docs: ADR-0044 relationship visualization (V1); accepted

Two-table connector diagrams (Style A) for relationships, resolving
ADR-0016 OOS-1 and the open half of requirements.md V1. Reach is
'relationship-relevant' (show relationship / show table / relationship
DDL echoes; incidental echoes keep prose). App-side rendering, width-
adaptive side-by-side vs vertical, compound-FK pair routing, bold box
title rows. Revised after a /runda DA pass corrected three inverted-
architecture claims (App-side rendering, untracked output width,
prose-in-worker show-relationship).

Index updated per ADR-0000.
This commit is contained in:
claude@clouddev1
2026-06-09 21:47:35 +00:00
parent b17148b467
commit bb02dfb752
2 changed files with 493 additions and 0 deletions
+492
View File
@@ -0,0 +1,492 @@
# ADR-0044: Relationship visualization (two-table connector diagrams)
## Status
Accepted (2026-06-09)
Resolves **ADR-0016 OOS-1** ("Relationship visualization — two
structures side-by-side with an arrow; its own ADR; will compose
`render_structure`"). Partially **supersedes ADR-0016 §5** (the
plain-text `References:` / `Referenced by:` relationship block) and
**extends ADR-0016 §4 and §6** (adds layout width-awareness and
per-span styling). Builds on ADR-0028 (styled output runs),
ADR-0043 (compound, list-based relationship endpoints), ADR-0013
(named 1:n relationships), and the V5/V5a `show` family. Honours
ADR-0009 (DSL conventions) and ADR-0002 (no engine name in
user-facing strings).
Closes the substantive open half of requirement **V1**
(`docs/requirements.md`): "a selected relationship as two tables
joined by a line."
## Context
The table-structure half of V1 is done: `show table <T>` renders a
box-drawn structure table (columns / types / constraints / indexes),
and relationships appear **as prose** beneath it —
`References:` / `Referenced by:` blocks formatted `A.col → B.col`
(`output_render.rs render_structure`, per ADR-0016 §5). The same
prose appears in V5a's `show relationship <name>` detail view.
The open piece is the **visual** form: drawing a relationship as two
tables side by side joined by a connecting line, with cardinality and
referential actions. ADR-0016 deliberately deferred this (OOS-1) and
**pre-sized the structure renderer to compose two of itself**. This
ADR cashes that in.
Three facts from the current architecture shape the design:
1. **Rendering is App-side; the worker returns structured data.**
*(Corrected from the initial draft after direct verification — an
earlier survey had this backwards.)* The database worker (`db.rs`)
returns `TableDescription` / `DataResult`; the **App** (`app.rs`)
calls the `output_render.rs` helpers to format them into
`OutputLine`s. Verified: every `render_structure` /
`render_data_table` call site is in `app.rs` (557, 1669, 1732,
1752, 1814, 1678, …); **none in `db.rs`**. Width and theme
therefore live App-side. The **one exception** is the V5/V5a
`show <kind> [<name>]` family: `do_show_one` / `show_list` build
prose `Vec<String>` **in the worker** from `RelationshipSchema` /
index metadata, carrying *no* `TableDescription` — so
`show relationship <name>` is the path that must be restructured
(§6). The output buffer is a flat `VecDeque<OutputLine>` — a
historical log of lines, not re-renderable widgets (live
re-rendering is V4 territory).
2. **Relationships are list-based (ADR-0043).** `RelationshipSchema`
/`RelationshipEnd` endpoints are `Vec<String>` column lists,
positionally paired. Single-column FKs are the one-element case;
the diagram must render compound FKs too.
3. **Only 1:n relationships exist.** Relationships are declared 1:n
(ADR-0013); m:n (requirement C4) is unbuilt, and there is no
UNIQUE-target 1:1. So cardinality is always **1 on the parent
(referenced) side, n on the child (referencing/FK) side**.
The user has chosen the design direction across three decisions
(recorded below as made): **visual style**, **where it appears**, and
**multi-relationship layout**.
## Decision
### 1. Scope and trigger — diagrams where relationships are the subject
Relationship diagrams replace the prose relationship form on the
surfaces where **relationships are the subject of the command**; no
new command or sigil is added (ADR-0009: one sigil, keyword grammar).
This is the user-chosen **"relationship-relevant"** reach (over a
global replacement of every structure echo, or a show-commands-only
scope — see the DA pass).
Diagram surfaces:
- **`show relationship <name>`** (V5a) renders **one** diagram: the
parent and child tables as **full structure boxes** joined by a
connector. The canonical base unit.
