Repository navigation
A file's identity: stat answers a volume and a node, through the kernel, the file server, libc and std - #851
Conversation
…el, the file server, libc and std
`toyos_abi::syscall::Stat` gains `volume` and `node`, the pair POSIX calls
`st_dev` and `st_ino`. Every file was one file to a program that asked: libc
zeroed both, so LLVM's `sys::fs::UniqueID` answered one identity for every
path and clang's `FileManager` handed back the first file it opened for every
later one.
Kernel: a file's node is its `FileId`, which the file cache allocates from a
counter it never rewinds and which ROOT and `/tmp` already keep one per file
for as long as it exists; its volume is its mount's index plus one, recorded
in the open file. Every other object answers 0.
File server: `fileserver::volume::Identities` numbers a volume's files by the
key the volume names a file by, minted on first ask and told every move and
removal: DATA keys by path (its format keeps no inode), carrying a directory's
keys at its rename and dropping a directory that existed only for what was
removed; FAT keys by where the entry lies (`Fat32::entry_offset`, the identity
`same_entry` already used), so both spellings of a name are one file. The
volume's number is drawn at each start and answered in `HELLO`; the client
sets bit 63 on it, below which every kernel volume lies, so no server can name
a kernel volume and a restarted server is a new volume.
libc fills `st_dev`/`st_ino`; std carries both in `FileAttr`, answers them as
`std::os::toyos::fs::MetadataExt::{dev, ino}`, and its `DirEntry::metadata`
stats the entry, since a listing carries no identity.
What the kernel opens not, a directory or a symlink, still has no identity,
nor does a listing entry: the issue is renamed to that remainder.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017cSFvbD35xJ2kGANVdm23C
e81e768 to
be007c6
Compare
|
fileserver-host-tests, part 1 of 1, run on tree 96435c7d (head be007c6's) |
|
guest-suite, part 1 of 1, run on tree 96435c7d (head be007c6's) |
|
ci-host, part 1 of 10, run on tree 96435c7d (head be007c6's) |
|
ci-host, part 2 of 10, run on tree 96435c7d (head be007c6's) |
|
ci-host, part 3 of 10, run on tree 96435c7d (head be007c6's) |
|
ci-host, part 4 of 10, run on tree 96435c7d (head be007c6's) |
|
ci-host, part 5 of 10, run on tree 96435c7d (head be007c6's) |
|
ci-host, part 6 of 10, run on tree 96435c7d (head be007c6's) |
|
ci-host, part 7 of 10, run on tree 96435c7d (head be007c6's) |
|
ci-host, part 8 of 10, run on tree 96435c7d (head be007c6's) |
|
ci-host, part 9 of 10, run on tree 96435c7d (head be007c6's) |
|
ci-host, part 10 of 10, run on tree 96435c7d (head be007c6's) |
|
hosted-clang-hello-on-830-merged, part 1 of 7, run on 46a27b9 (#830's local head) with this branch merged, merge fe5ffc5f2, not pushed |
|
hosted-clang-hello-on-830-merged, part 2 of 7, run on 46a27b9 (#830's local head) with this branch merged, merge fe5ffc5f2, not pushed |
|
hosted-clang-hello-on-830-merged, part 3 of 7, run on 46a27b9 (#830's local head) with this branch merged, merge fe5ffc5f2, not pushed |
|
hosted-clang-hello-on-830-merged, part 4 of 7, run on 46a27b9 (#830's local head) with this branch merged, merge fe5ffc5f2, not pushed |
|
hosted-clang-hello-on-830-merged, part 5 of 7, run on 46a27b9 (#830's local head) with this branch merged, merge fe5ffc5f2, not pushed |
|
hosted-clang-hello-on-830-merged, part 6 of 7, run on 46a27b9 (#830's local head) with this branch merged, merge fe5ffc5f2, not pushed |
|
hosted-clang-hello-on-830-merged, part 7 of 7, run on 46a27b9 (#830's local head) with this branch merged, merge fe5ffc5f2, not pushed |
|
Host mutations, each applied as a checked patch to tree 96435c7d (head be007c6's), The script: #!/bin/sh
# Each mutation: checked apply, build and run the fileserver's host tests, record the exit, restore.
cd /Users/jan/Dev/jan/toyos-statid || exit 1
S=/private/tmp/claude-502/-Users-jan-Dev-jan-toyos/80e90c3a-7449-46b8-84cc-b59e9dd31ae9/scratchpad/statid/mut
for p in "$S"/m*.patch; do
name=$(basename "$p" .patch)
git apply --check "$p" || { echo "$name: does not apply"; exit 1; }
git apply "$p"
cargo test -p fileserver --lib > "$S/$name.log" 2>&1
echo "$name EXIT=$?"
