Remove --download-mode and the diff download path (#792)

[Mikola Lysenko (mikolalysenko)]

Removes the diff download path and the --download-mode flag. Patch content is now always fetched as per-file blobs.

Why

On #792 the maintainer said: "We should try to clean up older and deprecated options we don't need any more." The decision on the issue is option A, done fully in v5. v5 is not stable yet, so we skip the one-major diff alias and remove the diff path and the flag in one PR. With diff gone, file would be the only value, so the flag would do nothing.

What changed

Core

• Deleted patch/diff.rs, the diff branch of apply, AppliedVia::Diff and PatchSources::diffs_path.
• Deleted DownloadMode, fetch_missing_sources and get_missing_archives from api/blob_fetcher.rs. The download loop fetches only blobs, and repair calls fetch_missing_blobs directly.
• Deleted ApiClient::fetch_diff and package::read_archive_filtered. The rest of package.rs stays, because vendor, hosted npm and vlt heal use it.
• Removed qbsdiff and its profile blocks (qbsdiff, bzip2, libbz2-rs-sys, suffix_array) from every Cargo.toml and from Cargo.lock.
• .socket/diffs is swept whole, like .socket/packages. cleanup_unused_archives is gone; only its own tests used it.
• apply_package_patch and apply_go_redirect no longer take a uuid argument. It only fed the diff lookup (about 40 call sites).
• Removed apply's mismatch-blob top-up. Staging now fetches every afterHash blob before apply runs, and apply stops before writing if any download fails, so that code could never fetch anything.

CLI

• --download-mode and SOCKET_DOWNLOAD_MODE are removed with no alias.
• Staging in fetch_stage.rs fetches blobs only, and repair has no second download pass.
• appliedVia is always "blob".

User-visible changes

• Passing --download-mode to any command is now clap's unknown-argument error (exit 2). A leftover SOCKET_DOWNLOAD_MODE is ignored.
• A cold-cache apply/get/scan/repair makes no /diffs/ requests.
• .socket/diffs/ is no longer read. repair, scan --prune, remove and rollback sweep every archive in it. The cleanup output still says "N unused diff archives" when it removes old ones.
• --json appliedVia no longer has the "diff" value.
• Kept for shape stability: repair's Downloaded/Verified event keeps details.mode, and the patch_fetched telemetry event keeps download_mode. Both are now always "file". Dropping them is the alternative if reviewers prefer it.

Docs

• CLI_CONTRACT.md: removed the flag and env rows, added a "removed in v5.0" note, made appliedVia "blob" only, updated the GC, prune and repair-event text, and changed the defaults example to name --vendor-source.
• docs/migrating-to-v5.md: added a Retired spellings row and rewrote the closing note about .socket/diffs/.
• CHANGELOG.md is unchanged.

Tests

• Deleted diff_e2e.rs, diff_created_file_e2e.rs and every diff and download-mode case.
• Added parse tests that apply --download-mode file|diff and repair --download-mode file exit 2.
• Added cold-cache checks to apply_network, the fetch-stage covgap test and the repair lifecycle test: no request path contains /diff and no .socket/diffs is created. The repair test also checks that a stale .socket/diffs/<referenced-uuid>.tar.gz is swept.
• cargo test -p socket-patch-core --no-fail-fast: 6119 passed, 1 failed. The failure is utils::digest::tests::production_digests_go_through_the_helpers, which also fails on main and names files this PR doesn't touch.
• cargo test -p socket-patch-cli: the lib, the 35 integration targets this PR touches, rollback, vendor, spawn_env_hygiene, in_process_remove_repair_lifecycle and the coverage_fix_*_silent_mute_exit targets all pass. The lib, rollback, spawn_env_hygiene, in_process_python_envs and the core api::client tests were re-run after rebasing onto main.
• cargo tree shows no qbsdiff.
• Not run locally: docker e2e and a full cargo test --workspace. CI covers those.
• Clippy with -D warnings reports one warning, an unused unix_default in python_crawler.rs. It comes from main, not this PR.

Review findings fixed

• tests/spawn_env_hygiene.rs still listed the deleted diff_created_file_e2e.rs in PENDING_RAW_SPAWNS, which would have failed no_new_bare_binary_spawns. Removed it.
• Repair's download-message unit tests used the diff-archive noun for output repair can no longer produce. They now use the blob noun.
• Removed stale comments that pointed to deleted code in rollback.rs, rollback_multicopy_blob_gate.rs, in_process_remove_repair_lifecycle.rs, golang_local.rs and args.rs.
• rustfmt also reformatted a few unrelated lines in vendor.rs and telemetry.rs. They are harmless.

Coordination

#966 touches the same files (args.rs, get.rs, scan/mod.rs, repair.rs, both docs, the cli_parse_* tests). Whichever PR lands second rebases, and both add rows to the Retired spellings table. Blob retry (#676) now covers one artifact kind, because diffs are gone.

Closes #792
Closes #791 (there is no --download-mode value left to validate)

🤖 Generated with Claude Code

---

Note

Medium Risk
This changes the core patch fetch/apply path and is a breaking CLI removal; behavior depends on blob staging succeeding where diff archives previously covered gaps.

