refactor: add support for atomic rename events on linux

This commit is contained in:
CJ van den Berg 2026-02-20 10:44:00 +01:00
parent 8ffcb0e732
commit 7672eb3c51
Signed by: neurocyte
GPG key ID: 8EB1E1BB660E3FB9

View file

@ -13,6 +13,8 @@ pub const EventType = enum {
created, created,
modified, modified,
deleted, deleted,
/// Only produced on macOS and Windows where the OS gives no pairing info.
/// On Linux, paired renames are emitted as a { "FW", "rename", from, to } message instead.
renamed, renamed,
}; };
@ -23,6 +25,7 @@ const SpawnError = error{ OutOfMemory, ThespianSpawnFailed };
/// Watch a path (file or directory) for changes. The caller will receive: /// Watch a path (file or directory) for changes. The caller will receive:
/// .{ "FW", "change", path, event_type } /// .{ "FW", "change", path, event_type }
/// where event_type is a file_watcher.EventType tag string: "created", "modified", "deleted", "renamed" /// where event_type is a file_watcher.EventType tag string: "created", "modified", "deleted", "renamed"
/// On Linux, paired renames produce: .{ "FW", "rename", from_path, to_path }
pub fn watch(path: []const u8) FileWatcherError!void { pub fn watch(path: []const u8) FileWatcherError!void {
return send(.{ "watch", path }); return send(.{ "watch", path });
} }
@ -120,17 +123,34 @@ const LinuxBackend = struct {
} }
} }
fn drain(self: *LinuxBackend, parent: tp.pid_ref) !void { fn drain(self: *LinuxBackend, allocator: std.mem.Allocator, parent: tp.pid_ref) !void {
const InotifyEvent = extern struct { const InotifyEvent = extern struct {
wd: i32, wd: i32,
mask: u32, mask: u32,
cookie: u32, cookie: u32,
len: u32, len: u32,
}; };
// A pending MOVED_FROM waiting to be paired with a MOVED_TO by cookie.
const PendingRename = struct {
cookie: u32,
path: []u8, // owned by drain's allocator
};
var buf: [4096]u8 align(@alignOf(InotifyEvent)) = undefined; var buf: [4096]u8 align(@alignOf(InotifyEvent)) = undefined;
var pending_renames: std.ArrayListUnmanaged(PendingRename) = .empty;
defer {
// Any unpaired MOVED_FROM means the file was moved out of the watched tree.
for (pending_renames.items) |r| {
parent.send(.{ "FW", "change", r.path, EventType.deleted }) catch {}; // moved outside watched tree
allocator.free(r.path);
}
pending_renames.deinit(allocator);
}
while (true) { while (true) {
const n = std.posix.read(self.inotify_fd, &buf) catch |e| switch (e) { const n = std.posix.read(self.inotify_fd, &buf) catch |e| switch (e) {
error.WouldBlock => return, error.WouldBlock => break,
else => return e, else => return e,
}; };
var offset: usize = 0; var offset: usize = 0;
@ -143,22 +163,50 @@ const LinuxBackend = struct {
std.mem.sliceTo(buf[name_offset..][0..ev.len], 0) std.mem.sliceTo(buf[name_offset..][0..ev.len], 0)
else else
""; "";
const event_type: EventType = if (ev.mask & IN.CREATE != 0)
.created var full_buf: [std.fs.max_path_bytes]u8 = undefined;
else if (ev.mask & (IN.DELETE | IN.DELETE_SELF) != 0) const full_path: []const u8 = if (name.len > 0)
.deleted try std.fmt.bufPrint(&full_buf, "{s}/{s}", .{ watched_path, name })
else if (ev.mask & (IN.MODIFY | IN.CLOSE_WRITE) != 0)
.modified
else if (ev.mask & (IN.MOVED_FROM | IN.MOVED_TO | IN.MOVE_SELF) != 0)
.renamed
else else
continue; watched_path;
if (name.len > 0) {
var full_buf: [std.fs.max_path_bytes]u8 = undefined; if (ev.mask & IN.MOVED_FROM != 0) {
const full_path = try std.fmt.bufPrint(&full_buf, "{s}/{s}", .{ watched_path, name }); // Park it, we may receive a paired MOVED_TO with the same cookie.
try parent.send(.{ "FW", "change", full_path, event_type }); try pending_renames.append(allocator, .{
.cookie = ev.cookie,
.path = try allocator.dupe(u8, full_path),
});