- **`show table <T>`** keeps T's full structure box at the top
(unchanged from ADR-0016 §5), then a **Relationships** section:
**one compact connector diagram per relationship**, stacked
vertically (§4).
- **Relationship DDL auto-shows** — the structure echo after
`add 1:n relationship`, `drop relationship`, and a future
`modify relationship` (C3a) — render their relationships as
compact diagrams (§4), since the user just acted on a relationship.
Prose-retained surfaces (**unchanged** from ADR-0016 §5):
- **Incidental DDL auto-shows** — the structure echo after
`create table`, `add`/`drop`/`rename`/`change column`,
`add`/`drop index` — keep the terse `References:` /
`Referenced by:` prose. A simple `add column` on a heavily-related
table should not print a wall of diagrams.
So this **partially supersedes ADR-0016 §5**: the prose block is
replaced by diagrams on the relationship-subject surfaces and
retained on incidental ones. No information is lost on either —
relationship name, endpoints, cardinality, and `on delete` /
`on update` actions all appear in both forms.
Mechanically (§6) this is a **render mode** on `render_structure`
(`Diagram` vs `Prose`), selected by the calling command; the items/
side panel (`ui.rs render_items_panel`) is a navigation tree and is
untouched.
A future **user-configurable setting** (e.g. always-prose /
always-diagram / auto-by-width, for small-screen users) is a clean
follow-up and is **out of scope here** (OOS-7) — the per-command
default above is the v1 behaviour.
### 2. The base diagram — Style A (structure + connector)
A relationship diagram is two boxes plus a connector. Reading
convention, applied **uniformly** (this is what makes
multi-relationship lists scannable):
- **Child (FK holder) on the left, parent (referenced) on the
right.** The connector arrow always points **left → right**, in the
direction of the reference (child *references* parent).
- **Cardinality** sits at the connector ends: **`n`** at the child
(left) end, **`1`** at the parent (right) end.
- **Referential actions** (`on delete …`, `on update …`) label the
connector beneath the line. `no action` is shown verbatim
(ADR-0009 wording), abbreviated to the action keyword when space is
tight.
#### 2.1 Box anatomy — the table name must stand out
Every box (full or compact) carries the table name as a **bold title
row separated from the columns by a rule** (`├─┤`), styled in the
table-name theme class (§5). This directly addresses the requirement
that a name never read as just another column row.
Full structure box (used in `show relationship`): title row, then the
ADR-0016 column/type listing, with key markers (§2.2):
```
┌──────────────────────┐
│ orders │ ← bold title row (table name)
├──────────────┬───────┤
│ id (PK) │ int │
│ customer_id ●│ int │ ● marks the FK column in this relationship
│ total │ real │
└──────────────┴───────┘
```
Compact box (used in `show table`'s stacked list): title row, rule,
then **only the column(s) participating in this relationship** — the
focal table's full structure is already shown above, so the compact
box stays small and the focal table is not redrawn in full N times:
```
┌──────────────┐
│ orders │ ← bold title row
├──────────────┤
│ customer_id ●│ only the FK column(s) for this relationship
└──────────────┘
```
#### 2.2 Key markers
- The PK is annotated `(PK)` (consistent with the existing structure
view's constraint column).
- The **endpoint columns of *this* relationship** are marked with a
filled dot `●` adjacent to the column name — on the child FK
column(s) and the parent referenced column(s). The connector
attaches at these `●` rows.
#### 2.3 The connector (single-column)
`show relationship Customers_Orders` (child `orders.customer_id`
→ parent `customers.id`):
```
orders customers
┌──────────────────────┐ ┌─────────────────────┐
│ orders │ │ customers │
├──────────────┬───────┤ 1 ┌──●│ id (PK) │
│ id (PK) │ int │ │ ├─────────────┬───────┤
│ customer_id ●│ int │ n ───────────┘ │ name │ text │
│ total │ real │ │ email │ text │
└──────────────┴───────┘ └─────────────┴───────┘
on delete cascade · on update no action
```
**Connector routing.** The two `●` endpoint rows are generally at
different heights. The connector leaves the child box's right edge at
the child-`●` row, travels horizontally into a **gutter channel**
between the boxes, jogs vertically to the parent-`●` row, then enters
the parent box's left edge with an arrowhead. Box tops are aligned.
(Exact glyphs/spacing are pinned by insta snapshots in
implementation; the mockups here are representative, not normative.)
