Repository navigation
Fix review process and improve ontology materialization - #340
Merged
Merged
Conversation
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.
leefaus
force-pushed
the
fix/ancestral-bug
branch
from
November 15, 2026 16:33
6b2c9fb to
5ebe7d8
Compare
vinceblock99
marked this pull request as ready for review
November 15, 2026 21:50
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
force-pushed
the
fix/ancestral-bug
branch
from
November 15, 2026 22:11
1ef23bf to
4fb69ba
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This pull request adds comprehensive support for retrospective publication reviews in the Atomic CLI and core libraries. It introduces a new
publication_reviewmodule 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:
publication_reviewmodule inatomic-canonical, defining types and validation logic for publication reviews, including schema versioning, review subject structure, and artifact hashing. [1] [2]IntentGraphCLI command, allowing users to include a prepared review as context when generating intent graphs. [1] [2]Intent Graph and Evidence Attribution:
RPC and Service Integration:
review-shaclfeature foratomic-repositoryin the CLI's dependencies.Triage Output and User Interface:
Documentation and Vocabulary:
icebox(set aside) status.