vendor/github.com/fsnotify/fsnotify/.mailmap

+Chris Howey <howeyc@gmail.com> <chris@howey.me>
+Nathan Youngman <git@nathany.com> <4566+nathany@users.noreply.github.com>

vendor/github.com/fsnotify/fsnotify/CHANGELOG.md

+# Changelog
+
+Unreleased
+----------
+Nothing yet.
+
+1.7.0 - 2023-10-22
+------------------
+This version of fsnotify needs Go 1.17.
+
+### Additions
+
+- illumos: add FEN backend to support illumos and Solaris. ([#371])
+
+- all: add `NewBufferedWatcher()` to use a buffered channel, which can be useful
+  in cases where you can't control the kernel buffer and receive a large number
+  of events in bursts. ([#550], [#572])
+
+- all: add `AddWith()`, which is identical to `Add()` but allows passing
+  options. ([#521])
+
+- windows: allow setting the ReadDirectoryChangesW() buffer size with
+  `fsnotify.WithBufferSize()`; the default of 64K is the highest value that
+  works on all platforms and is enough for most purposes, but in some cases a
+  highest buffer is needed. ([#521])
+
+### Changes and fixes
+
+- inotify: remove watcher if a watched path is renamed ([#518])
+
+  After a rename the reported name wasn't updated, or even an empty string.
+  Inotify doesn't provide any good facilities to update it, so just remove the
+  watcher. This is already how it worked on kqueue and FEN.
+
+  On Windows this does work, and remains working.
+
+- windows: don't listen for file attribute changes ([#520])
+
+  File attribute changes are sent as `FILE_ACTION_MODIFIED` by the Windows API,
+  with no way to see if they're a file write or attribute change, so would show
+  up as a fsnotify.Write event. This is never useful, and could result in many
+  spurious Write events.
+
+- windows: return `ErrEventOverflow` if the buffer is full ([#525])
+
+  Before it would merely return "short read", making it hard to detect this
+  error.
+
+- kqueue: make sure events for all files are delivered properly when removing a
+  watched directory ([#526])
+
+  Previously they would get sent with `""` (empty string) or `"."` as the path
+  name.
+
+- kqueue: don't emit spurious Create events for symbolic links ([#524])
+
+  The link would get resolved but kqueue would "forget" it already saw the link
+  itself, resulting on a Create for every Write event for the directory.
+
+- all: return `ErrClosed` on `Add()` when the watcher is closed ([#516])
+
+- other: add `Watcher.Errors` and `Watcher.Events` to the no-op `Watcher` in
+  `backend_other.go`, making it easier to use on unsupported platforms such as
+  WASM, AIX, etc. ([#528])
+
+- other: use the `backend_other.go` no-op if the `appengine` build tag is set;
+  Google AppEngine forbids usage of the unsafe package so the inotify backend
+  won't compile there.
+
+[#371]: https://github.com/fsnotify/fsnotify/pull/371
+[#516]: https://github.com/fsnotify/fsnotify/pull/516
+[#518]: https://github.com/fsnotify/fsnotify/pull/518
+[#520]: https://github.com/fsnotify/fsnotify/pull/520
+[#521]: https://github.com/fsnotify/fsnotify/pull/521
+[#524]: https://github.com/fsnotify/fsnotify/pull/524
+[#525]: https://github.com/fsnotify/fsnotify/pull/525
+[#526]: https://github.com/fsnotify/fsnotify/pull/526
+[#528]: https://github.com/fsnotify/fsnotify/pull/528
+[#537]: https://github.com/fsnotify/fsnotify/pull/537
+[#550]: https://github.com/fsnotify/fsnotify/pull/550
+[#572]: https://github.com/fsnotify/fsnotify/pull/572
+
+1.6.0 - 2022-10-13
+------------------
+This version of fsnotify needs Go 1.16 (this was already the case since 1.5.1,
+but not documented). It also increases the minimum Linux version to 2.6.32.
+
+### Additions
+
+- all: add `Event.Has()` and `Op.Has()` ([#477])
+
+  This makes checking events a lot easier; for example:
+
+	    if event.Op&Write == Write && !(event.Op&Remove == Remove) {
+	    }
+
+	Becomes:
+
+	    if event.Has(Write) && !event.Has(Remove) {
+	    }
+
+- all: add cmd/fsnotify ([#463])
+
+  A command-line utility for testing and some examples.
+
+### Changes and fixes
+
+- inotify: don't ignore events for files that don't exist ([#260], [#470])
+
+  Previously the inotify watcher would call `os.Lstat()` to check if a file
+  still exists before emitting events.
+
+  This was inconsistent with other platforms and resulted in inconsistent event
+  reporting (e.g. when a file is quickly removed and re-created), and generally
+  a source of confusion. It was added in 2013 to fix a memory leak that no
+  longer exists.
+
+- all: return `ErrNonExistentWatch` when `Remove()` is called on a path that's
+  not watched ([#460])
+
+- inotify: replace epoll() with non-blocking inotify ([#434])
+
+  Non-blocking inotify was not generally available at the time this library was
+  written in 2014, but now it is. As a result, the minimum Linux version is
+  bumped from 2.6.27 to 2.6.32. This hugely simplifies the code and is faster.
+
+- kqueue: don't check for events every 100ms ([#480])
+
+  The watcher would wake up every 100ms, even when there was nothing to do. Now
+  it waits until there is something to do.
+
+- macos: retry opening files on EINTR ([#475])
+
+- kqueue: skip unreadable files ([#479])
+
+  kqueue requires a file descriptor for every file in a directory; this would
+  fail if a file was unreadable by the current user. Now these files are simply
+  skipped.
+
+- windows: fix renaming a watched directory if the parent is also watched ([#370])
+
+- windows: increase buffer size from 4K to 64K ([#485])
+
+- windows: close file handle on Remove() ([#288])
+
+- kqueue: put pathname in the error if watching a file fails ([#471])
+
+- inotify, windows: calling Close() more than once could race ([#465])
+
+- kqueue: improve Close() performance ([#233])
+
+- all: various documentation additions and clarifications.
+
+[#233]: https://github.com/fsnotify/fsnotify/pull/233
+[#260]: https://github.com/fsnotify/fsnotify/pull/260
+[#288]: https://github.com/fsnotify/fsnotify/pull/288
+[#370]: https://github.com/fsnotify/fsnotify/pull/370
+[#434]: https://github.com/fsnotify/fsnotify/pull/434
+[#460]: https://github.com/fsnotify/fsnotify/pull/460
+[#463]: https://github.com/fsnotify/fsnotify/pull/463
+[#465]: https://github.com/fsnotify/fsnotify/pull/465
+[#470]: https://github.com/fsnotify/fsnotify/pull/470
+[#471]: https://github.com/fsnotify/fsnotify/pull/471
+[#475]: https://github.com/fsnotify/fsnotify/pull/475
+[#477]: https://github.com/fsnotify/fsnotify/pull/477
+[#479]: https://github.com/fsnotify/fsnotify/pull/479
+[#480]: https://github.com/fsnotify/fsnotify/pull/480
+[#485]: https://github.com/fsnotify/fsnotify/pull/485
+
+## [1.5.4] - 2022-04-25
+
+* Windows: add missing defer to `Watcher.WatchList` [#447](https://github.com/fsnotify/fsnotify/pull/447)
+* go.mod: use latest x/sys [#444](https://github.com/fsnotify/fsnotify/pull/444)
+* Fix compilation for OpenBSD [#443](https://github.com/fsnotify/fsnotify/pull/443)
+
+## [1.5.3] - 2022-04-22
+
+* This version is retracted. An incorrect branch is published accidentally [#445](https://github.com/fsnotify/fsnotify/issues/445)
+
+## [1.5.2] - 2022-04-21
+
+* Add a feature to return the directories and files that are being monitored [#374](https://github.com/fsnotify/fsnotify/pull/374)
+* Fix potential crash on windows if `raw.FileNameLength` exceeds `syscall.MAX_PATH` [#361](https://github.com/fsnotify/fsnotify/pull/361)
+* Allow build on unsupported GOOS [#424](https://github.com/fsnotify/fsnotify/pull/424)
+* Don't set `poller.fd` twice in `newFdPoller` [#406](https://github.com/fsnotify/fsnotify/pull/406)
+* fix go vet warnings: call to `(*T).Fatalf` from a non-test goroutine [#416](https://github.com/fsnotify/fsnotify/pull/416)
+
+## [1.5.1] - 2021-08-24
+
+* Revert Add AddRaw to not follow symlinks [#394](https://github.com/fsnotify/fsnotify/pull/394)
+
+## [1.5.0] - 2021-08-20
+
+* Go: Increase minimum required version to Go 1.12 [#381](https://github.com/fsnotify/fsnotify/pull/381)
+* Feature: Add AddRaw method which does not follow symlinks when adding a watch [#289](https://github.com/fsnotify/fsnotify/pull/298)
+* Windows: Follow symlinks by default like on all other systems [#289](https://github.com/fsnotify/fsnotify/pull/289)
+* CI: Use GitHub Actions for CI and cover go 1.12-1.17
+   [#378](https://github.com/fsnotify/fsnotify/pull/378)
+   [#381](https://github.com/fsnotify/fsnotify/pull/381)
+   [#385](https://github.com/fsnotify/fsnotify/pull/385)
+* Go 1.14+: Fix unsafe pointer conversion [#325](https://github.com/fsnotify/fsnotify/pull/325)
+
+## [1.4.9] - 2020-03-11
+
+* Move example usage to the readme #329. This may resolve #328.
