Changes touching this path

• a land run from a lane seeds the target/ of the lane with the release builds of registry dependencies that the primary target/ holds, before its pre-land gates: the new loot-first cargo_seed copies a unit only when the Cargo.lock of the position names its package as a registry package and not as a path package, under a shared lock on the cargo build locks of the source profile directory (skipping one a build holds), copies rather than hard-links (a same-name rebuild was measured writing fingerprint files and a proc-macro dll and pdb in place), renames each fingerprint into place last, overwrites nothing the lane holds and keeps file times, and prints a seed: line and a seed span; it never refuses, and is skipped under --skip-tests and in the primary. measured in fresh copies of a lane, twice each: the perf gate release build 70.67 and 70.07 s cold against 14.94 and 15.36 s seeded, the loot-wasm wasm32 release build 21.34 and 21.09 s against 7.43 and 7.36 s, for a seed of 5.06 and 5.11 s (1.43 GB). debug is not seeded: the primary debug builds predate the line-tables dev profile, and a seed of them took cargo test --no-run from 336 Compiling lines to 326 and saved no time, so a lane that never built Rust still pays that build. pinned in 7 cargo_seed tests, one the hunt experiment on a scratch workspace (two lanes over one seed each rebuild their own path crates, while a whole-target copy hands a third lane the primary binary), and a land_via order census; the rename census names the new rename, and was red until it did. red under named mutations, each restored, over the 7: path packages copied (4 passed and 3 failed), the fingerprint not copied (5 and 2), destination hashes not consulted (6 and 1), a held lock ignored (6 and 1); file times left unset stays green on Windows, whose copy keeps them; the seed moved after the pre-land gates (0 and 1 over the census). cargo test green, 4823 passed over 144 test result lines with 14 ignored. no deploy owed; lands seed once the primary loot-first is rebuilt; it also files lsoupnqm (woyvmtwq) e23e4b8b · dbf3dbe6…
• serve::held_tips takes the tips of a have over the ancestry the forge graph holds rather than over direct parents, so a have that skips versions no longer has a manifest read for each version it names: on the offer_cost fixture a have naming every other version of a 20-change line, its head included, had 10 manifests read and now has 1, pinned in the new a_gapped_have_has_the_manifests_of_its_tips_over_the_graph_read, and ADR 0075 says so. the record_deposit doc states what defines its writers, the one door to the deposit ledger of a position, which takes the remote, instead of naming callers, a list that had left out the runner deposit. judgement calls: bundle_walk takes have, wants and depth as one Ask and drops its clippy::too_many_arguments allow, the rideable closure called in place is the helper serve::rideable, the seed: line prints MiB, and the cargo_seed doc says nothing is overwritten only of what the destination held when the seed read it, since the seed takes none of the lane build locks. not taken: sharing the batch loop of a push deposit with the runner deposit, which differ in what they seal, how they bound a call and what they record. red under named mutations, each restored, over the 17 offer_cost tests: tips over direct parents (16 passed and 1 failed), the named versions counted among their own ancestors (13 and 4). cargo test green, 4827 passed over 144 test result lines with 14 ignored. owes a forge deploy for the fewer manifest reads (xukurtyx) 30c6668f · dbf3dbe6…diff
• a land seeds the debug builds of registry dependencies as woyvmtwq seeded the release ones, from target/lane-seed under the primary, which the lands fill themselves: once a land from a lane that ran cargo test has published and freed the harbor, it copies the registry debug builds of that lane into target/lane-seed by the same rules, registry packages only by the Cargo.lock of the landed tree and again by that of the landing tree, copies under the cargo build locks, now taken shared on the source and exclusive on the destination, and nothing the destination holds already; the next land seeds its lane from there, where the debug builds of the primary predated the line-tables dev profile and matched no unit. measured in fresh copies of a lane, seven interleaved samples each of cargo test --locked --no-run --workspace at CARGO_BUILD_JOBS=4: a median of 162.7 s cold and 107.9 s seeded, 336 Compiling lines against 19, for a seed of 2.3 s over 387-391 units and 941-960 MiB, where the lane target/debug was 7.7 GiB; filling an empty seed directory took 2.6 s and a filled one copies nothing; after two hops every native link path under a target directory named the seeded copy, none the lane or the seed directory. the seed span on the timing line covers both profiles and the copy back closes a seed-refresh span. pinned in the new cargo_seed tests a_seed_directory_hands_the_next_lane_no_workspace_build (a lane into the seed directory and from there into a lane whose changed core is older than the seed, which rebuilds it and runs its own code) and a_destination_a_reader_holds_is_not_copied, and the orchestrator tests a_land_seeds_each_profile_from_its_source_and_copies_its_debug_builds_back and the_seed_directory_is_refreshed_after_the_harbor_is_freed. red under named mutations, each restored, over the 12 selected seed tests: path packages copied (4 failed and 8 passed), no destination lock (1 and 11), the destination locked shared (1 and 11), debug seeded from the target/debug of the primary (1 and 11), the copy back into the target of the primary (1 and 11), the copy back under the harbor (1 and 11). cargo test green, 4846 passed over 144 test result lines with 15 ignored. owes no deploy; lands use it once the release loot-first of the primary is rebuilt. it also carries comments on wwmonzor and rmwmkyvu and files qptwwkqx and uzkouqqx (lolzxwpk) f2cb7126 · dbf3dbe6…diff

Renames are not followed. loot's tree maps a path to an address, so a rename is a delete and an add. This list is the history of the name, not of the bytes.