Overview
v5 removes diff-based patch downloads and the global --download-mode / SOCKET_DOWNLOAD_MODE controls. Passing the flag is a parse error; patch bytes are always loaded as per-file blobs under .socket/blobs/.

The apply/repair/staging pipeline is blob-only: fetch_stage no longer fetches or honors .socket/diffs/ archives, repair drops the secondary “created-file blob” pass, and apply removes the on-demand mismatch blob top-up (staging must fetch all needed afterHash blobs up front). qbsdiff and related lockfile deps are dropped. JSON appliedVia is always "blob"; repair telemetry still reports details.mode: "file" for compatibility.

Obsolete storage is swept, not read: leftover .socket/diffs/ and package archives are treated as junk and removed by repair, remove, scan --prune, and related GC—matching updated CLI_CONTRACT.md and e2e tests that assert no /diff API calls on cold cache.

^{Reviewed by Cursor Bugbot for commit 36e534e. Configure here.}

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

[agent] CI status: all green on 1357bcf (552 pass, 6 skipped).

• Fixed: CodeQL rust/cleartext-logging alert in crates/socket-patch-cli/tests/remove/remove_duality_invariants.rs. The new assert message interpolated the archive uuid, so it now labels the entries A/B instead (3be03b0).
• Infra, not this PR: hosted-e2e failed when corepack couldn't fetch yarn from registry.npmjs.org. The Windows Gradle multi-project e2e failed on a Maven Central 403. The CodeQL Analyze (rust) SARIF upload also failed, and default-setup runs can't be re-run, so an empty commit (1357bcf) retriggered the checks. All of them passed on the new head.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

[agent] Merged main, CI green; ready for review.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Ready for review (burn-down agent).

• Head: 9fc3cd30571fa4e08089138f0272e167e966ff3d
• CI: 520/520 check runs green (success/skipped/neutral) on this head, mergeable, no conflicts.
• Bugbot: reviewed this head (Cursor Bugbot check: success), no unresolved review threads.
• Changelog: untouched.

Nothing specific flagged for the reviewer beyond the PR description.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

bugbot run

---

Generated by Claude Code

[cursor]

Stale Bugbot comment from a previous run.

[Mikola Lysenko (mikolalysenko)]

bugbot run

---

Generated by Claude Code

[cursor]

Stale Bugbot comment from a previous run.

[Mikola Lysenko (mikolalysenko)]

Ready for review at 024c78b4d.

• CI: every check on this head green; merges cleanly into current main (88d5c598).
• Bugbot: reviewed this head, no new issues; no open review threads.
• Large deletion (+506/-5096 over 84 files): worth checking nothing outside the --download-mode / diff-download path went with it.
• Slack: not announced (this run's Slack connector has no send tool); the next run should retry.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

[final reviewer] Merged main at d06ce6f6 to clear a conflict in crates/socket-patch-core/src/api/client.rs. main had renamed is_valid_uuid to is_uuid_shaped, including inside fetch_diff, which this PR deletes. I kept the deletion. No new non-merge commits since the re-look request for 6773eaa and df96c49e. cargo build (core and cli) and the api::client tests (125) pass. clippy and rustfmt report nothing in the file.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Burn-down agent: the main merge at d06ce6f didn't compile the cli test targets. main's new tests/in_process_target_ambiguity.rs (#1034) sets GlobalArgs::download_mode, and this PR deletes that field. Pushed 36e534e, which drops that one line; cargo check -p socket-patch-cli --all-targets, workspace clippy, and the core api:: + target-ambiguity/get/apply-parse tests pass locally. I removed Ready for review until CI is green on the new head.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

bugbot run

---

Generated by Claude Code

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 36e534e. Configure here.}

[Mikola Lysenko (mikolalysenko)]

Ready for review (burn-down agent).

• Head: 36e534ec
• CI: every check green except main-wide failures that also fail on main at a8e9397: e2e (ubuntu-latest, e2e_safety_pnpm) (withdrawn free minimist@1.2.2 patch, Merge queue blocked: production no longer publishes free minimist@1.2.2 patch #1293; fixes in Repin live minimist@1.2.2 suites to republished patch 642d7f02 (#1293) #1301/Follow the republished minimist canary patch #1302), the vlt native (*) compat legs, and the ci-ok roll-up.
• Bugbot: reviewed this head, no findings; no open review threads.
• No CHANGELOG.md changes.
• Removes the --download-mode flag and diff download path (user-visible CLI change; 85 files).

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

[final reviewer] I disarmed auto-merge. Tanmay Singla (@Tanmay182003), three non-merge commits landed after your approval at 9fc3cd30, head now b9a36289:

• 6773eaa6 (product code, commands/fetch_stage.rs +7/−1 with a test): under --silent, a fatal staging failure now prints the per-blob hash and reason lines before the error, because the removed diff path used to print them. This addresses Bugbot 4223475496. --json stays silent on stderr.
• df96c49 drops --download-mode from the repair org-fallback test.
• 36e534e drops the removed download_mode field from main's new tests/in_process_target_ambiguity.rs, so the test targets compile after the main merge.

Please take another look at 6773eaa6 in particular and re-approve when you're happy.

---

Generated by Claude Code