Feat/macos port clean#202
Open
dlicudi wants to merge 19 commits into
Open
Conversation
Spike proving the live fuse3 backend works on macOS via macFUSE, so no fuser port is needed: - system/filesystem: widen statvfs block counts (u32 on macOS) to u64 - fuse3/shared: add macOS-only crtime/flags fields to FileAttr literals - fuse3/filesystem: implement open/release; macFUSE lacks FUSE_NO_OPEN_SUPPORT and forwards ENOSYS from open to read() - add ignored macos_mount_spike integration test (mount, read, readdir)
- add system::unmount::unmount_fuse: platform-abstracted unmount (fusermount on Linux, umount -f on macOS) used by both the panic handler and the SpawnedMountHandle Drop fallback, replacing the duplicated fusermount shell-outs - verify the production Fuse3OrthoUnionFS mounts on macFUSE end-to-end (mount, read scenery file, readdir, unmount) in macos_mount_spike - cover unmount_fuse with a fast non-mountpoint unit test The async MountHandle::unmount path remains the normal (fast) teardown; the shell-out helper is only the panic/Drop fallback, so macOS force- detaches rather than risk blocking.
macFUSE rejects some read replies with EINVAL under load; the previously pinned fuse3 commit treated that as fatal and tore down the mount. Point at dlicudi/fuse3 fix/macos-reply-error-resilience, which is samsoir's configurable-background plus a patch that drops the bad reply and keeps the session alive.
is_mounted() only read /proc/mounts, which doesn't exist on macOS, so it always returned false and the umount fallback in unmount_sync never ran. The fuse3 task completing also doesn't prove macFUSE released the mount. Make is_mounted() platform-correct (mount(8) on macOS) and always verify + escalate to umount -f rather than trusting task completion. Ends the stale-mount that blocked every restart.
BSD split on macOS supports neither --version nor --bytes, causing 'split command failed' panics in the publisher on stock macOS. Probe availability with a portable no-op (split -b 1 /dev/null) and invoke with -b <size>, which both GNU and BSD variants accept. Part naming (.aa/.ab suffixes) is identical across variants, so manifests are unaffected. Fixes test_build_archive_splits_large_file on macOS.
The X-Plane installer writes x-plane_install_12.txt to ~/Library/Preferences on macOS, not ~/.x-plane (Linux). Auto-detection of the installation and Custom Scenery directory silently failed on macOS, so 'packages install' could not create overlay symlinks without a manually configured custom_scenery_path. Add a macos cfg branch to get_install_reference_path() and update the config docs that hardcoded the Linux path.
The run command only read packages.custom_scenery_path and xplane.scenery_dir from config, while the packages commands also fall back to detect_custom_scenery(). With both keys unset, run failed with a configuration error even when the X-Plane installation is discoverable. Add the same auto-detect fallback for consistency. Required for out-of-the-box startup on macOS, where the install reference file detection was previously broken (see prior commit) and setup left both keys empty.
Replace the mutable branch reference to dlicudi/fuse3 with an immutable rev pin. The branch was rebased onto upstream master (which now includes samsoir's configurable-background PR samsoir#136), carrying the single reply-resilience patch pending as an upstream PR. Required for the macOS port: macFUSE can transiently reject reply writes under load; without the patch the whole session tears down.
A run that dies without unmounting (crash, SIGKILL, system sleep) leaves a dead mount on the mountpoint; the next mount attempt fails with ENXIO 'Device not configured' and the user must umount by hand. Detect the corpse before mounting and force-unmount it. Detection must use opendir(2), not stat(2): macOS serves cached attributes for a dead macFUSE mountpoint, so stat succeeds while reading the directory fails with ENXIO (verified against a real killed mount; a metadata()-based probe silently missed it). Linux's equivalent dead-mount errno ENOTCONN is classified too. Also make the mount-failure hints platform-aware: the previous text suggested apt/fusermount on macOS; now it points at macFUSE install, Privacy & Security approval, and umount. Verified end-to-end on macOS: mount, SIGKILL, restart -> WARN 'Stale FUSE mount detected from a previous run; force-unmounting' -> healthy mount.
