D352 - The harvest took names because the work was naming

decided · 2026-08-27

docs/REFERENCES.md says it plainly: "Symbol names, and nothing else… It contains no implementation, no structure layouts, and no constants." The FreeBSD checkout is shallow, blob:none-filtered and sparse to lib/ for exactly that reason.

That was right. The naming loop needs names - a NID is a hash of one - and 22 MB bought 2,497 of them.

The work has moved from naming to implementing, and implementing needs numbers. Eight functions went in before this and the gap cost three answers in one session (D343, D350):

• SIGPIPE was recovered from the guest's own call argument, which was luck.
• A sysctl MIB was recorded as [1, 14, 8, 0] with its meaning left open.
• errno was left unset, because ENOENT's value was not derivable from anything lawful here - and setting an invented one is how a wrong constant gets copied.

Sockets would have been much worse. socket(AF_INET, SOCK_STREAM, 0) cannot be mapped onto a host socket without knowing what those are, and a wrong value creates the wrong kind of socket: a silent, late failure of precisely the shape principle 3 exists to stop.

Widening is within the boundary, not an exception to it

A #define AF_INET 2 in a public header is an interface fact - the same category as a symbol name, and arguably more clearly so, because it is the ABI. It is not implementation and not a structure carrying creative choices. FreeBSD is BSD-2-Clause and CLAUDE.md names it oracle #1, "lawful, citable, and the strongest reference available". The old restriction was self-imposed and tighter than the principle requires.

Sparse to lib/ plus sys/sys, sys/netinet and include: 22 MB to 25 MB. 903 constants from six headers into crates/orbistoun-hle/data/abi-constants.toml.

The measurement and the citation agree, and neither came from the other

Four constants were confirmed against measurements taken before the harvest existed:

constant	header	measured independently
SIGPIPE	13	klogsrv passed 0xd to signal
CTL_KERN	1	MIB[0] of the dumped call
KERN_PROC	14	MIB[1]
KERN_PROC_PROC	8	MIB[2], from a caller the symbol table names find_pid

KERN_PROC_PROC is commented "only return procs", so [1, 14, 8, 0] is enumerate all processes - which is what a function called find_pid would ask for. D350's open question is answered, and answered by two routes that never touched each other.

Two things that keep it honest

They are FreeBSD's numbers, not the target's. The target is FreeBSD-derived, which is why they are worth having and also why they are not facts about it. Each is published about FreeBSD and assumed about the guest; the data file says so in its own header, and a guest passing a value that disagrees is what would show it.

Read, never retyped. abi_constant parses the file. A value copied into Rust would be untraceable - a reader could no longer tell a harvested constant from a remembered one, which is the entire distinction known_by keeps. A test pins SOL_SOCKET at 0xffff precisely because it is 1 on several other platforms: a table built from recall would differ there first, and that test would catch it.

The guest confirms the whole path. main.c:278:sysctl: error 0 became error 2 - harvested ENOENT, through this project's __error, into the guest's own diagnostic.