implements PLN-0132
Intrepid/Prometeu/Studio/pipeline/pr-master Build started...
JaCoCo Coverage #### Project Overview No changes detected, that affect the code coverage. * Line Coverage: 62.51% (18192/29102) * Branch Coverage: 53.15% (7093/13345) * Lines of Code: 29102 * Cyclomatic Complexity: 11715 #### Quality Gates Summary Output truncated.
Test / Build skipped: 15, passed: 663

Add PBS find-references as the request-local inverse of definition
identity on FrontendLanguageService. Usage sites are identifier
tokens whose definition locations intersect the cursor set.
Empty results stay honest when definition is empty, including
virtual stdlib paths. Honor includeDeclaration in the LSP bridge.

Housekeep DSC-0040 with LSN-0069.
This commit is contained in:
2026-09-21 08:54:11 +01:00
parent 4b72b2a0dd
commit 9458312d73
21 changed files with 729 additions and 68 deletions
@@ -1,62 +0,0 @@
---
id: AGD-0043
ticket: pbs-lsp-find-references
title: PBS LSP Find References
status: open
created: 2026-07-15
resolved:
decision:
tags: [studio, lsp, vscode, compiler-pbs, editor, references]
---
## Pain
Domain owner: `studio/lsp`
PBS users cannot ask where a symbol is used, so changes to functions, services, structs, constants, and stdlib-facing APIs require manual search that confuses overloads and unrelated text matches.
## Context
Completion and hover already resolve symbols in local editor context, but the LSP does not expose references and the compiler/editorial layer does not yet provide project-wide usage locations keyed by semantic identity.
## Open Questions
- [ ] Which declarations need reference tracking first: local functions, methods, types, constants, imports, stdlib symbols, or all of them?
- [ ] How should overloads and same-name symbols in different scopes be disambiguated?
- [ ] Should references include declarations, write/read classification, or only usage sites in the first wave?
## Options
### Option A - Text search with filters
- **Approach:** Search workspace text for the selected identifier and filter obvious false positives by token kind.
- **Pro:** Simple and fast for early demos.
- **Con:** Incorrect for overloads, scope, shadowing, fields, methods, imports, and generated/stdlib surfaces.
- **Maintainability:** Poor; later semantic references would duplicate and replace it.
### Option B - Semantic usage index
- **Approach:** During semantic analysis, emit definition and usage sites tied to stable symbol identities and expose `textDocument/references`.
- **Pro:** Correct foundation for rename, call hierarchy, diagnostics UX, and code actions.
- **Con:** Requires compiler-facing usage modeling and careful invalidation.
- **Maintainability:** Strong; one source of truth for all symbol usage features.
## Tradeoffs
Reference accuracy matters more than early breadth. False positives are worse than missing unsupported categories because users will trust rename and references for safe edits.
## Recommendation
Prefer Option B, but scope the first wave to project-owned declarations whose identity is already resolved reliably.
## Discussion
References should follow go-to-definition because both need the same identity and location model.
## Resolution
Ainda em aberto.
## Next Step
Define first-wave symbol categories and whether declarations are included in reference results.