Skip to content

observer: forget partial write, and wrong messages from synapse_forget_rule and synapse_record_learning #212

Closed

Description

@Alter-Igor

Bugs from the QA and architect passes on PR #210 (#38). They are listed separately so they are not lost inside the owner decision on #211.

  1. Partial write on forget. forgetLearning reads and checks every store first (fix(opencode): a refused forget changes no rule store; document mirrored project rules (#38) #210). But if the global write succeeds and the project write then fails, the call returns 0 while the global copy is already tombstoned (marked deleted). On Windows a rename can fail with EPERM or EBUSY when another process holds the file. Fix ideas: roll back the first write, or return a "partial" result the tool can report.
  2. Wrong message on a refused forget. synapse_forget_rule says "No rule matched" when forgetLearning returned 0 because a store could not be read. It should say the stores could not be checked and nothing was changed.
  3. False "recorded" message. If no store can be written (all unreadable or all writes fail), recordLearning still returns the entry, and synapse_record_learning says "Successfully recorded" though nothing was saved.

Activity

  1. Alter-Igor commented on Oct 17, 2026

    @Alter-Igor
    OwnerAuthor

    Update from PR #210:

    Items 1 and 2 are still open.

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

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions