archive_read_only Should Create Bare Git Repositories

ID: 6af6

Problem

When archive_read_only = true is set with archive_all = true, the relay accepts repository announcements but does NOT create bare git repositories or sync git data. This is a bug.

Current behavior:

src/nostr/builder.rs:179 explicitly skips bare repository creation for AcceptArchive announcements
Comment says: "Don't create bare repository for archived announcements"
Result: /persistent/grasp/full-archive/git/ directory is empty despite accepting announcements

Expected behavior:

archive_read_only = true should:
- ✅ Accept repository announcements (working)
- ✅ Create bare git repositories (BROKEN)
- ✅ Sync git data from remote clone URLs (BROKEN)
- ❌ Reject git pushes (working - read-only)

Impact:

Archive relays cannot actually archive git repositories
Only Nostr events are stored, not the actual git data
Defeats the purpose of GRASP-05 archive mode

Plan

Phase 1: Fix bare repository creation for archive mode
- Remove the skip logic in src/nostr/builder.rs:179
- Ensure bare repos are created for AcceptArchive announcements
- Verify sync works for archived repos
Phase 2: Add test coverage
- Create integration test for archive_read_only mode
- Verify bare repos are created
- Verify git data is synced
- Verify pushes are rejected
Phase 3: Update documentation
- Clarify archive_read_only behavior in docs
- Add examples of proper archive configuration

Progress

2026-01-19 [Session 13:20]

Identified bug: archive_read_only skips bare repo creation
Root cause: Line 179 in src/nostr/builder.rs explicitly skips creation
User reported empty /persistent/grasp/full-archive/git/ directory
Created issue to track fix

2026-01-19 [Session 14:45]

Started work: Created worktree 6af6-archive-read-only-should-create-bare-repos
Ready to begin test creation and implementation

2026-01-19 [Session 15:30]

FIXED: Modified src/nostr/builder.rs:169-208 to create bare repos for AcceptArchive
FIXED: Modified src/config.rs:564-575 to respect custom git_data_path in memory backend
COMPLETED: Test tests/archive_read_only.rs now passes bare repo creation check
REMAINING: Git sync not working yet - repos created but data not synced from clone URLs
Decision: Bare repo creation bug is FIXED. Git sync is a separate issue that needs investigation
All unit tests pass (369 passed)
Changes:
- AcceptArchive now calls ensure_bare_repository() just like Accept does
- Memory backend respects NGIT_GIT_DATA_PATH when explicitly set (fixes test isolation)
- Added test coverage for memory backend path behavior

2026-01-19 [Session 16:00]

ROOT CAUSE FOUND: Git sync not triggered for archive announcements
- Purgatory sync requires events in purgatory to trigger (state/PR events)
- Archive announcements have NO purgatory events initially
- has_pending_events() returns false → sync skipped
- collect_needed_oids() returns empty → sync skipped
SOLUTION: Added direct git fetch for archive announcements
- Created AnnouncementPolicy::sync_archive_git_data() method
- Spawns async task to fetch all refs from clone URLs
- Uses git fetch +refs/heads/*:refs/heads/* +refs/tags/*:refs/tags/*
- Tries each clone URL until one succeeds
COMPLETED: All tests pass (369 unit + all integration tests)
- test_archive_read_only_creates_bare_repo now fully passes
- Bare repo created ✅
- Git data synced ✅
- Pushes rejected ✅
Changes:
- src/nostr/policy/announcement.rs:111-205: Added sync_archive_git_data() method
- src/nostr/builder.rs:193-216: Spawn sync task for AcceptArchive announcements
- Archive mode now works end-to-end: accepts announcements, creates repos, syncs git data

2026-01-19 [Session 17:00]

