add ptsname_r shim for macos#4
Conversation
on macos provided bash, by default it outputs; The default interactive shell is now zsh. To update your account to use zsh, please run `chsh -s /bin/zsh`.
|
Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA). View this failed invocation of the CLA check for more information. For the most up to date status, view the checks section at the bottom of the pull request. |
ethanpailes
left a comment
ethanpailes
left a comment
There was a problem hiding this comment.
Thank you so much for this patch!
Overall, it looks great, I mostly just have feedback about getting the change to follow project standards when it comes to saftey.
| if let Some(fd) = self.pty { | ||
| // Safety: the vector's memory is valid for the duration | ||
| // of the call | ||
| unsafe { |
There was a problem hiding this comment.
I'd like to maintain the style that every unsafe block has a saftey comment above it.
There was a problem hiding this comment.
I mistakenly moved this one, updated; 559aa75
| //! Based on: https://tarq.net/posts/ptsname-on-osx-with-rust/ | ||
|
|
||
| #[cfg(any(target_os = "macos"))] | ||
| pub unsafe fn ptsname_r( |
There was a problem hiding this comment.
Please add a "Saftey" comment that explains the formal precondtions that any call site must meet in order to use this function correctly (should be pretty easy, just stuff like "buf needs to point to allocated memory" and "buflen can't be longer than the allocation" and "fd must be an open file descriptor").
| return *libc::__error(); | ||
| } | ||
|
|
||
| let mut len = 0; |
There was a problem hiding this comment.
Totally fine to leave it like this, but for this sort of null-seaking, memchr (either the pure rust crate https://crates.io/crates/memchr, or the libc function https://docs.rs/libc/0.2.177/libc/fn.memchr.html) will do this a lot faster because they take advantage of SIMD hardware acceleration. It definitely won't matter for performance since the strings are so small here, so this is mostly just me being nerd sniped.
There was a problem hiding this comment.
TIL thank you. To not bring a new dependency, went for the libc one; 3bca717
| return libc::ERANGE; | ||
| } | ||
|
|
||
| libc::memcpy( |
There was a problem hiding this comment.
Let's use std::ptr::copy (which is how you pronounce libc::memmove in rust) here. It is safer because it can handle overlapping ranges, and on modern branch predicting CPUs the extra branch it requires is undetectable in microbenchmarks.
| let master_fd = unsafe { libc::posix_openpt(libc::O_RDWR | libc::O_NOCTTY) }; | ||
| assert!(master_fd >= 0, "Failed to open master PTY"); | ||
|
|
||
| unsafe { |
There was a problem hiding this comment.
Please add a Saftey comment. I know it's just a test, but having them on every block makes auditing easier.
There was a problem hiding this comment.
Updated each unsafe section in the tests, not sure about the wording though. How all these sounds like? 559aa75
|
|
||
| let mut buf = [0u8; 2]; | ||
| let result = | ||
| unsafe { ptsname_r(master_fd, buf.as_mut_ptr() as *mut libc::c_char, buf.len()) }; |
| #[test] | ||
| fn test_ptsname_r_invalid_fd() { | ||
| let mut buf = vec![0u8; 1024]; | ||
| let result = unsafe { ptsname_r(-1, buf.as_mut_ptr() as *mut libc::c_char, buf.len()) }; |
| #[cfg(any(target_os = "macos"))] | ||
| #[test] | ||
| fn test_ptsname_r_null_buffer() { | ||
| let master_fd = unsafe { libc::posix_openpt(libc::O_RDWR | libc::O_NOCTTY) }; |
| let master_fd = unsafe { libc::posix_openpt(libc::O_RDWR | libc::O_NOCTTY) }; | ||
|
|
||
| if master_fd >= 0 { | ||
| unsafe { |
| let result = unsafe { ptsname_r(master_fd, std::ptr::null_mut(), 1024) }; | ||
|
|
||
| unsafe { | ||
| libc::close(master_fd); |
|
Looks great. Thanks! Looks like there is a bit of lint bitrot (https://github.com/shell-pool/shpool_pty/actions/runs/19173539720/job/54819707170?pr=4) that needs fixing and then we can merge. |
|