The reply-resilience patch is now submitted upstream as Sherlock-Holo/fuse3#137. Once merged, repoint the dependency to Sherlock-Holo/fuse3 (which also retires main's samsoir/fuse3 branch pin, merged upstream as samsoir#136).
Update platform support in README and CLAUDE.md: macOS works with macFUSE (Apple Silicon tested, multi-hour X-Plane flights). Notes the fuse3 fork rev pin and its pending upstream PR (Sherlock-Holo/fuse3#137).
Owner
|
Thanks for this, a huge high-quality contribution that I would like progress towards main safely. The macOS-specific reasoning is exactly right (the Process: This is shaping up as a 0.5.0 line, and we still have 0.4.x work to finish on main, so I'm going to spin up a 0.5 integration branch and I'll ask you to retarget this PR onto it. I'll also get CI running on it (it didn't trigger here) and add a macOS job, since right now none of the Must-fix:
Should-fix for "complete" macOS support:
Dependency note: nice work getting onto upstream-master + a single patch. That's a better position than our current samsoir/fuse3 pin. One ask: so we're not depending on a personal fork that could move, I'll mirror 849e84c into our org fork (or we wait for #137 upstream) before this ships. Personal note: I really appreciate this contribution. If we can address the points above, macOS support can be a major component of the 0.5.0 release. I am still traveling until early July, so unable to do any local testing until I return. When I am back home I will start laying the ground for 0.5.x development. |
Owner
|
Quick housekeeping note: I edited the PR description to qualify the fuse3 references. The fuse3 bullet's bare The same bare refs still live in two commit messages ( |
Fuse3PassthroughFS::open() hardcoded FOPEN_DIRECT_IO for virtual inodes (with a locally redefined constant), bypassing the platform-aware VIRTUAL_DDS_OPEN_FLAGS that Fuse3OrthoUnionFS already used. On macOS that reintroduces the macFUSE mmap crash: X-Plane memory-maps DDS textures and faults with EXC_BAD_ACCESS when a direct_io file is mapped, since direct_io has no page cache for mmap to use. Lift FOPEN_DIRECT_IO and VIRTUAL_DDS_OPEN_FLAGS into shared.rs as the single source of truth and have both filesystems report it from open(), so the macOS page-cache-not-direct_io rule cannot drift between mounts. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Mount detection was implemented twice: SpawnedMountHandle::is_mounted (cross-platform, ported in this branch) and manager's check_mount_status (Linux-only, returned Unknown on macOS). So `packages list` always showed "Unknown" mount status on macOS even for correctly mounted packages. Extract one platform-aware is_path_mounted() into system::unmount alongside unmount_fuse / is_stale_fuse_mount (the platform mount-table module), move the pure /proc/mounts and mount(8) parsers + their tests there, and have both SpawnedMountHandle::is_mounted and check_mount_status delegate to it. macOS now reports real Mounted/NotMounted status instead of Unknown. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
fuse3 instruments every FUSE handler with an INFO span, so at the default `info` level it emits a log line per syscall. On macOS, Finder / Spotlight / the open-save panel constantly stat the mount, and combined with the debug build's FmtSpan::CLOSE + pretty() format this floods the log at ~10 MB/s (observed: a 187 GB xearthlayer.log) and burns ~1.5 cores formatting it. Pin `fuse3=warn` in the default and --debug filter arms (the --profile arms already did this), matching the established pattern. fuse3 goes quiet while xearthlayer's own logs are untouched; RUST_LOG still overrides. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
dlicudi
force-pushed
the
feat/macos-port-clean
branch
from
91c9f01 to
8ae9839
Compare
