Keep repeated backlinks in source order

[srid]

Repeated backlinks now render in the order they appear in Markdown. A note that links to the same target several times now carries source occurrence order through relation extraction and backlink grouping, so the rendered context list matches the original note. Closes #186.

The fix makes the ordering contract explicit instead of relying on IxSet iteration or derived record ordering:

• extract link occurrences in document order using the existing Pandoc inline-ref parser
• store each relation occurrence with a RelOrder
• sort backlink contexts by a named source-order key before grouping them for rendering
• cover both noteRels and modelLookupBacklinks with focused regressions

Validation

nix develop -c cabal test test --test-options='--match /Emanote.Model.Graph/ --match /Emanote.Model.Link.Rel/noteRels/'

Try it locally

nix run github:srid/emanote/fix/backlink-context-order

Generated by /do on Codex (model gpt-5).

[srid]

Hickey/Lowy Analysis

#	Lens	Finding	Disposition
1	Hickey	relOrder hides sort contract	Fixed in this PR
2	Hickey	Link occurrence lacks representation	Fixed in this PR
3	Lowy	Backlink order rides Ord	Fixed in this PR

Hickey rationale

The first implementation preserved source order by putting _relOrder before _relTo in Rel, which made the behavior depend on derived Ord and record field position. That was split from the ordering contract reviewers need to understand. The follow-up commit introduced RelOrder, an explicit Ord Rel, and a named relSourceOrder key.

The extraction path also carried a raw (url, attrs, context) tuple through multiple helpers. That tuple represented one domain concept, so the follow-up commit replaced it with LinkOccurrence and reused Emanote.Pandoc.Link.parseInlineRef for Pandoc link parsing.

Lowy rationale

Backlink ordering is user-visible behavior, so the boundary that assembles backlinks should own the ordering decision. The original patch made this behavior ride on low-level Rel representation and IxSet iteration. The follow-up change sorts repeated backlink rels by Rel.relSourceOrder before grouping, and adds a regression at modelLookupBacklinks so the tested boundary matches the rendered behavior.

[srid]

Evidence

Captured from http://127.0.0.1:21720/foo using a scratch notebook where source.md links to foo in this order: b, a, c.

Verified the Links to this page section renders backlink contexts in the same source order: b, a, c.

[srid]

/do results

Step	Status	Duration	Verification
sync	✓	1s	git fetch ok; forge=github; noGit=false
research	✓	5m 39s	Issue #186 maps to Emanote.Model.Link.Rel.noteRels and Emanote.Model.Graph.modelLookupBacklinks; rel extraction/storage lacks source occurrence order, causing repeated backlink contexts to render out of Markdown order.
branch	✓	0s	Created feature branch fix/backlink-context-order from origin/master.
implement	✓	4m 3s	Implemented ordered rel extraction with per-note occurrence indices, added regression for repeated [[Foo]] backlink contexts, removed unused pandoc-link-context dependency, and updated CHANGELOG.
check	✓	19s	nix develop -c cabal build all passed (bare cabal not on PATH, same configured command run inside project dev shell).
docs	✓	1s	CHANGELOG.md updated under Unreleased bug fixes; README.md and docs/ need no changes for this internal backlink-order fix.
fmt	✓	18s	just fmt passed after fourmolu formatted the changed Haskell files.
commit	✓	8s	Committed 2e77aff and pushed origin/fix/backlink-context-order.
hickey+lowy	✓	9m 55s	Hickey and Lowy reviews completed. Fixed Lowy/Hickey ordering-contract finding in 61fb989 and Hickey link-occurrence representation finding in ed0412d; no deferred findings.
police	✓	1m 24s	code-police clean after fixing readability doc comments in 8309c54; rules, fact-check, and elegance passes found no remaining issues.
test	✓	8s	Focused Haskell tests passed: Emanote.Model.Graph modelLookupBacklinks and Emanote.Model.Link.Rel noteRels regressions for issue #186.
create-pr	✓	50s	Draft PR #688 created and Hickey/Lowy analysis comment posted; no deferred findings.
ci	✓	4m 4s	vira ci passed for HEAD 8309c54; signoffs vira/aarch64-darwin and vira/x86_64-linux succeeded.
evidence	✓	1m 48s	Posted ## Evidence comment with screenshot asset: #688 (comment)
done	✓	0s	All /do steps passed; final status report prepared.
Total		31m 40s

Slowest step: hickey+lowy (9m 55s)

Optimization suggestions

• hickey+lowy dominated the run at 9m 55s. For tiny, low-risk edits, $do --minimal would skip those reviews; for model/rendering behavior like this one, keeping the default was worthwhile because it produced the explicit ordering contract commits.
• research took 5m 39s. A prompt that names suspected modules such as Emanote.Model.Link.Rel and Emanote.Model.Graph, or includes a minimal note pair, would trim the initial mapping time.
• ci took 4m 4s and passed on the first run. For future follow-up-only validation, resuming from the CI step after the branch is already pushed avoids repeating earlier workflow work.

Workflow completed at 2026-04-30T12:08:29Z.

[srid]

#692