[Japabu]

The owner placed transport in the SDK (2026-10-21): it is the session layer every userland client and server speaks, and std links it through toyos. This PR moves toyos-transport from lib/blockring/transport to sdk/transport. The package keeps its name and its contents are unchanged.

What moved where

• The whole crate moved, and nothing was split off. It is forbid(unsafe_code), no_std and makes no system call. Its adapters already live outside it: sdk/toyos/src/net/stream.rs, plus netstack's lent.rs and node streams.rs. So no part of it is a syscall adapter.
• Path updates. Everything that named the old path now names the new one:
  • the workspace member list
  • the path dependencies of toyos, toyos-blockring, netstack and netstack's node
  • the crate's own toyos-untrusted dependency (path only; its use of Untrusted is untouched)
  • PUBLISHED in build/sdkversion.rs
  • SYSROOT_SOURCES, SYSROOT_MANIFESTS, STD_MANIFESTS and the std patch.crates-io path in build/sysroot.rs
  • the registry-refusal test in build/toolchain.rs
  • the three issue records that cite files in the crate.
• No fork change. rust/library/Cargo.lock names toyos-transport by version only (no source path), and std reaches it through sysroot::std_patch, so no fork pin moves. The image build logs show std compiling toyos-transport from sdk/transport.
• No change needed for issues: four records of today's landings, checked against the code and the CI logs #960. issues: four records of today's landings, checked against the code and the CI logs #960 has since landed and is merged in here (496df25); its record names the crate, not its path.
• Out of scope. toyos-untrusted is untouched here; its removal follows as a stacked PR. Nothing is published to crates.io.

Checks (tree 191dd46, macOS aarch64 host)

The head, 496df25, adds only the merge of main (#958 crypto parity, #960 issue records); neither touches the transport, its consumers or the sysroot wiring, so these results stand.

check	command	exit	log
host suite	cargo run -- --ci host (87 steps)	0	ci-host.log
x86-64 image	cargo run -- --build-only --arch x86_64	0	build-x86_64.log
AArch64 image	cargo run -- --build-only --arch aarch64	0	build-aarch64.log
blockring guest row (QEMU x86-64)	cargo test --test toyos-build -- nvme_disk_boots	0	guest-pr1.log

nvme_disk_boots is a booted guest whose system volume is read through diskserver over blockring, which runs on these rings. The std stream path (stream_handles) is a member of the shared metal boot only. I requested no metal run: the move changes no source, so the guest bytes differ only in embedded source paths.

Risks

None known beyond the path itself. Anything outside this repository that hard-codes lib/blockring/transport would break; searches of the tree and of the Rust fork's library manifests found none.

🤖 Generated with Claude Code

[Japabu]

Review of #963 at 496df25 (round 1)

Architecture. This matches the owner's decision of 2026-10-21 that transport belongs in the SDK, with shared host-testable logic kept apart from syscall adapters. sdk/transport is no_std and forbid(unsafe_code). Its only dependencies are core and toyos-untrusted, and it contains no syscall, toyos_abi use or asm! (checked by grep over sdk/transport/src). The adapters remain in sdk/toyos/src/net/stream.rs and netstack, so moving the whole crate without splitting it is correct, and nothing is merged into toyos. The package name is unchanged and nothing is published.

Completeness.

• git grep -e 'blockring/transport' -e 'blockring.transport' at 496df25 finds nothing (exit 1).
• The remaining toyos-transport path references are the workspace member, the four path dependencies (toyos, toyos-blockring, netstack, netstack node), PUBLISHED, SYSROOT_SOURCES/SYSROOT_MANIFESTS/STD_MANIFESTS, the std-patch test, the toolchain.rs registry test and the three issue records. All of them point at sdk/transport.
• build/ci.rs (TRANSPORT) and the clippy feature coverage refer to the crate by package name, so they need no change.
• The crate's own toyos-untrusted relative path is corrected (../../lib/untrusted).

Fork. The claim that no fork change is needed holds. At the pinned rust commit 73f63e244, library/Cargo.lock lists toyos-transport 0.1.0 with no source, so the lockfile is unaffected by the path.

Evidence. ToyOSOrg/evidence 29a1f9ecc round1, at tree 191dd46:

• --ci host: 87 steps, EXIT=0.
• Both --build-only images: EXIT=0. The build log shows Compiling toyos-transport v0.1.0 (<tree>/sdk/transport).
• nvme_disk_boots: EXIT=0.

The later merge to 496df25 adds only #958 (lib/crypto, a crypto parity guest binary, Cargo.lock +1) and the #960 issue records (checked with git diff --stat 191dd4618 496df257d). None of it touches the transport, its consumers or the sysroot wiring, so the evidence still applies. For a pure rename this is enough; no metal run is needed.

Net. 15 lines changed in, 15 out, plus the rename. No production or test growth.

Findings

None.

LAND