If you don't have access to that lint output for some reason, you should be able to repro it by running |
They all should be fixed now. |
|
Not sure why that test is spinning. At this point is seems like it is going to time out. |
I think I broke something. On macOS tests pass but when I try to run on a Linux vm it just hangs. |
|
I think you might want to abort the run. It seems to be due to replacing |
|
I'm kinda supprised it ran for over an hour without timing out. |
|
My guess is |
| let mut cmd = Command::new("bash"); | ||
| cmd.env_clear(); | ||
|
|
||
| // On macOS, silence the deprecation warning; | ||
| // https://github.com/apple-oss-distributions/bash/blob/e86b2aa8e37a31f8fce56366d1abaf08a3fac7d2/bash-3.2/shell.c#L760-L765 | ||
| #[cfg(target_os = "macos")] | ||
| { | ||
| cmd.env("BASH_SILENCE_DEPRECATION_WARNING", "1"); | ||
| } |
There was a problem hiding this comment.
Looks like /bin/sh is just bash in disguise on macOS. Would it be OK to stick to bash and just silence this annoying warning?
There was a problem hiding this comment.
Yeah, that seems fine to me.
|
Found my way here from shell-pool/shpool#48 Looks like most if not all the feedback has been taken up. Do you guys need any help testing this? It would be great if @seruman wouldn't have to maintain a fork for folks like me! |
| let mut cmd = Command::new("bash"); | ||
| cmd.env_clear(); | ||
|
|
||
| // On macOS, silence the deprecation warning; | ||
| // https://github.com/apple-oss-distributions/bash/blob/e86b2aa8e37a31f8fce56366d1abaf08a3fac7d2/bash-3.2/shell.c#L760-L765 | ||
| #[cfg(target_os = "macos")] | ||
| { | ||
| cmd.env("BASH_SILENCE_DEPRECATION_WARNING", "1"); | ||
| } |
There was a problem hiding this comment.
Yeah, that seems fine to me.
|
@seruman in the future, when you finish responding to a review, you can click the little recycle icon on the upper right of the PR page to notify the reviewer that you are done responding and request a re-review. 
|
|
@ethanpailes thanks for the heads up, sorry that i totally forgot this PR. |
No worries, I forgot about it too. Thank you for the patch! I'll cut a new release so we can pick up the change in the main shpool repo. |
|
I got a little too eager on merging it looks like. I posted #5 as a followup. |
|
@seruman I just cut and pushed 0.3.2 to crates.io (https://crates.io/crates/shpool_pty) with this change. If you have a macos port for the main shpool repo that you want to merge in I would be happy to help you with that. |
thank you, yes i do. i'm a little caught up though, it might take some time before getting to it. |
|
Awesome! No rush |
This version includes macOS support via ptsname (vs ptsname_r). See: shell-pool/shpool_pty#4
This version includes macOS support via ptsname (vs ptsname_r). See: shell-pool/shpool_pty#4
This version includes macOS support via ptsname (vs ptsname_r). See: shell-pool/shpool_pty#4
This version includes macOS support via ptsname (vs ptsname_r). See: shell-pool/shpool_pty#4
This patch adds support for building and running on macos. As of this change, most of the test suite passes on macos, with the exception of some shell tests that just need to be tweaked and the motd pager tests, which have deeper issues. Full log of squashed commit messages: * use platform agnostic wather instead of inotify * add macos/bsd specific passwd fields * use std::env::current_exe instead of proc fs * add macos specific peer creds handling possibly the same in BSD * use getpeereid instead of LOCAL_PEERCRED socket opts basically the same according to manual * fix: use into_owned() to avoid double allocation * fix: canonicalize watched paths in ConfigWatcher File system watchers (inotify, FSEvents) report canonical paths, but ConfigWatcher was storing paths as provided by the caller. When a path contained symlinks (e.g., /var -> /private/var on macOS), event paths wouldn't match stored paths, causing config changes to be missed. Add canonicalize_path() to resolve symlinks in the existing portion