53 lines
2.7 KiB
Markdown
53 lines
2.7 KiB
Markdown
# ADR 0048: Modular Editor Composition and Architecture-Debt Ratchet
|
|
|
|
## Status
|
|
|
|
Accepted
|
|
|
|
## Context
|
|
|
|
The Content Browser and Inspector accumulated unrelated navigation, rendering, import, material,
|
|
transaction, and component-card behavior in files thousands of lines long. That made small changes
|
|
high-risk, encouraged duplicate UI, and obscured ownership. ADR 0012 states a zero-debt intent but
|
|
did not provide an enforceable point-of-change budget.
|
|
|
|
## Decision
|
|
|
|
Editor panels are composed from thin shells, domain modules, and UI-independent services. The
|
|
Inspector delegates every visible built-in component through `EditorComponentRegistry`; a
|
|
hard-coded type dispatcher is not a second authority. Material slot presentation uses a shared
|
|
view-model/action contract, and Content Browser and Material Library cards use one bounded asset
|
|
card/status model.
|
|
|
|
The machine-readable policy is `.codex/architecture.toml`. Its audit enforces 500 nonblank lines
|
|
for UI shells, 800 for other UI modules, and 1,200 for other production Rust modules. Existing
|
|
over-limit modules have frozen ceilings. A temporary exception requires a tracker issue, rationale,
|
|
hard maximum, extraction target, and expiry milestone. Selective verification and candidate CI run
|
|
the audit.
|
|
|
|
The M2 extraction is tracked by Gitea #68. The Content Browser and Inspector shells are now below
|
|
the 500-line budget, their domain behavior lives in bounded submodules, and no temporary #68
|
|
exception is active. Closing the issue still requires the selective and native acceptance evidence
|
|
defined by the milestone.
|
|
|
|
The Penpot-led material-inspector refinement remains inside #68's shared-UI scope. The reusable
|
|
visual primitives live in `ui/design_system`, while `ui/materials` owns the action-returning domain
|
|
panel. This keeps authored visual policy separate from Inspector world access and prevents the
|
|
first design-system migration from rebuilding a panel monolith. ADR 0049 owns the visual contract.
|
|
|
|
## Consequences
|
|
|
|
- New behavior is added through a sustainable extension seam instead of enlarging a monolith.
|
|
- Shrinking an over-limit module is always accepted; regrowth above its frozen baseline fails.
|
|
- Passing the audit is a no-regression statement; issue closure additionally requires functional
|
|
and native evidence for the extracted seams.
|
|
- Architecture work remains visible in the same tracker and milestone as the feature that exposes
|
|
it.
|
|
- ADR 0034 owns registry-driven component semantics; this ADR owns editor composition and the
|
|
mechanical debt guard.
|
|
|
|
Related decisions: [ADR 0012](0012-zero-tech-debt-editor.md),
|
|
[ADR 0034](0034-registry-driven-authoring-components.md), and
|
|
[ADR 0047](0047-editor-authored-asset-documents.md), and
|
|
[ADR 0049](0049-penpot-led-editor-visual-system.md).
|