851

Highway New Concepts: Revisited

We’ve overhauled the Fusion New Concepts docs with the goal of making the information more easily digestible. There’s a very pretty animated diagram that illustrates the difference between AOT and JIT strategies! Please check it out — especially if the information didn’t click with you previously

Arrow

this one is extra credit for those that love reading about “lower-level” technologies that are changing the face of the data industry. The Fusion engine relies heavily on Arrow. We have more to say on why we’re betting on Arrow and the future it enables. Until then, if you’re interested, I highly encourage to give Ian David and Matt’s article a read

https://arrow.apache.org/blog/2025/01/10/arrow-result-transfer/

🏁 Made it to the meme

The Fusion engineers got together for an offsite last week, but since they didn’t take a team picture here’s the Fusion engine…ears.