Skip to content

Harden indexing lifecycle, freshness, and T100 extraction#12

Draft
jordannn02 wants to merge 2 commits into
win4r:mainfrom
jordannn02:codex/codebase-memory-p0-clean
Draft

Harden indexing lifecycle, freshness, and T100 extraction#12
jordannn02 wants to merge 2 commits into
win4r:mainfrom
jordannn02:codex/codebase-memory-p0-clean

Conversation

@jordannn02

Copy link
Copy Markdown

Summary

  • add a source-owned executable capability contract exposed through --version, --manifest, and MCP tools/list
  • make repository indexing observable, cancellable, and backgrounded by default with job IDs and progress state
  • persist source/extractor freshness metadata and expose ready, stale_source, stale_extractor, missing_root, and partial states
  • add dry-run-only index pruning and preserve corrupt index evidence instead of deleting it implicitly
  • add T100/TIPTOP 4GL-family discovery and extraction support with regression coverage

Why

The executable, plugin metadata, and actual MCP tool surface could diverge without a runtime-owned contract. Long-running indexing also lacked a durable job lifecycle, and an index could appear ready even after its source root or extractor changed. T100/TIPTOP .4gl, .per, .42m, and .sch sources additionally needed explicit language and extraction support.

Impact

  • callers can verify the exact executable version, commit, build ID, checksum, tool list, and UI capability
  • indexing can be polled and cancelled without terminating the MCP server
  • stale or incomplete indexes are distinguishable from usable indexes
  • cleanup decisions are previewable and non-destructive
  • T100/TIPTOP source tracing gains functions, calls, table reads/writes, and language discovery coverage

Validation

  • fresh ASan/UBSan full test suite: 5638 passed, with no sanitizer markers
  • fresh production build with -Wall -Wextra -Werror
  • manifest assertions: schema 1, 17 tools, ui_assets=false, cancel_index and prune_index present, repo_map absent
  • source and installed plugin contract smoke tests passed against the sealed executable
  • git diff --check passed and the published worktree was clean

Commits

  • 4db7262 - implementation and regression coverage
  • 6559c16 - clean-build provenance

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant