Add Google Voice Takeout import support

[sternryan]

Summary

• New sync-gvoice command that imports SMS, MMS, and call records from Google Takeout Voice exports
• Follows the established adapter pattern, implementing gmail.API interface for plug-and-play integration
• Deterministic message IDs via SHA-256 for idempotent re-imports

New files

• internal/gvoice/client.go - gmail.API implementation over Takeout HTML
• internal/gvoice/parser.go - HTML conversation parser
• internal/gvoice/models.go - conversation and message types
• internal/gvoice/parser_test.go - parser tests
• cmd/msgvault/cmd/sync_gvoice.go - CLI command

Usage

msgvault sync-gvoice --takeout-dir ~/path/to/Takeout/Voice

Performance

• Indexes ~120k entries from ~50k files in ~6 seconds
• Full import at ~1,500 messages/sec

[roborev-ci]

roborev: Combined Review (cb5595d)

Verdict: The PR successfully introduces the Google Voice Takeout parser and sync pipeline, but requires crucial fixes for potential panics, data
races, and an O(N²) performance bottleneck before merging.

High Severity

• Location: internal/gvoice/client.go:153 (in buildIndex)
  **
  Problem:** The error path dereferences entry.ID after c.indexCallFile(...) returns an error. In that case, entry is nil, so a single malformed or unsupported call HTML file will panic the entire import instead of being skipped.
  Fix: Log the filename/path
  already in scope instead of entry.ID, or guard the dereference before logging.

• Location: internal/gvoice/client.go (in GetMessagesRawBatch)

Problem: Continuing after a GetMessageRaw error leaves nil pointers in the pre-allocated results slice. If callers iterate over the returned slice and assume all items are valid, it will trigger a panic.
Fix: Initialize results with zero length (make([]*gmail .RawMessage, 0, len(messageIDs))) and append messages on success.

Medium Severity

• Location: internal/gvoice/client.go (in
  getCachedMessages and buildIndex)
  Problem: The file cache (lastFilePath, lastMessages) and the lazy index initialization (c.indexBuilt) lack mutex protection. If the syncer fetches message batches concurrently, this will cause a data race.
  Fix: Protect the cache
  fields with a sync.Mutex and use sync.Once for buildIndex.

• Location: internal/gvoice/client.go (in GetMessageRaw)

  Problem: A linear scan over c.index is performed for every message fetched. Doing this for a large number of messages yields O(N²) complexity, which will cause severe CPU stalling during large syncs.
  Fix: Populate a map[string]*indexEntry during build Index() to allow O(1) lookups by ID.

• Location: internal/gvoice/client.go:219 and internal/gvoice/client.go: 425
  Problem: For 1:1 conversations, the code derives the other participant only by scanning for a non-Me message. If a Takeout thread contains only outbound messages
  , indexTextFile falls back to the owner's own number for the thread ID and buildTextMessage emits no recipient at all. That can merge unrelated sent-only threads together and lose addressing metadata.
  Fix: Use the conversation/file metadata (contactName, filename, or parsed conversation header) as
  the fallback participant when no inbound message exists, and add a test for sent-only threads.

• Location: internal/gvoice/client.go:449 and internal/g voice/client.go:461
  Problem: MMS attachments are detected during HTML parsing, but buildTextMessage only adds the mms label and generates a plain-text MIME body with no
  attachment parts. The import drops all image/video content from MMS messages.
  Fix: Build multipart MIME that includes the referenced media files (or persist them through the attachment pipeline) and cover this with tests for image/video MMS messages.

---

Synthesized from 3 reviews (agents: codex, gemini | types: default, security)