Skip to content

Fix review process and improve ontology materialization - #340

Merged
vinceblock99 merged 7 commits into
fix/triage-reviewfrom
fix/ancestral-bug
Nov 15, 2026
Merged

vinceblock99 merged 7 commits into
fix/triage-reviewfrom
fix/ancestral-bug

Conversation

@leefaus

@leefaus leefaus commented Nov 15, 2026

Copy link
Copy Markdown
Contributor

This pull request adds comprehensive support for retrospective publication reviews in the Atomic CLI and core libraries. It introduces a new publication_review module with detailed data structures and parsing logic, updates the intent graph and triage commands to recognize and handle review context and evidence, and improves the user interface and output to clearly distinguish between implementation and review evidence. Additionally, it enhances the RPC and CLI layers to support review preview validation and error handling, and updates test coverage and documentation accordingly.

Publication Review Support:

  • Added the new publication_review module in atomic-canonical, defining types and validation logic for publication reviews, including schema versioning, review subject structure, and artifact hashing. [1] [2]
  • Integrated review file input into the IntentGraph CLI command, allowing users to include a prepared review as context when generating intent graphs. [1] [2]
  • Added a new subcommand to the triage CLI for preparing, signing, and inspecting publication reviews. [1] [2]

Intent Graph and Evidence Attribution:

  • Updated intent graph rendering and annotation to distinguish between reviewed changes and implementation, including new summaries and section headings for reviewed evidence. [1] [2]
  • Added support for new review-related node types and colors in graph visualizations.
  • Extended test coverage for review context and evidence attribution in the intent graph. [1] [2]

RPC and Service Integration:

  • Modified RPC calls to accept and validate review documents, checking that the returned review preview matches the requested artifact and providing clear error messages if not. [1] [2] [3] [4]
  • Enabled the review-shacl feature for atomic-repository in the CLI's dependencies.

Triage Output and User Interface:

  • Improved triage report output to clearly indicate when an intent is set aside (icebox), showing historical criteria and reviewer information distinct from current implementation approval. [1] [2]

Documentation and Vocabulary:

  • Clarified documentation for intent status values, especially the meaning of the icebox (set aside) status.

@leefaus
leefaus changed the base branch from dev to fix/triage-review November 15, 2026 16:15
@leefaus
leefaus added this pull request to stack #341 November 15, 2026 16:16
leefaus added a commit that referenced this pull request Nov 15, 2026
Merge the published PR base without rewriting either branch. Preserve the review boundary, span capability fence and prototype feature guards. The only source-tree change from the prior PR head is the capability import ordering required by cargo fmt.
@vinceblock99
vinceblock99 marked this pull request as ready for review November 15, 2026 21:50
leefaus and others added 7 commits November 15, 2026 17:10
Remove the Unix-only refusal from the buffer-based text and binary span planners. Keep dependency, semantic-baseline, capability and journal validation unchanged. Exercise invalid-UTF8 binary edits, binary/text transitions, immutable-object replay and materialization without platform-specific test exclusions.
Recording a projected span edit no longer refuses off Unix, but the change
store still did: save_projected_change_without_overwrite had a non-Unix stub
that returned "durable projected change publication is not supported on this
platform", so every record that produced a projected edit failed on Windows.
Import and remote submission carried the same platform refusal.

Use one publication path on every platform: stage the bytes in a temporary
file, flush it, and publish with a no-clobber rename that never replaces a
different object. Two steps stay platform-specific, matching sync_dir and
sync_directory elsewhere in this crate: directories are only fsynced on
Unix, and an existing object is flushed through a writable handle where the
platform requires one. Off Unix the rename is still atomic, but the
directory entry is not explicitly flushed.

Drop the matching refusals from import_change_bytes and remote submission,
and run the publication tests on every platform; only the process-kill test
remains Unix-only.
A projected span edit must be the only semantic operation on its path, but
record also emitted a semantic set_mode for the same path whenever the file
still had a CRDT trunk, which every file from pre-0.20 history does. Apply
then refused the change with "projected semantic path has multiple FileOps".

Windows always hit this: it observes a writable file as 0o666, which differs
from the default mode of a file whose history carries no mode attribute, so
every edit of such a file also recorded a mode change. Any platform hits it
when a legacy file is edited and chmod-ed in the same record.

When the path carries a projected span edit, record the attribute on the
graph inode alone, as is already done for a file that has no semantic trunk.
When a modified file's CRDT cache cannot be used, record falls back to the
file's legacy span baseline. That fallback also ran for a path whose name
is being resolved, and span recording refuses such an edit, so the record
failed with "span recording does not yet support this edit".

Whether it happened depended on which claimant won the name, which follows
the change hashes and so varies between runs. It surfaced on Windows in
test_emptying_file_after_orphan_view_merge_removes_every_copy: a writable
file is observed there as 0o666, so the orphan view's change also records a
mode and depends on the change that first added the file, leaving two
claimants after the merge. Any platform reproduces it with a 0o666 file.

Leave a path that is being resolved on the graph route, as before the
legacy span fallback existed. The new test repeats the scenario so that
both winners are exercised.
@vinceblock99
vinceblock99 merged commit 86b3c3c into dev Nov 15, 2026
10 checks passed
@vinceblock99
vinceblock99 deleted the fix/ancestral-bug branch November 15, 2026 22:39
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.

2 participants