#### 2.4 Compound foreign keys (ADR-0043)
For an n-column FK, the n child `●` columns and n parent `●` columns
are each listed, and **one connector is routed per positional pair**.
A summary line states the pairing explicitly so it is unambiguous
even if the routed lines are visually close:
```
orders customers
┌──────────────────────────┐ ┌──────────────────────┐
│ orders │ 1 ┌●│ region (PK) │
├──────────────┬───────────┤ │ │ id (PK) │
│ cust_region ●│ text │ n ─────────┤ ├────────────┬─────────┤
│ cust_id ●│ int │ n ─────────┘ │ name │ text │
│ … │ … │ └────────────┴─────────┘
└──────────────┴───────────┘
(cust_region, cust_id) ──► customers.(region, id)
on delete cascade · on update no action
```
### 3. Width handling and the narrow-terminal fallback
The diagram is rendered **once, App-side, at the output-panel width
current when the command runs**. The `App` does **not** track the
panel width today — `note_output_viewport(visible_rows,
total_wrapped_rows)` records only row counts for scroll math
(verified; `App` has no width field). This ADR adds a
`last_output_width: u16` to `App`, set from `ui.rs` where the output
panel's `inner.width` is already computed (next to the existing
`note_output_viewport` call), with an `80` default before the first
render. The App-side render path then chooses layout per diagram:
- **Side-by-side** (§2) when both boxes plus the gutter fit the
available width.
- **Vertical stack** when they do not: parent box above, child box
below, connector running **downward** through a short vertical
channel, cardinality and actions labelling it. This keeps a real
diagram at any width instead of degrading to prose:
```
┌──────────────┐
│ orders │
├──────────────┤
│ customer_id ●│ n
└──────────────┘
│ on delete cascade
▼ on update no action
┌──────────────┐
│ customers │
├──────────────┤
│ id (PK)●│ 1
└──────────────┘
```
- **Last-resort helper.** Only when even a single vertical diagram
cannot fit (e.g. an extremely narrow pane, or a box wider than the
pane), `show table` emits a one-line pointer per relationship —
`<child> → <parent> · run `show relationship <name>` for the
diagram` — so nothing is silently dropped. `show relationship`
itself always renders (it is the detail view); if its single box is
wider than the pane, ratatui's existing right-edge handling applies
(ADR-0016 §4), no new truncation.
**No live reflow.** Because the output buffer is a historical line
log, resizing the terminal *after* a diagram is rendered does not
reflow it — identical to how every other rendered output behaves
today. Reflow-on-resize belongs to V4's re-renderable journal and is
explicitly out of scope (OOS-2).
This **extends ADR-0016 §4**: that ADR added no width-awareness
(relying on ratatui truncation); this ADR adds width-awareness for
the **layout decision** only. It still performs **no per-cell
truncation** (ADR-0016 OOS-4 stands).
### 4. `show table` multi-relationship layout — stacked compact diagrams
After T's full structure box, a **Relationships** heading, then one
**compact** diagram (§2.1) per relationship, stacked vertically.
Ordering matches the current prose order: **outbound** (T is the
child — "References") first, then **inbound** (T is the parent —
"Referenced by"), each group ordered by relationship name. The
child-left/parent-right rule (§2) is applied per diagram, so the
focal table appears on the left for its outbound relationships and on
the right for its inbound ones — consistent with the reference
direction.
```
orders
┌──────────────┬──────┐
│ orders │ │
├──────────────┼──────┤
│ id (PK) │ int │
│ customer_id │ int │
│ total │ real │
└──────────────┴──────┘
Relationships
┌──────────────┐ n 1 ┌─────────────┐
│ orders │────────────────│ customers │
├──────────────┤ on delete ├─────────────┤
│ customer_id ●│ cascade │ id (PK) ●│
└──────────────┘ └─────────────┘
┌──────────────┐ n 1 ┌─────────────┐
│ order_items │────────────────│ orders │
├──────────────┤ on delete ├─────────────┤
│ order_id ●│ cascade │ id (PK) ●│
└──────────────┘ └─────────────┘
```
Stacking (rather than a single focal-centred subgraph with fan-out
connectors) is chosen because: it never produces crossing lines; it
scales to any number of relationships via the pane's existing
vertical scroll; each diagram is only two boxes wide, so it fits
nearly any terminal; and it keeps the per-diagram width logic of §3
trivial. The cost — the focal table name repeats per diagram — is
mitigated by the compact box showing only the participating
column(s).
