Backport IGs may cost the ecosystem more than they're worth — and SQL on FHIR should live outside core for good — Sep 30, 2025
Mixed feelings about the whole backport approach
Nikolai used the R4 subscriptions backport IG as the example: it's all Basic resources and a pile of extensions, and he's not sure it doesn't bring more mess to implementations than value.
What servers actually do — my own included — is support R6 subscriptions with R4 resources and quietly accept it isn't conformant. I wonder whether saying that out loud is the cleaner path.
The consensus for SQL on FHIR was still to go R4-compatible, on the grounds that it isn't nearly as messy a case as subscriptions.
The core is being modularised
FHIR-I has started stripping resources out of core and into incubation IGs — the test resources have already been pulled out and turned into additional resources.
The driver is compatibility: from R6 onwards, core should hold only resources mature enough not to break, so anything at risk of breaking is better off outside.
Nikolai also noted the encouraging part he sees at FHIR-I now: people are deprecating elements and adding new ones instead of breaking things outright.
SQL on FHIR should live outside core forever
Listed in the core spec as an official additional resource — since additional resources have to be registered with HL7 — but not part of core and not tied to core's release cycle, so the group keeps control of its own versioning.
Separately, Graham has been asking for SQL on FHIR to plug into the FHIR test ecosystem, which could mean eventually rewriting the group's tests into the new test resource format.
Steve — two concrete asks
- Tried the run sandbox against his own server and it works, but his server is stateless, and the sandbox runs views over whatever's in its own database — he'd like a mode where you feed it a resource inline, and offered to write it
- Wants the SQL on FHIR tester in FHIRPath Lab moved onto the current
$runoperation so that everybody is testing the same way
Rust pile-up — pool the libraries
Steve released his FHIRPath implementation in Rust, and Nikolai pointed out that three of the organisations doing anything in Rust were on the call, plus at least one more building the same primitives elsewhere.
His suggestion: get together and pool the libraries in one direction rather than each maintaining a parallel version of the same hard thing.