GeezerVison

cursor Bot left a comment •

edited

Loading

Uh oh!

There was an error while loading. Please reload this page.

cursor Bot left a comment •

edited

Loading

Uh oh!

There was an error while loading. Please reload this page.

, #1066)

GC was reported four ways. repair and remove buried the sweep in
artifact-level event details with no byte count, while rollback and
scan --prune printed a hand-built `gc` object. The contract documented
summary.bytesFreed, summary.bytesDownloaded and events[].bytes, but no
command emitted any of them, so its GC jq recipe returned null.

- json_envelope::GcReport {removedBlobs, removedDiffArchives,
  removedPackageArchives, bytesFreed} is built from the three sweep
  passes and serialized identically everywhere: the envelope's new
  top-level `gc` (repair, remove), rollback's `gc` and scan's `gc`. The
  hand-written json! blocks are gone.
- summary.bytesFreed is always present and mirrors gc.bytesFreed.
  events[].bytes is set on the GC carrier event and on --update's
  downloaded event. summary.bytesDownloaded is dropped from the contract.
- repair's GC carrier event no longer bumps summary.removed/verified,
  matching remove: summary counters count patch entries, and the sweep
  totals live in `gc`.
- remove's human output now names the diff/package archives it sweeps.
- rollback --json: rolledBack and failed now span the agent, vendored
  and hosted legs (#1066). A run where something failed and nothing was
  rolled back, already original or not installed now reports
  status "error" with error.code rollback_failed instead of
  partial_failure. Exit codes are unchanged.
- CLI_CONTRACT.md: the envelope and PatchEvent schemas, the PatchAction
  "Emitted by" column, the per-command action matrix and the GC jq
  recipe now match the emitters. New unit tests pin the documented
  summary/gc/PatchEvent key sets against what json_envelope serializes.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

The --dry-run preview of scan --prune/--sync used its own vocabulary
(prunableManifestEntries, orphanBlobs, orphanDiffArchives,
orphanPackageArchives, revertableVendoredEntries, vendorOrphanDirs,
bytesReclaimable). It now prints the same keys as the wet pass and every
other GC-running command, counting what the pass would remove, and leaves
out only the keys a real pass alone can fill (keptVendoredEntries,
failedVendoredEntries, skipped, warnings). v5.0 MAJOR.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

All packages refused and nothing rolled back is a failed run since #1066.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

Assisted-by: Claude Code:claude-opus-5-5

The shared bun.lockb rollback helper expected status "error" for every
caller, so the hosted -> vendored -> hosted takeover leg went red: there
a manifest record makes the agent leg report the copy already original,
and the run is correctly a partial_failure. A branch rewrite had dropped
the earlier fix for this. Each caller now states which outcome it
expects, so the takeover leg checks partial_failure and the hosted-only
alias/transitive shapes keep checking rollback_failed.

Assisted-by: Claude Code:claude-opus-5-5

cursor Bot left a comment

Resolve against main's removal of the diff download path (#1049) and
the restore-blob GC fix (#1316): repair drops the created-file blob
pass and keeps the GcReport carrier; remove keeps the archive noun
loop; the contract keeps "update" and drops the removed paidRequired
status; the envelope contract test uses AppliedVia::Blob.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

main's new test read gc.prunableManifestEntries, which this branch
renamed to prunedManifestEntries for the dry-run preview.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

….lockb e2e

- json_envelope's contract tests read CLI_CONTRACT.md through one
  LF-normalized copy, so a Windows checkout (CRLF) no longer misses the
  ```jsonc fence (review thread on #1273; full-scope test windows).
- e2e_bun_lockb's shared-bundled case also has an agent copy that fails
  (hash_mismatch on a bundled copy the patch never touched); with #1066's
  all-leg counter, `failed` is the hosted refusal plus those agent failures.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

Resolve remove.rs imports: keep main's KeepCause and this branch's GcReport.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

main's #1278 e2e (sync_keeps_entry_whose_shared_cache_copy_is_still_patched)
reads the preview's prunableManifestEntries, which the one GC shape
renamed to prunedManifestEntries; the contract text it brought gets the
same rename.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

#1305's sync_keeps_entry_whose_shared_cache_copy_is_still_patched
asserted the scan --sync --dry-run preview via gc.prunableManifestEntries,
a key #1273 retires: under the one GC shape a dry run reports would-be
prunes as gc.prunedManifestEntries. The keep logic itself merged intact
(preview and wet pass both keep the still-patched itoa entry, and the wet
pass reports cargo_cache_patch_kept in gc.warnings[] with the purl and the
rollback remedy), so only the test read the wrong key. Assert the same
ryu-only list under the new key.

CLI_CONTRACT.md's scan --prune paragraph still named the retired preview
keys (prunableManifestEntries, revertableVendoredEntries,
vendorOrphanDirs); point it at the unified keys and note that the
preview, like every warnings[] entry, carries no cargo_cache_patch_kept.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

# Conflicts:
#	crates/socket-patch-cli/CLI_CONTRACT.md