### 5. Styling
Styling uses the **ADR-0028 styled-run mechanism** (`OutputSpan` /
`OutputStyleClass` on `OutputLine`), not raw text. New style classes
(theme-defined, legible light/dark per ADR-0016 §6 / NFR-7):
- **table name** — bold, accent colour (the "stand out" requirement);
- **key marker** (`●`, `(PK)`) — a distinct accent;
- **connector** (box-drawing line + arrowhead) — muted;
- **cardinality** (`1` / `n`) — emphasised;
- **referential action** label — muted/secondary.
This **extends ADR-0016 §6** (which set up `OutputKind::System`
styling but no per-element theming) by reusing the per-span
`OutputSpan` path ADR-0028 later introduced — which is **already
produced App-side** (`output_render.rs:299-332` builds styled runs
for the explain-plan tree; `app.rs` constructs `OutputLine`s with
`styled_runs: Some(..)`; `ui.rs:863` renders them). So **no
worker→UI contract change is needed**: the new `output_render`
diagram functions return styled lines directly on the App side
(§6).
### 6. Implementation
All rendering is **App-side** (per the corrected Context fact 1), in
`output_render.rs` (hand-rolled, ADR-0016 §7), returning styled
lines that `app.rs` pushes as `OutputLine`s with `styled_runs` set.
No worker `available_width` and no worker→UI contract change.
- **New renderer functions in `output_render.rs`**, composing the
existing box primitives and emitting styled spans (§5):
- `render_relationship_diagram(child, parent, rel, width, full) ->
Vec<OutputLine>` — `full` selects full vs compact boxes;
internally decides side-by-side vs vertical (§3) from `width`;
routes the connector(s) including compound pairs (§2.4).
- A shared helper for the box title row + key markers.
- The two call paths supply `width` from the new
`App::last_output_width` (§3).
- **`show table <T>` (already App-side).** The focal
`TableDescription` reaches `app.rs` and is rendered by
`render_structure` today. `render_structure` is refactored to emit
the focal structure box, then a **Relationships** section built
from the focal description's `RelationshipEnd`s (which already
carry neighbour table name + participating columns + cardinality +
actions — **enough for the compact box**; no neighbour
full-structure fetch needed). `render_structure` gains a
**relationship render mode** (`Diagram` | `Prose`); the caller
selects it (§1): `show table` and the relationship DDL echoes pass
`Diagram`, incidental DDL echoes pass `Prose`. The generic
`handle_dsl_success(command, description)` (`app.rs:1669`) already
has the `Command`, so the mode is a function of the command
variant; the incidental-only call sites (`app.rs:557/1732/1752/
1814`) pass `Prose`.
- **`show relationship <name>` (must be restructured).** Its
worker path `do_show_one` currently returns prose `Vec<String>`
from a `RelationshipSchema` only. To draw the **full** diagram the
App needs **both** endpoint `TableDescription`s. The relationship
detail reply is upgraded to carry the `RelationshipSchema` plus
both endpoints' `TableDescription`s (the worker already has
`do_describe_table`); the App renders the diagram. The
"No relationship named `X`." not-found line is preserved.
- **Self-referential FKs** (`parent_table == child_table`; supported
in the grammar): render as two boxes bearing the same table name
(child-left copy with the FK column, parent-right copy with the
referenced column), connector as usual — clearer than a self-loop
glyph in a TUI. Covered by a snapshot test.
- **Out of these surfaces:** the items/side panel (`ui.rs:582`
`render_items_panel`) is a navigation tree, not a relationship
view, and is unchanged.
- **Database engine name** never appears in any rendered string
(ADR-0002).
### 7. Testing
- **Insta snapshots** (Tier 2) pin exact rendered output for:
single-column 1:n (`show relationship`); compound FK; the
narrow-terminal vertical fallback; the last-resort helper line;
a self-referential FK; and a `show table` with both an outbound and
an inbound relationship (stacked compact list).
- **Unit tests** (`output_render.rs`): the side-by-side-vs-vertical
**width-threshold** decision at boundary widths; connector routing
for endpoints at differing row heights; compound-pair routing;
key-marker placement.
- **Tier-3 integration**: `show relationship <name>` and
`show table <T>` produce diagram output (not prose) end-to-end
through the worker.
- **Existing tests/code to update** (enumerated from a DA grep — the
supersession of ADR-0016 §5 is not abstract):
- `output_render.rs:121,135` — the `References:` / `Referenced by:`
prose-emitting code in `render_structure`;
- `output_render.rs:78` — the docstring deferral note (OOS-1);
- `output_render.rs:793` — unit test asserting `"Referenced by:"`;
- `src/snapshots/…render_structure_with_relationships.snap` — the
prose snapshot;
- `tests/it/walking_skeleton.rs:477,530` — integration asserts on
`"Referenced by:"` (and the comment at 433);
- `src/dsl/command.rs:984` — a comment referencing the prose (no
behaviour).
If the scope fork resolves to "all `render_structure` call sites,"
the DDL-echo snapshots for create-table / add-column / drop-index /
change-column auto-shows also churn — to be re-recorded with
`cargo insta`. No test is deleted to hide a regression; each change
is a deliberate format update.
### 8. Out of scope
- **OOS-1.** Live reflow-on-resize of already-rendered diagrams —
V4's re-renderable journal (requirement V4).
- **OOS-2.** Whole-database ER diagram / export (requirement V3).
- **OOS-3.** m:n relationships (requirement C4, unbuilt). Only the
existing 1:n form is rendered.
- **OOS-4.** ASCII fallback for terminals without box-drawing
(inherits ADR-0016 OOS-5).
- **OOS-5.** Per-cell colouring of column *data* values (ADR-0016
OOS-3 / NFR-5) — unrelated to relationship structure.
- **OOS-6.** Cell-level truncation with ellipsis (ADR-0016 OOS-4
stands).
- **OOS-7.** A user-configurable relationship-display setting
(always-prose / always-diagram / auto-by-width — useful for small
screens). The §1 per-command default is v1; the setting is a clean
later follow-up (user-flagged 2026-06-09).
## Consequences
- Requirement **V1** is fully satisfied: relationships render as
two-table connector diagrams in both `show relationship <name>`
(full) and `show table <T>` (stacked compact), superseding the
prose form.
- `output_render.rs` grows a relationship renderer that **composes**
the box primitives exactly as ADR-0016 anticipated; the worker→UI
show contract becomes styled-line-based, a small generalisation
reusable by other styled `show` output later.
- The worker gains an `available_width` input for `show` — the first
width-aware formatting in the codebase. The decision is a snapshot
at command time; no resize machinery is introduced.
- Compound FKs (ADR-0043) get their first dedicated visualization,
with explicit positional-pair labelling.
- The historical-log output model is preserved (no widget/reflow
model); V4 remains the home for a re-renderable journal.
- Box-drawing is required (no ASCII fallback), consistent with
ADR-0016.
## Devil's Advocate / runda pass (2026-06-09)
A planning-time DA pass (empirical, against the code) corrected three
foundational errors carried in from an initial code survey, and
surfaced one open fork:
1. **Rendering boundary was inverted** (fixed in Context 1 / §6):
`render_structure` / `render_data_table` run **App-side**
(`app.rs`), not in the worker. Rendering, width, and theme are all
App-side — simpler than the draft assumed.
2. **Width was claimed "already tracked"; it is not** (fixed in §3):
`note_output_viewport` records only rows. A new
`App::last_output_width` (set from `ui.rs`) is required.
3. **`show relationship <name>` renders prose in the worker** (fixed
in §6): `do_show_one` carries only a `RelationshipSchema`. The
detail reply must be upgraded to include both endpoint
`TableDescription`s for the full diagram.
4. **Styled-line contract was over-stated** (fixed in §5): styled
runs are already produced App-side (explain plan); no worker→UI
contract change.
**Resolved fork (user, 2026-06-09): "relationship-relevant" reach.**
Diagrams render on the surfaces where the relationship is the subject
(`show table`, `show relationship`, relationship DDL echoes);
incidental DDL echoes keep prose (§1) — avoiding a wall of diagrams
after an `add column`. A future user-configurable display setting is
noted as OOS-7. This also records V1's **deliberate scope expansion**
beyond its literal "a selected relationship" wording into
multi-relationship surfaces, to be reflected in `requirements.md`.
All design forks are now resolved and the architecture is corrected.
