GeezerVison

LLM Description written by Claude Code:claude-opus-5-5

Fixes #372

Root cause

vlt 1.3 records a new node flag bit, brotli = 4, in slot [0] of a vlt-lock.json node when it resolved a .tar.br alternate. The shared vlt node-line grammar (vendor/vlt_lock_text.rs::parse_node_entry_text) whitelists slot [0] ∈ {0,1,2,3}, so every brotli node fails to parse. Hosted, vendored, heal and VEX discovery all read nodes through that parser, so hosted refuses the lock as "not canonical" (exit 0, nothing redirected), and vendored fails with vendor_lockfile_version_unsupported.

Changes (b3996a6)

• Shared grammar: slot [0] is vlt's flag bitset 0–7 (node_flags); anything else is still refused.
• render_tuple_with_slots now takes the brotli bit explicitly, because slots [2]/[3] name one artifact and its hash. vlt derives the bit from slot [3]'s extension (tarballFormat, checked against @vltpkg/graph / @vltpkg/types 1.3.6). If slot [3] is omitted, vlt uses the stored bit to rebuild the URL. Callers:
  • hosted pin → hosted .tgz: bit cleared, so vlt ci keeps the lock byte-stable, as the issue verified.
  • vendored wiring → local dir: bit cleared.
  • upstream restore → registry .tgz integrity: bit follows the written URL (cleared when omitted).
  • vendored revert and superseding-carried-pin restore → the recorded original's bit.
• vlt_heal::reinstalls_after_removal: decided by the optional bit (flags <= 7 && flags & 1 == 0), so brotli prod/dev nodes (4, 6) are reinstalled.

Tests (red → green)

With only the old "0" | "1" | "2" | "3" grammar line restored, 3 of the 4 new tests fail. With the fix, all 4 pass:

• vlt_lock_text::node_line_grammar_accepts_vlt_1_3_brotli_flags (flags 4–7 parse; 8, 07, -1 still refused)
• vlt_lock_text::brotli_bit_follows_the_artifact_slot_three_names
• redirect::vlt::a_vlt_1_3_brotli_lock_is_pinned_with_the_brotli_bit_cleared (hosted: no refusal, bits cleared with dev kept, idempotent re-run, carried-pin restore puts bit 4 back)
• vendor::vlt_lock::a_vlt_1_3_brotli_lock_is_vendored_and_reverted_exactly (vendored wire + byte-exact revert)
• updated vlt_heal::only_prod_and_dev_nodes_are_reinstalled_after_removal

cargo test -p socket-patch-core --lib -- vlt: 171 passed, 1 failed. The failure is vlt_heal::an_unremovable_hidden_lock_keeps_every_store_entry, which depends on a 0o555 directory blocking writes, and this sandbox runs as root. The change doesn't touch it.

Local gate

• cargo fmt --all -- --check: clean
• cargo clippy --workspace --all-features -- -D warnings: clean
• cargo test --workspace --all-features --no-fail-fast: every vlt binary passes (e2e_redirect_vlt_build, e2e_vendor_vlt_build, e2e_vlt, e2e_safety_vlt, mode_migration_vlt). The only local failures don't involve vlt:
  • chmod-based tests (covgap_commands_vendor *_state_write_failure_*, vlt_heal::an_unremovable_hidden_lock_*). This sandbox runs as root, so a 0o555 directory doesn't block writes.
  • Go/PyPI e2e builds that hit ENOSPC because the sandbox disk filled up.
• The ignored real-vlt-1.2.0 matrix (--include-ignored vlt_pinned_matrix) can't fetch npmjs from the Rust test client in this sandbox. CI runs it: 460/466 checks green on b3996a6, 3 still in progress, 0 failed.

Per-issue checklist

• Hosted and vendored modes refuse vlt 1.3 locks whose nodes carry the new brotli flag (slot [0] = 4) #372 hosted refusal → a_vlt_1_3_brotli_lock_is_pinned_with_the_brotli_bit_cleared
• Hosted and vendored modes refuse vlt 1.3 locks whose nodes carry the new brotli flag (slot [0] = 4) #372 vendored refusal → a_vlt_1_3_brotli_lock_is_vendored_and_reverted_exactly
• Hosted and vendored modes refuse vlt 1.3 locks whose nodes carry the new brotli flag (slot [0] = 4) #372 heal flag check → only_prod_and_dev_nodes_are_reinstalled_after_removal

Follow-ups (not in this PR)

• The CI vlt matrix pins vlt 1.2.0, and its e2e fixtures use npmjs bytes that advertise no .tar.br alternates. A real-vlt-1.3 e2e would need a mock registry that serves alternates.

Generated by Claude Code