of a path (handling paths where the final components don't exist yet), and apply it when storing paths in the watcher. Include a regression test that creates a symlink and verifies events are received when watching through the symlink. * fix: correct test synchronization in ConfigWatcher The worker_ready() test helper signals when the worker is idle, but it was signaling even when there was a pending reload_deadline. This could cause tests to proceed before the debounce timeout fired, leading to flaky test results. Only signal idle when there's no pending work (reload_deadline is None). Update the debounce test to not call worker_ready() between writes, as that now correctly waits for the reload to complete. * chore: bump shpool_pty to 0.3.2 This version includes macOS support via ptsname (vs ptsname_r). See: shell-pool/shpool_pty#4 * test: fix integration tests for macOS - Forward test_hooks feature from shpool to libshpool crate - Shorten socket paths to avoid macOS 104-byte sun_path limit - Forward HOME and PATH env vars to attach subprocess (macOS typically lacks XDG_RUNTIME_DIR, needs HOME instead) - Use std::env::var directly to avoid unused import warning - Document that tests should run with --test-threads=1 on macOS due to FSEvents timing behavior under concurrent load * fix: handle EINVAL on set_read_timeout for macOS macOS returns EINVAL when setting socket timeout on a connection where the peer has already closed (documented in setsockopt(2)). This typically happens with daemon presence probe connections. * fix: handle ENOTCONN on socket shutdown for macOS On macOS, calling shutdown() on a Unix socket after the peer has disconnected returns ENOTCONN (error 57). This was causing the shell->client thread to exit with an error, which then caused re-attachment to fail. Add a shutdown_socket() helper that ignores ENOTCONN errors since the socket is already closed in that case. * test: fix race condition in move_multiple_levels_in_place Add worker_ready() synchronization before the rename operation, matching the pattern used in other config_watcher tests. Without this, the filesystem event from the rename could occur before the watcher is fully ready to receive it, causing flaky test failures on macOS where FSEvents timing differs from Linux's inotify. * test: fix lines_big_chunk_restore race condition * Wait for daemon-wrote-s2c-chunk event before sending the large command This avoids a race where the command is sent before the shell is ready to receive input. * Silence macOS bash deprecation warning via BASH_SILENCE_DEPRECATION_WARNING env var The warning contains 'd' which caused read_until(b'd') to stop early before reaching "food" in the restore buffer. * test: skip known-failing tests on macOS Skip 6 tests that fail on macOS due to platform differences: - prompt_prefix_zsh, prompt_prefix_fish: hard-coded Linux shell paths - motd_pager, motd_debounced_pager_*, motd_env_test_pager_*: PTY pager output doesn't reach client on macOS Update HACKING.md to document skipped tests. * docs: add safety comments for unsafe FFI calls * style: apply cargo fmt nightly formatting --------- Co-authored-by: Selman Kayrancioglu <selmankayrancioglu@gmail.com>
3 participants
Related to; shell-pool/shpool#183
I was not able to build shell-pool/shpool;
With a quick search I found these references:
This PR adds a small macOS-only shim so that from the caller’s perspective, it still looks like
ibc::ptsname_r, to make as little as change possible.I have little to no experience in Rust or systems programming, so this change
may not be idiomatic or does not even make sense in the first place so feel
free to update/feedback or even discard.
I've been using it on on my shpool fork for couple some time just fine.
Disclaimer; got help from LLM tools, especially for tests.
Had to change tests to use
shinstead ofbashfor portability as on my version of macOS -26.0.1-, Apple providedbashgreets you with;And seems like it is baked in.