The user accepted this revised ADR on 2026-06-09; status is
**Accepted** and implementation proceeds test-first.
+1
View File
@@ -49,3 +49,4 @@ This directory contains the project's ADRs, recorded per
- [ADR-0041 — Copy the output panel to the system clipboard](0041-copy-output-to-clipboard.md) — **Accepted 2026-06-02 (issue #11)**, amends ADR-0003's app-command registry (adds **`copy`** / `copy all` / `copy last`). The friction it removes: filing a bug report meant terminal-selecting the output panel and fighting wrapping/borders. New **app-level command** (sigil-free, both modes): `copy` / `copy all` copy the whole panel; `copy last` copies from the most recent echo line to the end. **Mechanism — OSC 52 *and* native (`arboard`), always both**, because OSC 52 acceptance is undetectable (no terminal ack), so a true "fall back when unsupported" can't be built: emit the OSC 52 escape (no new dep — `base64`+`crossterm`; works over SSH; tmux-passthrough-wrapped via `$TMUX`), then a best-effort native write whose failure is ignored (headless host — OSC 52 carried it); the two carry identical content. **Format — plain text verbatim as rendered** (tags, `✓`/`✗`, box-drawing) joined by `\n`, without viewport padding/wrapping; a drift-lock test pins `OutputLine::plain_text` to `render_output_line`. `arboard` added **`--no-default-features`** (drops the `image` crate; X11-only on Linux — `wayland-data-control` deliberately omitted as it ~doubles the dep tree and OSC 52 covers native-Wayland). Security: write-only, scans clean for arboard's tree (cargo audit / osv-scanner / grype), 1Password-maintained, minimal surface. OOS: Markdown export, selection/range, a keybinding, OSC 52 read, `screen` passthrough - [ADR-0041 — Copy the output panel to the system clipboard](0041-copy-output-to-clipboard.md) — **Accepted 2026-06-02 (issue #11)**, amends ADR-0003's app-command registry (adds **`copy`** / `copy all` / `copy last`). The friction it removes: filing a bug report meant terminal-selecting the output panel and fighting wrapping/borders. New **app-level command** (sigil-free, both modes): `copy` / `copy all` copy the whole panel; `copy last` copies from the most recent echo line to the end. **Mechanism — OSC 52 *and* native (`arboard`), always both**, because OSC 52 acceptance is undetectable (no terminal ack), so a true "fall back when unsupported" can't be built: emit the OSC 52 escape (no new dep — `base64`+`crossterm`; works over SSH; tmux-passthrough-wrapped via `$TMUX`), then a best-effort native write whose failure is ignored (headless host — OSC 52 carried it); the two carry identical content. **Format — plain text verbatim as rendered** (tags, `✓`/`✗`, box-drawing) joined by `\n`, without viewport padding/wrapping; a drift-lock test pins `OutputLine::plain_text` to `render_output_line`. `arboard` added **`--no-default-features`** (drops the `image` crate; X11-only on Linux — `wayland-data-control` deliberately omitted as it ~doubles the dep tree and OSC 52 covers native-Wayland). Security: write-only, scans clean for arboard's tree (cargo audit / osv-scanner / grype), 1Password-maintained, minimal surface. OOS: Markdown export, selection/range, a keybinding, OSC 52 read, `screen` passthrough
- [ADR-0042 — H1a parse-error pedagogy in the grammar-tree era](0042-h1a-parse-error-pedagogy-grammar-tree.md) — **Accepted 2026-06-03.** Continues **H1a** from ADR-0021 against the ADR-0024 grammar tree (ADR-0021's chumsky mechanism is dead). Records the **baseline already shipped** — per-command `usage:` block (38 `parse.usage.*` templates), available-commands fallback, structural "after `…`, expected …" wording, source-derived ident slot labels ("table name"/"column name"), curated `parse.custom.*` near-miss messages, and the ADR-0027/0033/0036 schema-aware `[ERR]` diagnostics — so H1a is *substantially* delivered at the intent level. Defines the remaining work as **(1)** a verified per-command **near-miss matrix** (`tests/typing_surface/` + `tests/it/parse_error_pedagogy.rs`) as the definition of done, test-first; **(2)** **friendlier literal expectation labels** — optional prose glosses on `Word`/`Punct`/`Flag` positions that *add* role context while always keeping the exact literal visible (e.g. "a filter clause: `where …` or `--all-rows`"); **(3)** **advanced-mode SQL** near-miss parity (RETURNING scope, CTE-arity positioning, `CROSS JOIN … ON`, INSERT…SELECT count) — **in scope**, kept distinct from ADR-0019 §OOS-2 which covers advanced-SQL *engine*-error sanitisation, a different layer. Catalog/anchor-phrase discipline (ADR-0019) preserved; no public API change. OOS: I3/I4, spell-correction, multi-error reporting, verbosity-gating the usage block - [ADR-0042 — H1a parse-error pedagogy in the grammar-tree era](0042-h1a-parse-error-pedagogy-grammar-tree.md) — **Accepted 2026-06-03.