git apply -R "$p"
done
git status --porcelainIts output: m1-never-remembersdiff --git a/userland/fileserver/src/volume.rs b/userland/fileserver/src/volume.rs
index fd85d7d1b..55b9d6193 100644
--- a/userland/fileserver/src/volume.rs
+++ b/userland/fileserver/src/volume.rs
@@ -155,11 +155,9 @@ impl<K: Ord> Identities<K> {
}
pub fn of(&mut self, key: K) -> u64 {
- *self.of.entry(key).or_insert_with(|| {
- let number = self.next;
- self.next += 1;
- number
- })
+ let _ = key;
+ self.next += 1;
+ self.next
}
/// The file at `from` is now at `to`, and whatever was at `to` is gone.Its run, whole: m2-data-rename-dropsdiff --git a/userland/fileserver/src/data.rs b/userland/fileserver/src/data.rs
index 697f7b71d..bfb027c62 100644
--- a/userland/fileserver/src/data.rs
+++ b/userland/fileserver/src/data.rs
@@ -710,7 +710,6 @@ impl<D: Disk> Volume for DataVolume<D> {
self.open.get_mut(&node).expect("indexed").path = to.to_string();
self.by_path.insert(to.to_string(), node);
}
- self.ids.moved(&from.to_string(), to.to_string());
self.forget_emptied(from);
Ok(())
}Its run, whole: m3-data-keeps-emptieddiff --git a/userland/fileserver/src/data.rs b/userland/fileserver/src/data.rs
index 697f7b71d..c1cb62845 100644
--- a/userland/fileserver/src/data.rs
+++ b/userland/fileserver/src/data.rs
@@ -289,7 +289,6 @@ impl<D: Disk> DataVolume<D> {
fn forget_emptied(&mut self, path: &str) {
let mut dir = parent(path);
while !dir.is_empty() && !self.is_dir(dir) {
- self.ids.forget(&dir.to_string());
dir = parent(dir);
}
}Its run, whole: m4-fat-rename-dropsdiff --git a/userland/fileserver/src/fat.rs b/userland/fileserver/src/fat.rs
index 846099147..34e74acdb 100644
--- a/userland/fileserver/src/fat.rs
+++ b/userland/fileserver/src/fat.rs
@@ -426,7 +426,7 @@ impl<D: Disk> Volume for FatVolume<D> {
// An entry the device will not show again is a file whose number goes
// with it: renumbered, never another file's.
match self.entry_offset(to) {
- Ok(to_key) => self.ids.moved(&from_key, to_key),
+ Ok(_) => self.ids.forget(&from_key),
Err(_) => self.ids.forget(&from_key),
}
releasedIts run, whole: m5-tree-move-dropsdiff --git a/userland/fileserver/src/volume.rs b/userland/fileserver/src/volume.rs
index fd85d7d1b..11e3e9f9c 100644
--- a/userland/fileserver/src/volume.rs
+++ b/userland/fileserver/src/volume.rs
@@ -190,7 +190,7 @@ impl Identities<String> {
}
for key in self.tree(from) {
let number = self.of.remove(&key).expect("listed by tree");
- self.of.insert(format!("{to}{}", &key[from.len()..]), number);
+ let _ = number;
}
}
Its run, whole: |
|
t14-staging-head, part 1 of 1, staged from tree 96435c7d (head be007c6's); the machine was not touched |
|
t14-staging-negative-control, part 1 of 1, staged from cc0315601: the base 1924919 with only the C case and its LICENSE count added; the machine was not touched |
|
**T14 reading, head |
|
T14 reading, negative control |
|
Review of #851 at be007c6 (tree 96435c7d) against BLOCKER
NOTE
SEND BACK |
…unt up once The review of be007c6 found the served volume word a SYS_RANDOM draw, distinct from another server's only with probability, when the kernel already mints a number distinct by construction: a pid names one process for the life of the machine (kernel/pure/proclife/pids.rs), and a restarted server gets a new one. HELLO's value2 is now the server's pid; the client still sets bit 63 on it, so no server names a kernel volume. The draw and its start-time expect are gone, and Stat's "no other file's" holds for served volumes without a condition. Vfs::volume resolved the target's mount a second time beside the open's own resolution. open_target and create_file now answer the volume with the FileId, from the one MountPoint they resolve, and Vfs::volume is deleted. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017cSFvbD35xJ2kGANVdm23C
|
Host mutation m6, the review's named one (FAT's rename keeps the displaced entry's number), applied as a checked patch to be007c6, Red, exit 101: The script: #!/bin/sh