COMPLETED: Issue fully resolved and documented
Documentation Updated: docs/reference/configuration.md
- Added clarification that archive_read_only DOES create bare repos
- Added clarification that git data IS synced from clone URLs
- Added "Git Sync Behavior" section with technical details
- Updated use case examples to show expected results
Test Coverage: Comprehensive integration test added
- tests/archive_read_only.rs: Tests bare repo creation, git sync, and push rejection
- All 369 unit tests pass
- All integration tests pass
Summary of Fix:
- Bug: archive_read_only was accepting announcements but NOT creating repos or syncing git data
- Root cause: Explicit skip logic in builder.rs + no sync trigger for archive announcements
- Solution: Remove skip logic + add direct git fetch for archive announcements
- Result: Archive mode now fully functional - accepts, creates, syncs, and enforces read-only
All phases complete: Ready to commit and close issue

2026-01-19 [Session 18:30] - ARCHITECTURE RETHINK

PROBLEM IDENTIFIED: Previous sync_archive_git_data() approach was wrong
- Direct git fetch from clone URLs trusts git servers blindly
- No validation against Nostr events (state events)
- Bypasses the security model of purgatory sync
- "Naughty git servers" could provide incorrect state
CORRECT ARCHITECTURE: Use existing purgatory sync infrastructure
- Archive relay syncs from bootstrap relay (relay-to-relay sync)
- State events are synced via negentropy/REQ
- State events go to purgatory (waiting for git data)
- Purgatory sync fetches git data from clone URLs
- Git data is validated against Nostr state events
CHANGES MADE:
- Removed sync_archive_git_data() method from src/nostr/policy/announcement.rs
- Removed spawn of custom sync in src/nostr/builder.rs
- Updated test to use relay-to-relay sync (TestRelay with bootstrap_relay_url)
- Made TestRelay::start_with_archive_and_sync() public for tests
TEST UPDATES:
- test_archive_read_only_creates_bare_repo: Now uses source relay + archive relay
- test_archive_without_state_events_does_not_sync_git: Verifies security model
- All 369 unit tests pass
- All archive integration tests pass
SECURITY MODEL:
- Archive mode only syncs git data when there are state events to validate against
- This protects against "naughty git servers" providing incorrect state
- Same security guarantees as normal relay operation
LIMITATION DOCUMENTED:
- Archive mode requires a bootstrap relay to sync state events
- Without state events, git data is NOT synced (by design - security)

2026-01-19 [Session 19:00] - READY FOR MERGE

SQUASHED COMMITS: All 4 commits squashed into single comprehensive commit
- Commit: f191261 "fix: archive_read_only creates bare repos and syncs git data (GRASP-05)"
- Comprehensive commit message includes problem, root cause, solution, security model, test coverage
ALL TESTS PASS: 37 tests passed (including both archive integration tests)
- test_archive_read_only_creates_bare_repo: End-to-end sync flow verified
- test_archive_without_state_events_does_not_sync_git: Security model verified
READY FOR REVIEW AND MERGE:
- Single clean commit for easy review
- Comprehensive test coverage
- Documentation updated
- Security model validated
- All phases complete

2026-01-19 [Session 20:00] - FINAL REVIEW AND MERGE

BUILD AGENT REVIEW: Reviewed test implementation for quality and isolation
- Test isolation: ✅ Uses random temp directories and ports - no conflicts possible
- Test complexity: ✅ Bare repo check is necessary and justified - validates the actual fix
- Recommendation: Keep current implementation - well-designed and ready to merge
ISSUE COMPLETE: All work committed, tests pass, ready to merge to main
- Commit bf70d72: "fix: archive_read_only creates bare repos for archived announcements"
- Test coverage comprehensive and properly isolated
- No changes needed - merging to main

Notes

Related code: src/nostr/builder.rs:169-195 (AcceptArchive handling)
Related code: src/nostr/events.rs:434-437 (GRASP-05 validation)
User configuration: NixOS with archiveAll = true and archiveReadOnly = true
Test should verify both event storage AND git repository creation

8.2 KiB Raw Blame History