ADR 0067 — @block directive

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

Context

ADR-0066 (column-scoped data block) established that a sheet's data block is the maximal connected rectangle of marker cells extended through adjacent non-empty cells, and that the engine supports at most one such block per sheet at 0.7.x. The single-block restriction was intentional — multi-block detection was deferred until concrete use cases emerged and an explicit syntactic boundary marker was in the spec.

Two patterns surface concrete need for multi-block:

1. Side-by-side tables — a single sheet holding two unrelated data blocks, possibly bound to different sources (e.g., customer list on the left, product list on the right; or daily sales by region in two parallel columns).
2. Vertically stacked tables — two blocks on the same sheet at different starting rows, each iterating its own source (e.g., a header summary block at row 4 with one source and a detail block at row 20 with another).

The Phase 1 implicit auto-detection rule cannot disambiguate these cases — multiple disconnected marker clusters raise xl3/expression/bracket-outside-block at parse time today. An explicit directive is needed so authors can declare each block's geometry and the engine has unambiguous boundaries to render against.

Considered Options

A. @block directive with three grammar forms (this ADR). Authors declare each block via a directive cell placed above the block, with optional Excel-native range syntax for explicit boundaries:

• {{ @block }} — bare; col-range inferred from marker cells below
• {{ @block A:D }} — explicit col-range, row-range auto-detected
• {{ @block A2:D7 }} — fully explicit rectangle

B. @table instead of @block. Same semantics, Excel-Table- inspired naming. Familiar to spreadsheet authors but creates a vocabulary mismatch with spec/internal "data block" terminology.

C. No directive — auto-detect everything. Use a more aggressive implicit clustering algorithm that can return multiple blocks per sheet. Smaller spec surface but prone to silent reclassification when authors edit templates (a stray [col] reference could silently create a new block).

D. Range-only spec (no bare form). Always require @block A:D or @block A2:D7. More verbose but zero ambiguity.

Option A chosen. Naming aligns with internal DataBlock type and spec's existing "data block" usage (B rejected for vocabulary consistency). The three-form grammar gives authors a typing-cost gradient: bare for the common case, col-range when the data row template doesn't fill the intended block extent, full-rect when authors want zero ambiguity (e.g., in heavily reviewed templates or auto-generated output).

The implicit auto-detection from ADR-0066 stays — @block is opt-in, used only when an author needs explicit boundaries (disambiguation, multi-block, range extension). Templates without any @block keep working unchanged.

Decision

A new directive is added to the spec's directive list:

@block                  — bare; col-range auto-detected from {{...}} markers
@block <col-range>      — Excel column-letter range, e.g., A:D
@block <full-range>     — Excel A1-style rectangle, e.g., A2:D7

Grammar:

block_directive   ::= "@block" ( WS excel_range )?
excel_range       ::= col_letters ":" col_letters                          # "A:D"
                   | col_letters digit+ ":" col_letters                    # "A2:D" — start-row + col-range
                   | col_letters digit+ ":" col_letters digit+             # "A2:D7" — full rectangle
col_letters       ::= [A-Z]+

A:D7 and A:D7 with mismatched col-letter style (lower-case, mixed-case) raise xl3/directive/invalid-syntax — @block arguments follow standard Excel cell-reference conventions.

Semantics (placement and scope)

A @block directive cell sits in a cell whose row is strictly above the block's first row. The directive's column position and the optional range argument together determine the block's bounding box:

• Bare {{ @block }}:

  • Block start row = first row below the directive's row that contains a marker cell ({{ ... }}).
  • Block end row = last consecutive marker-containing row.
  • Block col-range = bounding box of marker cells in those rows, extended through adjacent non-empty cells per ADR-0066.

• Col-range {{ @block A:D }}:

  • Block col-range = explicit [A..D].
  • Block start row / end row = auto-detected same as bare.
  • Marker cells inside [A..D] MUST be present in the inferred row range; if none exist, xl3/block/empty-table is raised.

• Full-rect {{ @block A2:D7 }}:

  • Block row range = [2..7] explicit.
  • Block col-range = [A..D] explicit.
  • Marker cells inside [A2:D7] MUST be present; otherwise xl3/block/empty-table.

Interaction with @source / @filter / @sort / @group / @top / @repeat / @join

These directives are NOT folded into @block — they remain independent directives that the engine attaches to the relevant block via the per-block proximity rule (ADR-0069). An author can combine:

{{ @block A:D }} {{ @source Customers }} {{ @filter [Status] = "VIP" }} {{ @sort [Revenue] desc }}
{{ [Account] }}  {{ [Name] }}  {{ [Region] }}  {{ [Revenue] }}

Both @source, @filter, @sort attach to the @block A:D they share a row with (col-overlap + same row direction matching). The directive ordering on the row doesn't matter; spec evaluation order is fixed (ADR-0029).

Consequences

Backward compatibility — strictly additive. Templates with zero @block directives continue to work via Phase 1's implicit cluster detection (ADR-0066). Authors only need @block when they want explicit boundaries (multi-block, range extension, or self-documenting clarity).

One new error code — xl3/block/empty-table, raised when a @block rectangle contains no marker cells. The existing xl3/expression/bracket-outside-block (ADR-0066) is reused for the "orphan marker outside any block" case when multi-block is in play (see ADR-0068). G3's 30-day catalog-frozen clock resets to today.

Grammar additions. block_directive, excel_range, col_letters are appended to spec/language.md's grammar fragment alongside group_directive / subtotal_directive (ADR-0058).

Spec surface — three new forms. Deliberately narrower than "any Excel range can become a block declaration" — @block only accepts column-range or rectangle forms, not whole-row forms like 2:7 (rows-only is hard to read at a glance, and the bare form covers the common case where the row range is obvious from markers).

Fixture coverage — fixtures 149 (block-col-range-explicit) and 150 (block-full-rect-explicit) exercise the explicit forms; 146 (multi-block-explicit-two-tables) exercises the bare form in a multi-block sheet. Together they pin the three grammar forms against the conformance corpus.

Multi-block enables but does not require multi-source. A single source can feed multiple blocks (the same source's rows iterated in two output regions). Per-block @source makes multi-source explicit when needed.

References

• ADR-0066 — column-scoped data block; primitive this directive builds on.
• ADR-0029 — directive composition and evaluation order; @block slots into the existing directive pipeline.
• ADR-0058 — @subtotal row composition; precedent for adding new directives with their own grammar production.
• Future ADR-0068 — multi-block detection (strict mode rules).
• Future ADR-0069 — per-block directive scoping (proximity rule).