# Each mutation: checked apply, build and run the fileserver's host tests, record the exit, restore.
cd /Users/jan/Dev/jan/toyos-statid || exit 1
S=/private/tmp/claude-502/-Users-jan-Dev-jan-toyos/80e90c3a-7449-46b8-84cc-b59e9dd31ae9/scratchpad/statid-r2/mut
for p in "$S"/${1:-m}*.patch; do
name=$(basename "$p" .patch)
git apply --check "$p" || { echo "$name: does not apply"; exit 1; }
git apply "$p"
cargo test -p fileserver --lib > "$S/$name.log" 2>&1
echo "$name EXIT=$?"
git apply -R "$p"
done
git status --porcelainIts output (run as m6-fat-displaced-keptdiff --git a/userland/fileserver/src/fat.rs b/userland/fileserver/src/fat.rs
index 846099147..ac4501325 100644
--- a/userland/fileserver/src/fat.rs
+++ b/userland/fileserver/src/fat.rs
@@ -403,11 +403,6 @@ impl<D: Disk> Volume for FatVolume<D> {
self.level(node)?;
}
let from_key = self.entry_offset(from)?;
- let displaced = match self.entry_offset(to) {
- Ok(key) => Some(key),
- Err(SyscallError::NotFound) => None,
- Err(e) => return Err(e),
- };
let replaced = self.fs.replace_rename(from, to).map_err(|e| {
if let Some(stranded) = &e.stranded {
println!("fileserver: {to} could not be put back and is under {stranded}");
@@ -420,9 +415,6 @@ impl<D: Disk> Volume for FatVolume<D> {
self.open.get_mut(&node).expect("indexed").path = to.to_string();
self.by_path.insert(to.to_string(), node);
}
- if let Some(key) = displaced {
- self.ids.forget(&key);
- }
// An entry the device will not show again is a file whose number goes
// with it: renumbered, never another file's.
match self.entry_offset(to) {Its run, whole: |
|
ci-host, part 1 of 9, run at head de60926, exit 0 |
|
ci-host, part 2 of 9, run at head de60926, exit 0 |
|
ci-host, part 3 of 9, run at head de60926, exit 0 |
|
ci-host, part 4 of 9, run at head de60926, exit 0 |
|
ci-host, part 5 of 9, run at head de60926, exit 0 |
|
ci-host, part 6 of 9, run at head de60926, exit 0 |
|
ci-host, part 7 of 9, run at head de60926, exit 0 |
|
ci-host, part 8 of 9, run at head de60926, exit 0 |
|
ci-host, part 9 of 9, run at head de60926, exit 0 |
|
guest-suite, part 1 of 1, run at head de60926, exit 0 |
|
t14-staging-head, part 1 of 1, staged from head de60926, exit 2 (staged); the machine was not touched |
|
Guest negative controls g1 and g2, staged for the T14 (not yet run; the machine is the orchestrator's). Both members are shared-boot members, which the QEMU suite does not run, so these can be judged only on the T14. Each patch was applied with What each must turn red, in one boot:
g1-served-volume-constant--- a/userland/fileserver/src/main.rs
+++ b/userland/fileserver/src/main.rs
@@ -599,7 +599,7 @@
let granted = if client.writes { RIGHT_WRITE } else { 0 };
// The volume is this start's pid, which the kernel never issues
// again: a restart numbers its files afresh, so it is another volume.
