Release v0.1.29 - #15
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ac7f67c190
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| let is_non_usage_error = self.generation == 2 | ||
| && self.has_error_object | ||
| && self.input.is_none() | ||
| && self.output.is_none() | ||
| && self.reasoning.is_none() | ||
| && self.cache_read.is_none() | ||
| && self.cache_write.is_none() | ||
| && self.cost_usd.is_none(); |
There was a problem hiding this comment.
When an in-flight OpenCode V2 assistant row already has a structured error object but time.completed is still absent, this predicate classifies it as a non-usage error and removes it before the adapter's existing completed_at_ms.is_none() live-write deferral can run. In that state the collector reconciles the session counters as ready recovery instead of DeferredLiveWrite, so a refresh can establish a partial baseline from an actively written row; require a completed timestamp before taking this error-recovery path, or pass the row through the normal deferred logic.
Useful? React with 👍 / 👎.
| for conversation in conversations | ||
| .iter() | ||
| .filter(|conversation| conversation.variant == AntigravityProductVariant::Cli) | ||
| .filter(|conversation| is_sqlite_artifact(&conversation.path)) |
There was a problem hiding this comment.
When a CLI conversation directory contains both session.db and a newer legacy session.pb, ConversationIndex keeps only the newest artifact for that stem, so this new post-index filter drops the selected .pb and the older valid .db is never opened. Full collection can then report a complete baseline while omitting that conversation's SQLite usage; filter or prefer artifacts before the stem-level dedupe, or have the index retain the .db candidate for SQLite collection.
Useful? React with 👍 / 👎.
1 participant
Summary
Verification