} else if (ev.mask & IN.MOVED_TO != 0) {
// Look for a paired MOVED_FROM.
var found: ?usize = null;
for (pending_renames.items, 0..) |r, i| {
if (r.cookie == ev.cookie) {
found = i;
break;
}
}
if (found) |i| {
// Complete rename pair: emit a single atomic rename message.
const r = pending_renames.swapRemove(i);
defer allocator.free(r.path);
try parent.send(.{ "FW", "rename", r.path, full_path });
} else {
// No paired MOVED_FROM, file was moved in from outside the watched tree.
try parent.send(.{ "FW", "change", full_path, EventType.created });
}
} else if (ev.mask & IN.MOVE_SELF != 0) {
// The watched directory itself was renamed/moved away.
try parent.send(.{ "FW", "change", full_path, EventType.deleted });
} else { } else {
try parent.send(.{ "FW", "change", watched_path, event_type }); const event_type: EventType = if (ev.mask & IN.CREATE != 0)
.created
else if (ev.mask & (IN.DELETE | IN.DELETE_SELF) != 0)
.deleted
else if (ev.mask & (IN.MODIFY | IN.CLOSE_WRITE) != 0)
.modified
else
continue;
try parent.send(.{ "FW", "change", full_path, event_type });
} }
} }
} }
@ -227,7 +275,8 @@ const MacosBackend = struct {
} }
} }
fn drain(self: *MacosBackend, parent: tp.pid_ref) !void { fn drain(self: *MacosBackend, allocator: std.mem.Allocator, parent: tp.pid_ref) !void {
_ = allocator;
var events: [64]std.posix.Kevent = undefined; var events: [64]std.posix.Kevent = undefined;
const immediate: std.posix.timespec = .{ .sec = 0, .nsec = 0 }; const immediate: std.posix.timespec = .{ .sec = 0, .nsec = 0 };
const n = std.posix.kevent(self.kq, &.{}, &events, &immediate) catch return; const n = std.posix.kevent(self.kq, &.{}, &events, &immediate) catch return;
@ -350,7 +399,8 @@ const WindowsBackend = struct {
} }
} }
fn drain(self: *WindowsBackend, parent: tp.pid_ref) !void { fn drain(self: *WindowsBackend, allocator: std.mem.Allocator, parent: tp.pid_ref) !void {
_ = allocator;
var bytes: windows.DWORD = 0; var bytes: windows.DWORD = 0;
var key: windows.ULONG_PTR = 0; var key: windows.ULONG_PTR = 0;
var overlapped_ptr: ?*windows.OVERLAPPED = null; var overlapped_ptr: ?*windows.OVERLAPPED = null;
@ -453,13 +503,13 @@ const Process = struct {
var err_msg: []const u8 = undefined; var err_msg: []const u8 = undefined;
if (try cbor.match(m.buf, .{ "fd", tp.extract(&tag), "read_ready" })) { if (try cbor.match(m.buf, .{ "fd", tp.extract(&tag), "read_ready" })) {
self.backend.drain(self.parent.ref()) catch |e| self.logger.err("drain", e); self.backend.drain(self.allocator, self.parent.ref()) catch |e| self.logger.err("drain", e);
self.backend.arm(); self.backend.arm();
} else if (try cbor.match(m.buf, .{ "fd", tp.extract(&tag), "read_error", tp.extract(&err_code), tp.extract(&err_msg) })) { } else if (try cbor.match(m.buf, .{ "fd", tp.extract(&tag), "read_error", tp.extract(&err_code), tp.extract(&err_msg) })) {
self.logger.print("fd read error on {s}: ({d}) {s}", .{ tag, err_code, err_msg }); self.logger.print("fd read error on {s}: ({d}) {s}", .{ tag, err_code, err_msg });
self.backend.arm(); self.backend.arm();
} else if (builtin.os.tag == .windows and try cbor.match(m.buf, .{"FW_poll"})) { } else if (builtin.os.tag == .windows and try cbor.match(m.buf, .{"FW_poll"})) {
self.backend.drain(self.parent.ref()) catch |e| self.logger.err("drain", e); self.backend.drain(self.allocator, self.parent.ref()) catch |e| self.logger.err("drain", e);
self.backend.arm(); self.backend.arm();
} else if (try cbor.match(m.buf, .{ "watch", tp.extract(&path) })) { } else if (try cbor.match(m.buf, .{ "watch", tp.extract(&path) })) {
self.backend.add_watch(self.allocator, path) catch |e| self.logger.err("watch", e); self.backend.add_watch(self.allocator, path) catch |e| self.logger.err("watch", e);