- let volume = u64::from(toyos_abi::syscall::getpid().0);
+ let volume = 0;
return Answer::Reply(Reply { value: granted, value2: volume, ..Reply::ok() });
}
if self.clients[&id].window.is_none() {g2-kernel-volume-constant--- a/kernel/src/vfs.rs
+++ b/kernel/src/vfs.rs
@@ -112,7 +112,7 @@
/// says: its filesystem's index plus one, so 0 stays "no volume" and every
/// kernel mount is below a file server's bit 63.
fn volume(&self) -> u64 {
- self.fs as u64 + 1
+ 1
}
}
|
|
t14-staging-guest-mutations, part 1 of 1, staged from e3c0c50ac (de60926 with g1 and g2), exit 2 (staged); the machine was not touched; git's progress carriage returns made newlines |
|
T14 reading, head |
|
T14 reading, mutation |
|
Review of #851, round 2, at de60926 against Earlier BLOCKERs
BLOCKERNone. NOTE
LAND |
, #851 and #853, into consent tests/common/qemu.rs: Profile carries both Desktop and MetalNoX2apic (#841). userland/compositor/src/session.rs: the crate import takes main's list, without CURSOR_PX (#814), plus the branch's PROMPT_POLL_HANDLES. Cargo.lock is main's, re-resolved: grants, and filepicker's toyos-manifest and the build's toyos-desktop. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017cSFvbD35xJ2kGANVdm23C
Main now carries the loader's relocation (#836), stat identity (#851), libc's realpath (#854), the kernel's directory identity (#855) and the second bench (#852). ALL_CONFIGS keeps consentcase and hostedclangcase; the host-tools issue takes main's os-release row. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017cSFvbD35xJ2kGANVdm23C
Takes up
issues/libc-stat-answers-one-serial-number-for-every-file.mdand renames it to the part that remains,issues/the-kernel-numbers-no-directory-symlink-or-listing-entry.md. The issue's two citations inissues/toyos-builds-itself.mdmove with it.Design
The shape
toyos_abi::syscall::Statgains two words,volumeandnode. Together they are the file's identity. The pair is the same through every path and handle that reaches the file, for as long as the file exists and its volume stays mounted, and no other existing file has it.volume = 0(and thennode = 0) means the object is not a file on any volume. libc fillsst_dev = volumeandst_ino = node. std'sFileAttrcarries both, andstd::os::toyos::fs::MetadataExt::{dev, ino}answers them. The file server's wire (toyos::fs) answers the same pair for a served file:Reply::nodeis in every reply that carries a stat, and the volume comes once per connection inHELLO'svalue2. Sotoyos::fs::Stathas both too.The volume number space is split so the two minters cannot collide. A kernel mount's volume is its filesystem's index in the VFS plus one (ROOT 1,
/tmp2), which is below 2^63. A file server's volume is its own pid. The client (toyos::fs::Dir) sets bit 63 (toyos::fs::SERVED_VOLUME) on whatever the server answers, so no server can name a kernel volume, whatever it says. Two served volumes are distinct by construction:kernel/pure/proclife/pids.rsholds that a pid names one process for the life of the machine and is never issued again, so two servers, or one server before and after a restart, never share a volume.The contract is "for as long as it exists and its volume stays mounted", which is what POSIX promises.
st_inois unique within a file system, and nothing promises it across a remount (Linux's vfat renumbers at every mount). A file server that restarts counts as a new mount: it is a new process with a new pid. So no file from before the restart shares a pair with a file from after it, and a file that lives through the restart gets a new pair.What each filesystem answers (surveyed in the tree)
/system)ReadOnlyBcacheFsAdapterFileIdname_to_idkeeps oneFileIdper name for the whole boot. The file cache allocatesFileIds from a counter it never rewinds, and drops a file only once it is deleted, which never happens on ROOT/tmpTmpFsFileIdrenameand retired atdelete; the same allocator/home,/apps, …)fileserver,DataVolumerename(a directory's carries everything under it) and dropped atunlink/rmdirbcachefs/src/fs.rs's leaf value is type, name, size, mtime, extents), so nothing on disk names a file across a renamefileserver,FatVolumeFat32::entry_offset), moved byrenameand dropped atunlink/rmdirsame_entry), which is what makesLOG/A.TXTandlog/a.txtone file; a rename writes the entry somewhere elsefileserver,AbsentNo ToyOS filesystem has hard links.