** Continues **H1a** from ADR-0021 against the ADR-0024 grammar tree (ADR-0021's chumsky mechanism is dead). Records the **baseline already shipped** — per-command `usage:` block (38 `parse.usage.*` templates), available-commands fallback, structural "after `…`, expected …" wording, source-derived ident slot labels ("table name"/"column name"), curated `parse.custom.*` near-miss messages, and the ADR-0027/0033/0036 schema-aware `[ERR]` diagnostics — so H1a is *substantially* delivered at the intent level. Defines the remaining work as **(1)** a verified per-command **near-miss matrix** (`tests/typing_surface/` + `tests/it/parse_error_pedagogy.rs`) as the definition of done, test-first; **(2)** **friendlier literal expectation labels** — optional prose glosses on `Word`/`Punct`/`Flag` positions that *add* role context while always keeping the exact literal visible (e.g. "a filter clause: `where …` or `--all-rows`"); **(3)** **advanced-mode SQL** near-miss parity (RETURNING scope, CTE-arity positioning, `CROSS JOIN … ON`, INSERT…SELECT count) — **in scope**, kept distinct from ADR-0019 §OOS-2 which covers advanced-SQL *engine*-error sanitisation, a different layer. Catalog/anchor-phrase discipline (ADR-0019) preserved; no public API change. OOS: I3/I4, spell-correction, multi-error reporting, verbosity-gating the usage block
- [ADR-0043 — Compound-primary-key foreign-key references (T3)](0043-compound-pk-foreign-key-references.md) — **Accepted + implemented 2026-06-09** (all four forks confirmed at the recommended option: full-PK matching, house-style uniform lists, parenthesized DSL syntax, bare-SQL-FK auto-expansion). Closes `requirements.md` **T3** `[x]` — the relationship model went list-based across six layers (single-column preserved, no migration), DSL `from P.(a,b) to C.(x,y)` + SQL `FOREIGN KEY (a,b) REFERENCES P(x,y)` parse/execute/enforce, 12 tests in `tests/it/compound_fk.rs`. Closes the open leg of `requirements.md` **T3**: a foreign key that *references* a parent's compound primary key. A 2026-06-09 audit found single-column FK woven through ~1520 sites (metadata table, `RelationshipSchema`, `project.yaml` `RawEndpoint`, both grammar surfaces, executor FK-DDL emission, per-column type-compat, display) — earns an ADR, not an inline build. **Decision:** reference the parent's **full** compound PK, matched **positionally** to an equal-length child column list, per-pair `fk_target_type` compat (ADR-0011, element-wise); DSL `from <P>.(a, b) to <C>.(x, y)` (single form unchanged), SQL `FOREIGN KEY (x, y) REFERENCES P(a, b)` (extend the existing one-cap lists; bare table-level FK auto-expands to the parent PK when arities match). **Storage — no migration (back-compat not required, user-confirmed 2026-06-09; no installed base):** the relationship endpoint joins the list convention `project.yaml` *already* uses — `columns: [a, b]` like `primary_key: [id]` and index `columns: [...]` (the endpoint was the lone scalar `column:` holdout); the metadata `TEXT` columns are unchanged and store the list **comma-joined** (`a,b`; the bare name for single — safe because identifiers are `[A-Za-z0-9_]+`). No F3 migrator, no version bump; accepted trade-off is that a pre-change `project.yaml` with relationships won't load (clean cutover). In-memory model goes list-based (`Vec<String>`) through all six layers; the enforced FK is the rebuilt child-table DDL (`FOREIGN KEY (a,b) REFERENCES P(x,y)`), one relationship = one undo step (ADR-0013). Genuine forks escalated: matching policy (full-PK vs subset), storage (house-style uniform lists vs normalized table), DSL syntax (parenthesized vs repeated-dotted), bare-SQL-FK auto-expansion. OOS: subset/non-PK (UNIQUE-targeted) FK references; any single-column behaviour change - [ADR-0043 — Compound-primary-key foreign-key references (T3)](0043-compound-pk-foreign-key-references.md) — **Accepted + implemented 2026-06-09** (all four forks confirmed at the recommended option: full-PK matching, house-style uniform lists, parenthesized DSL syntax, bare-SQL-FK auto-expansion). Closes `requirements.md` **T3** `[x]` — the relationship model went list-based across six layers (single-column preserved, no migration), DSL `from P.(a,b) to C.