A 3.0.0 release branch, the last ballot tickets, and self-documenting external tables — Nov 17, 2026
A release branch for 3.0.0
John Grimes opened with how the spec will be versioned from here. The existing ballot tag becomes a release branch for 3.0.0, all ballot PRs are retargeted to it, and main carries on with new work such as terminology in SQL and external tables. John will also move a couple of editorial and example fixes into the 3.0.0 branch.
Main will not get a plain 3.1.0 version, so CI builds are never mistaken for a release. Gino Canessa listed the official labels - snapshot, ballot, draft and CI build - and the group leaned toward a "-draft" suffix. Gino prefers the dash form because the label rules as written break once a counter goes past nine. He also named the cost of two branches: every fix has to land on both, and some get missed.
The last ballot tickets
John has now put a PR and a proposed disposition comment on every ballot ticket. Once the last three are discussed, Gino will move the tickets to proposed dispositions and file a single block vote, which gives everyone a week to review. Technical corrections need no vote, but tickets filed as change requests still do, even when they are really technical corrections.
- Shareable ViewDefinition (FHIR-58974). The profile only requires url today. It will be tightened to match CRMI Shareable Library, adding publisher, version and description. Nobody objected, and John will open the PR.
- Primitives-only constraint (FHIR-59164). date and uri were missing from the invariant on the Tabular ViewDefinition, with no reason found. The group agreed to merge the existing PR #406.
- FHIRPath version (FHIR-58971). The ticket asks for an element that declares the FHIRPath version. Gino is not convinced it is needed now that FHIRPath 3.0 is published, so the spec could simply say it targets 3.0. Nikolai Ryzhikov thinks the SQL on FHIR version matters more, and the resource version already tells you that. The question may come back briefly next meeting.
I'm assuming we're just gonna say, "Yeah, we're targeting FHIRPath 3.0," and that's it.
The experimental FHIRPath functions section also needs rewording, since those functions are now part of FHIRPath 3.0. They stay in a separate section so conformance is not affected.
Logical models for external tables
Gino demoed a playground that generates logical models - StructureDefinitions - from the tables of a SQLite database, so the schema documents itself and SQL queries and ViewDefinitions can be built on top. For him the main value is descriptions of tables and columns, and he sees it applying to EMR exports such as Epic's EHI export.
Right now if you get one of those dumps, it's just a bunch of tables, and there's no descriptions on them. Being able to add the descriptions to me is critical, because then it's a documented way of exporting your schema.
Nikolai suggested the same annotation could describe the output of a ViewDefinition, and John liked the idea of a runner operation that returns the logical model for a view's output, which would help create a target table for the results. Nikolai estimates Gino's work and his own external tables proposal are about 80% the same, and Gino will reconcile the two. Gino is also looking at mapping more ElementDefinition properties - bindings, min/max, mustHaveValue for null flags - and John asked which parts of ElementDefinition should be disallowed.
SQL types and dialects
The prototype keeps the SQL type and dialect in extensions, which Gino calls rough drafts. The group agreed a SQL type and an optional dialect are needed. John sees value in more than one dialect, for example ANSI SQL plus a vendor one, and Nikolai added that CSV and Parquet outputs would differ in types too.
SQL types are not really a code, they're a grammar, right?
Nikolai noted that most ORMs and schema DSLs just use a free string for the type. Gino will try a complex extension on the ElementDefinition type that holds a dialect and a string, and compare it with using mappings. John suggested ElementDefinition could also carry details like varchar length, and Gino plans to add that next.
Bulk export responses
Nikolai asked Adam Culbertson whether the bulk export response could move to Parameters, in line with how SQL on FHIR defines its async pattern, and would deprecate the custom JSON. Gino explained that the bulk community, which has this in production, will not change it: the response stays a Binary-style JSON with an extension pointing to a logical model, and the form of that extension is still being argued. The bulk work is nearly done and will publish as is, so any change needs a ticket right away.
Adam is concerned about provider directory and scheduling use cases. He may avoid bulk publish for them, or pair it with subscriptions or streaming JSON.