std::fs::linkisUnsupported, libc'slinkis refused, and neither format can represent one. So "distinct for distinct files, hard links included" is met by distinct entries. A filesystem that gains links will key its numbers by what the links share.The file server's numbers live in one host-tested type,
fileserver::volume::Identities<K>. It maps the volume's key for a file to a number from a counter that never rewinds. A number is minted on first ask, so a file nobody stats costs nothing. The map is bounded by the files that exist: every removal drops its key, and on DATA, where a directory exists only while something is under it, an unlink or rename that empties a directory drops that directory's key.What stays without an identity, and is recorded
The kernel opens no directory and no symlink (
SYS_OPENrefuses both), andSYS_READDIRanswers no identity. So on/systemand/tmp, libc'sstatof a directory andlstatof a symlink answerst_dev = st_ino = 0, and std answers the same. Pipes, consoles and the other non-file kernel objects answer 0 as well, andd_inostays the listing offset. That is what the renamed issue now records; it stays open with its exit.Roast
Cost. Two
u64s inStat(the ABI change). Oneu64in the kernel's open file (its volume), which the open's own mount lookup answers:Vfs::open_targetandVfs::create_filereturn the volume together with theFileId, so there is no second lookup. One map per served volume. One word in the file server's reply, and one in itsHELLOanswer (its pid, oneSYS_GETPIDper connection). Two lines in libc. A two-method trait in std.DirEntry::metadataon ToyOS becomes anlstatof the entry, as it is on Unix, because a listing carries no identity. Before, it answered the listing's size with mtime 0.What it saves. clang's
FileManagerkeys every file it opens on(st_dev, st_ino)(LLVM'ssys::fs::UniqueID). When every file has the same pair, the first file it opens answers for every later one. That is #830'shosted_clang_helloreadinghello.cback in place ofstdio.h. The same pair is howcpandmvrefuse to copy a file onto itself, how a tree walker sees a cycle, how SQLite keys its locks, and howfind -xdevstays on one volume.Simpler shapes, refused:
MetadataExtall want two words; packing them saves nothing.FileIds are already unique across kernel mounts. It would save one field and losest_dev, which is what tells mounts apart.Nodeas the identity. ANodeis minted at an open and gone at the last close, so one file opened twice in turn would be two files.Rust std.
std::os::toyos::fs::MetadataExt { fn dev(&self) -> u64; fn ino(&self) -> u64; }uses Unix's names and meanings, implemented overFileAttr::{volume, node}. The fork needs no change. std's ToyOS backend issdk/std, which the fork'slibrary/stdincludes by#[path]from each worktree's own tree (src/sysroot.rs), so this branch builds its own std.What changed
toyos-abi:Stat::{volume, node};fstatstarts fromStat::default().OpenFileState::volume, whichops::opentakes from the sameMountPointits open resolves (Vfs::open_targetandVfs::create_fileanswer(FileId, volume), withMountPoint::volumeas filesystem index plus one).ops::fstatanswers the volume and theFileIdfor a file, and 0 for every other object.toyos::fs:SERVED_VOLUME;Reply::node;Dirkeeps the connection's volume fromHELLOwith bit 63 set;Stat::{volume, node}.fileserver:Volume::identityandIdentities<K>. DATA keys by path. FAT keys by entry offset, through a newtoyos_fat32::Fat32::entry_offset, which makessame_entry's identity public.Absentkeys by position. The server answers the identity inOPEN,STAT/LSTATandFSTAT; a fid keeps the identity it was opened with, which renames keep too. InHELLOthe server answers its pid as the volume.fstat, and sostat, fillst_devandst_ino.FileAttr::{volume, node}, from both the kernel's stat and a server's;MetadataExt.DirEntry::metadatais anlstatof the entry, andDirEntryloses its unused size.Tests and why each tier