detect_total_memory() only parsed /proc/meminfo (Linux); on macOS it fell through to the 8GB fallback. The setup wizard sizes the memory cache as RAM/12, so every Mac got memory_size ≈ 682 MB regardless of actual RAM — a quietly undersized config (e.g. a 48 GB Mac should get ~4 GB). Add a macOS arm reading hw.memsize via sysctl(8), with a pure, unit-tested parser. Other platforms keep the 8GB fallback. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
detect_storage_type() was Linux-only (/sys/block rotational); on macOS it warned and returned None, so Auto always fell back to the SSD profile — under- using NVMe (64 vs 256 concurrent I/O) on every Apple Silicon Mac. Add a macOS arm: resolve the cache path to its backing device with df, query diskutil info, and map Solid State + Protocol to a profile (Apple Fabric / PCI-Express / NVMe -> Nvme, other SSD -> Ssd, spinning -> Hdd). Output parsing is split into pure, unit-tested functions; unclassifiable devices return None so the caller keeps the safe SSD default. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The diagnostics report's collectors read Linux-only sources (/proc/cpuinfo, /proc/meminfo, /etc/os-release, `ip route`), so on macOS the Operating System, Hardware, and Network sections came out blank — useless for Mac bug reports. Add macOS branches: CPU/memory via sysctl, OS name via sw_vers, default interface via `route -n get default`. Output parsing is a pure, unit-tested function; Linux collectors are unchanged. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ixes Repoint the dlicudi/fuse3 pin from the reply-resilience rev (849e84c) to feat/macos-mount-options (74abf55), which adds two macOS fixes on top: - comma-join mount options so all `-o` flags actually apply on macFUSE (the builder dropped every option after the first) - log recoverable reply drops at debug, not warn — macOS file browsers (Finder) crawl the mount and produce a harmless ~40/sec stream of "dropping a failed fuse reply" warnings; demoting to debug silences the flood while the session-continues handling is unchanged. Verified on macOS: reply-drop log rate 0/sec after the bump. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Author
|
@samsoir just a quick note, pushed a few more changes to this branch, but still work in progress. Looking forward to 0.5 branch, I can rebase this macOS port branch to 0.5 once it's available. For the fuse3 fork, I agree best to rely on your fork; my experimental branch is only pinned to my fuse3 fork for purposes of development/testing. Will post again once I've had a chance to pick this up again and review remaining items. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Ports XEarthLayer to macOS. The branch is strictly portability: no feature
changes, no behavior changes on Linux. Tested on Apple Silicon (M5 Max,
macOS 26, macFUSE 5.2) against stable X-Plane 12.4.2 with multi-hour
flights and no session losses.
Developed with AI assistance (Claude); all changes flight-tested by me.
Changes
mount verification (
system/unmount.rs)SIGKILL leaves a dead mount; detection must use opendir, not stat,
because macOS serves cached attributes for dead macFUSE mountpoints)
on macOS instead of apt/fusermount)
(~/Library/Preferences/x-plane_install_12.txt)
splitinvocation in the publisher (probe and-b)runfalls back to Custom Scenery auto-detection like the packagescommands already did
submitted upstream as fix(macos): don't kill the session on a transient reply error Sherlock-Holo/fuse3#137 (rebased on upstream
master, which already includes feat(init): add configurable max_background and congestion_threshold to ReplyInit Sherlock-Holo/fuse3#136). Once
fix(macos): don't kill the session on a transient reply error Sherlock-Holo/fuse3#137 merges, one commit
repoints to upstream and also retires the samsoir/fuse3 branch pin
currently on main.
Related Issues
dependency, pending)
Test Plan
make pre-commitpasses (fmt + clippy + tests)xearthlayer run --airport <ICAO>mounts, streamstiles, and unmounts cleanly on Ctrl-C; 2h and 4h+ real X-Plane
flights without crashes
observe "Stale FUSE mount detected" warning and healthy remount
build and tests)
Checklist
cargo clippy