Changes touching this path

• the LOOT_NET_TIMING line now gives the time to the headers and then the time the body took, with the body size as read, and is written when the body ends, where it gave only the time to the headers and the Content-Length, so a 52 MB /pull-grants body of 132 s read as a 3.9 s fetch; a body dropped before its end is written as not read to its end, with the bytes read of those announced, and a request that got no headers says so. the forge was measured, not changed: /pull-grants builds its whole answer from one Postgres query before the headers, nginx does not compress it, and over a copy of the primary store two whole pulls moved 55,502,699 bytes at 490 and 466 KB/s while a site asset from the same box came at 559 and 438 KB/s and Cloudflare at 25 MB/s, so the rate is the path from the VPS. pinned by a slow-body test host: red under named mutations, each restored: the line written at the headers (3 failed and 8 passed), reads not counted (1 and 10), a dropped response writing nothing (1 and 10), bytes() ending no line (1 and 10), the body time taken as the headers time (2 and 9). cargo test green, 4926 passed over 145 test result lines with 15 ignored. CONTEXT.md and ADR 0057 say so. owes no deploy (opxlsyom) 64a20f99 · dbf3dbe6…

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.