(x,y)` + SQL `FOREIGN KEY (a,b) REFERENCES P(x,y)` parse/execute/enforce, 12 tests in `tests/it/compound_fk.rs`. Closes the open leg of `requirements.md` **T3**: a foreign key that *references* a parent's compound primary key. A 2026-06-09 audit found single-column FK woven through ~1520 sites (metadata table, `RelationshipSchema`, `project.yaml` `RawEndpoint`, both grammar surfaces, executor FK-DDL emission, per-column type-compat, display) — earns an ADR, not an inline build. **Decision:** reference the parent's **full** compound PK, matched **positionally** to an equal-length child column list, per-pair `fk_target_type` compat (ADR-0011, element-wise); DSL `from <P>.(a, b) to <C>.(x, y)` (single form unchanged), SQL `FOREIGN KEY (x, y) REFERENCES P(a, b)` (extend the existing one-cap lists; bare table-level FK auto-expands to the parent PK when arities match). **Storage — no migration (back-compat not required, user-confirmed 2026-06-09; no installed base):** the relationship endpoint joins the list convention `project.yaml` *already* uses — `columns: [a, b]` like `primary_key: [id]` and index `columns: [...]` (the endpoint was the lone scalar `column:` holdout); the metadata `TEXT` columns are unchanged and store the list **comma-joined** (`a,b`; the bare name for single — safe because identifiers are `[A-Za-z0-9_]+`). No F3 migrator, no version bump; accepted trade-off is that a pre-change `project.yaml` with relationships won't load (clean cutover). In-memory model goes list-based (`Vec<String>`) through all six layers; the enforced FK is the rebuilt child-table DDL (`FOREIGN KEY (a,b) REFERENCES P(x,y)`), one relationship = one undo step (ADR-0013). Genuine forks escalated: matching policy (full-PK vs subset), storage (house-style uniform lists vs normalized table), DSL syntax (parenthesized vs repeated-dotted), bare-SQL-FK auto-expansion. OOS: subset/non-PK (UNIQUE-targeted) FK references; any single-column behaviour change
- [ADR-0044 — Relationship visualization (two-table connector diagrams)](0044-relationship-visualization.md) — **Accepted 2026-06-09** (revised post-`/runda` DA pass). Resolves **ADR-0016 OOS-1** and closes the open half of `requirements.md` **V1** ("a selected relationship as two tables joined by a line"). Renders a relationship as **Style A** (two structure boxes + connector). **Reach = "relationship-relevant"** (user-chosen over global / show-only): diagrams on the surfaces where the relationship is the *subject*`show relationship <name>` (one full diagram), `show table <T>` (T's structure box then a **Relationships** section of **stacked compact** per-relationship diagrams — chosen over a focal-centred subgraph: no crossing lines, scales via scroll, two-boxes-wide fits any terminal), and relationship DDL echoes (`add`/`drop`/`modify relationship`); incidental DDL echoes (`add column`, `drop index`, `change column`, plain `create table`) keep the terse prose, via a `Diagram`|`Prose` render mode on `render_structure`. Reading convention **child(FK)-left / parent-right, arrow →, `n`…`1` cardinality**, applied uniformly; every box gets a **bold title row + rule** so the name can't read as a column. **Compound FKs** (ADR-0043) route one connector per positional pair + an explicit pairing line. **Width-aware** (first in the codebase) but **App-side**: `render_structure`/diagram rendering runs in `app.rs` (the worker only returns `TableDescription`s), a new `App::last_output_width` (set from `ui.rs`) drives side-by-side vs a **vertical-stack** fallback + last-resort "run `show relationship`" pointer; rendered once at command time, **no live reflow** (V4). `show relationship`'s worker path (`do_show_one`, prose-only) is restructured to return both endpoint `TableDescription`s. Styling reuses **ADR-0028** App-side styled runs (new classes: table-name/key/connector/cardinality/action) — no worker→UI contract change. **Partially supersedes ADR-0016 §5** (prose block replaced on relationship-subject surfaces, retained on incidental ones); extends §4 (layout width-awareness, still no cell truncation) and §6 (per-span theming). Tests: insta snapshots (single, compound, vertical fallback, helper line, self-referential, multi-rel `show table`) + width-threshold/routing unit tests + Tier-3 wiring; enumerated prose-fallout updates (`output_render.rs:121/135/793`, the relationships snapshot, `walking_skeleton.rs:477/530`). A `/runda` DA pass corrected three inverted-architecture claims (App-side rendering, untracked width, prose-in-worker show-relationship) before acceptance. OOS: user-configurable display setting (OOS-7), live reflow (V4), whole-DB ER export (V3), m:n (C4), ASCII fallback (ADR-0016 OOS-5)