본문으로 건너뛰기

ADR 0068 — Multi-block detection (strict mode)

  • Status: accepted
  • Date: 2026-05-24
  • Spec target: XTL 0.1
  • Affects: language.md (Data Blocks); impl (parser); 1 new error code

Context​

ADR-0066 established column-scoped data blocks but limited a sheet to at most one block, with disconnected marker clusters raising xl3/expression/bracket-outside-block at parse time. ADR-0067 introduced the @block directive for explicit boundary declaration.

The remaining design question: when both implicit detection and explicit @block directives are in play on the same sheet, how does the engine decide which cells belong to which block? Two competing models:

  • Permissive (mixed-mode) — explicit @block directives claim their rectangles; any leftover marker cells outside all explicit blocks form an additional implicit block subject to ADR-0066's cluster rules. Friendly for migrating an existing single-block template by adding a sidebar via just one new @block on the sidebar.
  • Strict — if a sheet has any @block directive, ALL marker cells on that sheet MUST be inside some @block rectangle. Mixed mode is forbidden; the implicit auto-detection runs only when zero @block directives are declared.

Permissive is friendlier per-migration; strict is friendlier per- review and per-spec.

Considered Options​

A. Strict mode (this ADR). One-of:

  • ZERO @block on sheet → ONE implicit block per ADR-0066's rules. Cluster-completeness check applies (multiple disconnected clusters raise xl3/expression/bracket-outside-block).
  • ≥1 @block on sheet → all marker cells must lie inside some @block rectangle. Orphan markers raise xl3/expression/bracket-outside-block. No implicit cluster detection runs.

B. Permissive (mixed) mode. @block directives claim their rectangles; leftover marker cells form additional implicit blocks subject to ADR-0066 cluster checks. Up to N+M total blocks per sheet (N explicit + M implicit).

C. No-mix forbidden — sheets are either all-implicit or all-explicit. Same as strict A but framed as a sheet-mode toggle. Effectively equivalent to A.

Option A chosen. Strict mode trades a small one-time migration cost (when an author wants a second block, they must add @block to the existing block too, not just the new one) for substantial review and debugging clarity:

  • Reading a template's first @block instantly tells the reader "this sheet uses explicit blocks; check the others." No silent hidden implicit block.
  • Adding a stray marker reference somewhere on the sheet immediately raises an error pointing at the location, rather than silently extending or creating a new implicit cluster.
  • Conformance testing the "any @block present" rule is a one-line predicate; mixed mode requires per-cell membership tracking across N explicit + M implicit blocks.

The migration cost is bounded: a real-world transition adds one @block directive cell to the existing block (one keystroke effort). Existing templates without @block continue to work unchanged via ADR-0066 implicit detection.

Decision​

The following becomes normative spec text in language.md's "Data Blocks" section (extending the existing single-block description):

A sheet's data block(s) are determined by which directives the author declared:

Implicit mode (no @block directives on sheet). Exactly one data block per sheet, detected via the marker-cluster rule (ADR-0066). If two or more disconnected marker clusters exist, xl3/expression/bracket-outside-block is raised.

Explicit mode (one or more @block directives on sheet). Each @block directive defines one data block whose row and column ranges are determined by the directive's form (bare / col-range / full-rect; ADR-0067). All marker cells ([Column] references outside aggregates) on the sheet MUST lie inside some @block rectangle; orphan markers raise xl3/expression/bracket-outside-block and identify the cell location.

@block rectangles MUST NOT overlap. Overlap (row or column intersection between any two @block declarations on the same sheet) raises xl3/block/overlap at parse time and identifies the two blocks.

Mode determination​

The parser checks the sheet's cells for @block directives before running implicit cluster detection:

  • If at least one @block cell exists, mode = explicit.
  • Otherwise, mode = implicit (ADR-0066 detection runs as today).

The two modes never coexist on the same sheet by construction; this is the "strict" choice.

Multi-block sheet layout examples​

Side-by-side two tables (both bound to default source):

{{ @block A:C }}                          {{ @block E:H }}
{{ [Account] }} {{ [Name] }} {{ [Total] }} {{ [Sku] }} {{ [Price] }} {{ [Stock] }} {{ [Vendor] }}

Vertically stacked two tables (different sources):

{{ @block }} {{ @source Customers }}
{{ [Account] }}  {{ [Name] }}

(gap rows here are fine — they're between blocks, not inside one)

{{ @block }} {{ @source Vendors }}
{{ [Vendor] }}  {{ [Country] }}

Mixed orientation — explicit + an outside-cell side note (no implicit second block; the static side note isn't a block):

{{ @block A:D }}                                  side note (outside)
{{ [Customer] }} {{ [Revenue] }} {{ [Region] }} {{ [Status] }}

Cross-block reference semantics (informational)​

Per ADR-0066, outside cells are preserved at their original row positions and their formula text is preserved verbatim. With multi-block, "outside" means outside all blocks on the sheet.

If an outside cell references a cell inside a block (e.g., a side note formula =B5), the engine does not rewrite the reference — the block-expansion may have shifted block-interior cells, so the reference can become stale. This is documented in ADR-0066 and applies identically in multi-block mode. Workaround patterns: use whole-column references (SUM(B:B)), or place the formula inside a block's column range so it shifts together with the data.

Block-to-block references (a cell in block A referencing a cell in block B) are similarly user-managed; the engine doesn't track or guarantee cross-block reference validity.

Consequences​

Backward compatibility — strictly additive. Implicit-mode templates (no @block) are unchanged. The xl3/block/overlap error is new but can only be raised by templates using @block — which didn't exist before this ADR.

Two error codes affected:

  • xl3/expression/bracket-outside-block (existing per ADR-0066) — its meaning is extended: in explicit mode it also fires for marker cells outside any declared @block.
  • xl3/block/overlap (new) — added to catalog. G3 30-day clock resets to today (compound reset with ADR-0067's xl3/block/empty-table).

Migration cost is bounded. Adding a second block to an existing single-block template requires the author to also add @block to the existing block (one cell). No other change is required.

Implementer note. The mode-determination pass should run before cluster detection so the parser doesn't waste work computing implicit clusters that will be discarded in explicit mode. The overlap check requires O(B²) rectangle intersection across the sheet's @block set (B = number of @block directives per sheet); in practice B ≤ 5 or so for realistic templates, so the quadratic cost is negligible.

Conformance fixtures:

  • 146-multi-block-explicit-two-tables — two @block directives, different col-ranges, default source.
  • 147-multi-block-different-sources — per-block @source binding.
  • 148-multi-block-different-start-rows — stacked vertically.
  • 151-block-overlap-error — xl3/block/overlap.
  • 152-block-empty-table-error — xl3/block/empty-table (per ADR-0067; covers the multi-block case where one block has no markers).

145-block-bracket-outside-error (already shipped in 0.7.x) was written assuming the Phase 1 single-block enforcement and remains valid: in implicit mode it still raises for disconnected clusters.

References​

  • ADR-0066 — column-scoped data block; cluster detection rule.
  • ADR-0067 — @block directive grammar.
  • ADR-0069 — per-block directive scoping (companion to this ADR).
  • ADR-0015 — append-only error code catalog principle (G3 clock reset note above).