cargo test -p fileserver, inside--ci host):Identitieskeeps a number across a move, never hands out a removed or displaced number again, and moves a directory's keys with it without touching keys that only share its prefix.tests/testcases/tinycc/210_stat_identity.c,tests/toyos-rust-tests/src/bin/stat_identity.rs). Both are shared-boot members, which this tree runs on the T14 (shared_metal), the tier before QEMU; the QEMU suite does not run them./tmpfiles differ. One file matches through two spellings and through its descriptor, before and after a rename./system/bin/cat(a link) is the same file as/system/bin/toybox, andtoyboxis notsupervisor./systemand/tmpare two volumes.MetadataExton/tmp, DATA (/home) and FAT (/log). Two files differ. Each of these answers the same pair as the file: a path, a symlink (except on FAT, which has none), an upper-case spelling (FAT), the open handle, the listing's entry, and a rename. Two served directories differ. The three mounts are three volumes; the served ones have bit 63 set and the kernel's does not.SYS_FSTATanswer through libc. The second is a served file's pair crossing the file server, the wire and std. Both exist only in a running system.cc0315601:192491965plus only the case, its.expectand the LICENSE count). It is red on the T14: judge EXIT=1, 0 passed, 1 failed, and the case itself exited 2,ccheck'sCASE_FAILED, as when the case returns 1 because libc zeroes both words.mainhas not moved from192491965, so this control stands for de60926. The Rust test cannot be built on the base at all (there is noMetadataExtand noSERVED_VOLUME).e3c0c50ac, which is de60926 plus two checked patches, in the throwaway worktree../toyos-statid-mut:stat_identitymust fail on "two mounts are one volume".210_stat_identitymust fail on "/system and /tmp are two volumes".fat::tests::a_file_is_one_number_through_spellings_and_renamesred on its last assertion. There, the file made later in a freed slot answers 3, the number ofc.txt, which the rename displaced. That is the defect the review described, so the existing test already forces it, and no new test was added.hosted_clang_hello(below). Before this branch, itsFileManagerhanded backhello.cforstdio.h(hosted_clang_hello: ToyOS's own clang and ld.lld compile and link a C program in a guest, byte-identical to the host's, on demand #830's recorded failure). With this branch, clang readsstdio.hand stops at the next problem.Gates
At head de60926:
cargo run -- --ci hostcargo test --test toyos-buildcargo test --test toyos-build -- --metal --metal-readback <scratch>/statid-r2/metal-head stat_identitye3c0c50ac,… metal-mut stat_identityAt be007c6, whose FAT, DATA and
Identitiescode de60926 leaves unchanged:run.sh m6cc0315601)In the posted logs, the home directory's user name is replaced by
<user>. In the g1+g2 staging log, git's progress carriage returns are turned into newlines. Nothing else is changed.#830's
hosted_clang_hellowith this branch merged (evidence only, at be007c6)46a27b925(#830's local head) was merged with this branch in a throwaway worktree (fe5ffc5f2, not pushed; the worktree is removed) and run withcargo test --test toyos-build -- hosted_clang_hello, exit 1. It built the hosted clang and LLD against this branch's sysroot (key23a68a7033e0f028). In the guest, clang got past the old failure: it now readsstdio.h, nothello.c. It stopped at the include after that:Log: parts 1 2 3 4 5 6 7.
A hypothesis, not measured.
stddef.his in clang's resource directory,/system/share/hosted-clang/lib/clang/<v>/include, which is a directory on ROOT, and every ROOT directory still answers(0, 0). clang'sFileManager::getDirectoryRefkeys directories on the same pair, so every directory is oneDirectoryEntryto it. Header search's duplicate removal keys search directories on that entry, which would drop the resource directory as a duplicate ofsysroot/include. That is the remainder this branch records, and #830's next blocker.What I am unsure of
e3c0c50ac0 passed, 2 failed, red as required.stddef.hcause above is read from clang's source, not measured.DirEntry::metadatais now one stat per entry. A walker that asks it of every entry pays a round trip each (served) or four syscalls each (kernel). Before, it paid nothing and got a wrong mtime.by_pathkeys open files by the spelling they were opened with (unchanged here), so one file opened asAand asais two nodes. Their identity is one, keyed by entry offset.issues/fstats-answer-is-declared-twice.mdstill says the kernel's copy is "threeu64s". After this branch it is five, and the branch grows both hand-kept declarations without the size and offset assertion that is that issue's exit. The issue is left as it is, because this branch does not take it up and no rule names it, and a pull request edits no other existing issue.Net lines: 22 files, +642 −102. Of the additions, 167 are the two guest members and their
.expect, and about 115 are host tests.T14 RUN REQUESTED:
<scratch>/statid-r2/metal-head/request.txt(head de60926) and<scratch>/statid-r2/metal-mut/request.txt(g1+g2, from the worktree../toyos-statid-mut, kept for it).🤖 Generated with Claude Code
https://claude.ai/code/session_017cSFvbD35xJ2kGANVdm23C