[Japabu]

Two records named comments that say something false: that the kernel uses toyos-fat32, and what syscalls the ABI no longer declares used to do. This corrects the comments, corrects the same false claim in root README.md, and deletes both records. Comments and README text only; no code changes.

record	action	evidence
three-comments-say-the-kernel-uses-toyos-fat32	two comments corrected (build/fatformat.rs module doc, populate's doc in build/image.rs); record deleted	git grep -i fat32 -- kernel finds nothing. The third comment, lib/fat32/Cargo.toml's header, was already corrected on main
comments-still-name-syscalls-the-abi-deleted	three module docs cut down to what is true of the test now (abuse_connect_flood.rs, abuse_pipe_owner.rs, shm_release_reclaims.rs); record deleted. shm_release_reclaims.rs's remaining claim about handle count was false too and is corrected	The record's exit git grep -w -E 'SYS_WAITPID|SYS_CONNECT|SYS_AUDIO_SUBMIT|SYS_PIPE_OPEN|SYS_SOCKET_CREATE|SYS_LISTEN|SYS_RELEASE_SHARED' -- kernel system tests finds nothing. The fourth comment, on Pipe::backing, is gone with that function. A mapping holds the region (kernel/src/object/shm.rs header), on_zero_handles revokes only an aperture, and SharedMemory::drop unmaps before its handle closes, so the doc now says a region's frames go back once its last handle and last mapping have gone

Records: 752 on main at 34cb4c918, 750 on this branch.

Decisions a reviewer may want to question

• abuse_pipe_owner.rs loses more than the syscall names. Its second and third paragraphs described the test by contrast with the sweep it replaced ("the property that replaced it", "the sweep cannot survive the fail-fast flip"). Without the first paragraph those have no referent, so they are rewritten to state the property and why each attack is a child of its own.
• abuse_connect_flood.rs's opening history is cut too (the queue that had no depth, the eager 2 MiB rings), to the contract its assertions hold: a connect past the depth is ResourceExhausted, a full queue costs under a quarter of eager rings, and the first byte on a live connection takes a frame.
• Root README.md is corrected here rather than recorded. Lines 199-201 said the running kernel appends its log through its own FAT32 write path, and lines 305-306 that the kernel writes a log file per boot. logkeeper writes it (system/logkeeper/src/store.rs, through the file server). It is the same false claim the first record names.

Checks

• cargo clippy -p toyos-build --all-targets -- $ADOPTED -D warnings (the $ADOPTED lints of build/clippy.rs) at 42ffcadab: exit 0 (development host). toyos-build is the only touched package host clippy lints: tests/guest/rust-tests is outside the host workspace and has no clippy shape, and README and issues/ are not code. The merge queue runs the full host checks.
• cargo run -- --ci host lint at 7a8f3202b: exit 0. The build/ comments are unchanged since. Two later full runs ended red on the development host, not on this change: no space left on device, then a missing temporary directory under the kernel's aarch64 shape, a package this branch does not touch. The run above supersedes them.
• Doc comments and README text only: no guest row can change; none was run.
• Branch merges cleanly with main at b56693152, which holds issues: sixteen records whose defect, code or test is gone are deleted #1220, issues: two duplicate pairs merged, and two records renamed for the test each still owes #1221 and issues: records name the programs and files the tree has #1222 (git merge-tree), and the merged tree cites neither deleted slug.

🤖 Generated with Claude Code

[Japabu]

Review of #1224 at 7a8f3202b (base 34cb4c918)

Assessment. +20/−90. Every changed line in build/ and tests/ is a //! or /// comment; no code token changed. Both records' exits hold:

• git grep -i fat32 -- kernel finds nothing.
• The SYS_* exit grep, widened to sdk lib build, finds nothing.
• Pipe::backing is gone.
• lib/fat32/Cargo.toml's header no longer names the kernel.

The two comments in build/ are now true: toyos-fat32 has no format path (lib/fat32/src/lib.rs, "No formatting"), and refused_writes.rs judges with toyos-fat32-check. One comment the PR keeps is false, and one discovery is left unrecorded.

Findings

BLOCKER: tests/guest/rust-tests/src/bin/shm_release_reclaims.rs:3-4 keeps "A region's life is its handle count, and the zero-handle hook is where the mappings go."

• Both halves are false.
• kernel/src/object/shm.rs:7-11: "A mapping holds the object, and a handle holds nothing of a mapping: the object's frames go back when the last of its handles and mappings has."
• SharedMemObject::on_zero_handles calls revoke, which returns immediately for anything but an aperture (shm.rs:312-328).
• The SDK unmaps in SharedMemory::drop before its handle closes (sdk/toyos/src/shm.rs:162-167).
• The PR says these docs are cut "to what is true of the test now", and this sentence is the one that is not.
• Fix: delete it, or say that a region's frames go back once its last handle and last mapping have gone.

BLOCKER: the root README.md discovery is not recorded.

• The PR's "Not changed" section finds the same false claim this PR closes a record for: "the running kernel then appends its log … through its own FAT32 write path" (README.md:199-201), and "the kernel writes a log file per boot" (README.md:305-306).
• logkeeper writes the log.
• Leaving it out of scope is fine, but root CLAUDE.md requires an unrelated discovery to be recorded, and a PR body is not the tracker.
• Fix: file a record, or correct the two sentences.

SUGGESTION: as the PR notes, abuse_connect_flood.rs:3-16 still tells history: the queue that had no depth, the discarded bool, the eager 2 MiB rings. Root CLAUDE.md puts history in commits. Cutting it to the contract (a connect past the depth is ResourceExhausted, and a pipe takes memory only as it holds bytes) costs a few lines now.

Checked and accepted:

• abuse_pipe_owner.rs's rewrite matches End::Fault(Fault::Handle) (sdk/abi/src/end.rs:32) and the test's children.
• abuse_connect_flood.rs's new paragraph matches main(), which floods its own port::create().
• I did not rerun the reported --ci host lint result. A comment-only change gives no reason to doubt it.
• The branch merges cleanly with issues: sixteen records whose defect, code or test is gone are deleted #1220, issues: two duplicate pairs merged, and two records renamed for the test each still owes #1221, issues: records name the programs and files the tree has #1222 and main.

SEND BACK

[Japabu]

Orchestrator verification at 42ffcad, after review r1's SEND BACK (comment-only blockers).

The diff against main changes no code token: every changed Rust line is a comment (checked by filtering the diff), the rest is README.md and two deleted issue records. Blocker 7 is resolved (shm_release_reclaims.rs now says a mapping keeps the region alive and the SDK unmaps before closing the handle); blocker 8 is resolved in the README (logkeeper writes the log, at both places). Clippy on toyos-build with warnings denied is clean at this head. Queueing.