+
+## [1.4.8] - 2020-03-10
+
+* CI: test more go versions (@nathany 1d13583d846ea9d66dcabbfefbfb9d8e6fb05216)
+* Tests: Queued inotify events could have been read by the test before max_queued_events was hit (@matthias-stone #265)
+* Tests:  t.Fatalf -> t.Errorf in go routines (@gdey #266)
+* CI: Less verbosity (@nathany #267)
+* Tests: Darwin: Exchangedata is deprecated on 10.13 (@nathany #267)
+* Tests: Check if channels are closed in the example (@alexeykazakov #244)
+* CI: Only run golint on latest version of go and fix issues (@cpuguy83 #284)
+* CI: Add windows to travis matrix (@cpuguy83 #284)
+* Docs: Remover appveyor badge (@nathany 11844c0959f6fff69ba325d097fce35bd85a8e93)
+* Linux: create epoll and pipe fds with close-on-exec (@JohannesEbke #219)
+* Linux: open files with close-on-exec (@linxiulei #273)
+* Docs: Plan to support fanotify (@nathany ab058b44498e8b7566a799372a39d150d9ea0119 )
+* Project: Add go.mod (@nathany #309)
+* Project: Revise editor config (@nathany #309)
+* Project: Update copyright for 2019 (@nathany #309)
+* CI: Drop go1.8 from CI matrix (@nathany #309)
+* Docs: Updating the FAQ section for supportability with NFS & FUSE filesystems (@Pratik32 4bf2d1fec78374803a39307bfb8d340688f4f28e )
+
+## [1.4.7] - 2018-01-09
+
+* BSD/macOS: Fix possible deadlock on closing the watcher on kqueue (thanks @nhooyr and @glycerine)
+* Tests: Fix missing verb on format string (thanks @rchiossi)
+* Linux: Fix deadlock in Remove (thanks @aarondl)
+* Linux: Watch.Add improvements (avoid race, fix consistency, reduce garbage) (thanks @twpayne)
+* Docs: Moved FAQ into the README (thanks @vahe)
+* Linux: Properly handle inotify's IN_Q_OVERFLOW event (thanks @zeldovich)
+* Docs: replace references to OS X with macOS
+
+## [1.4.2] - 2016-10-10
+
+* Linux: use InotifyInit1 with IN_CLOEXEC to stop leaking a file descriptor to a child process when using fork/exec [#178](https://github.com/fsnotify/fsnotify/pull/178) (thanks @pattyshack)
+
+## [1.4.1] - 2016-10-04
+
+* Fix flaky inotify stress test on Linux [#177](https://github.com/fsnotify/fsnotify/pull/177) (thanks @pattyshack)
+
+## [1.4.0] - 2016-10-01
+
+* add a String() method to Event.Op [#165](https://github.com/fsnotify/fsnotify/pull/165) (thanks @oozie)
+
+## [1.3.1] - 2016-06-28
+
+* Windows: fix for double backslash when watching the root of a drive [#151](https://github.com/fsnotify/fsnotify/issues/151) (thanks @brunoqc)
+
+## [1.3.0] - 2016-04-19
+
+* Support linux/arm64 by [patching](https://go-review.googlesource.com/#/c/21971/) x/sys/unix and switching to to it from syscall (thanks @suihkulokki) [#135](https://github.com/fsnotify/fsnotify/pull/135)
+
+## [1.2.10] - 2016-03-02
+
+* Fix golint errors in windows.go [#121](https://github.com/fsnotify/fsnotify/pull/121) (thanks @tiffanyfj)
+
+## [1.2.9] - 2016-01-13
+
+kqueue: Fix logic for CREATE after REMOVE [#111](https://github.com/fsnotify/fsnotify/pull/111) (thanks @bep)
+
+## [1.2.8] - 2015-12-17
+
+* kqueue: fix race condition in Close [#105](https://github.com/fsnotify/fsnotify/pull/105) (thanks @djui for reporting the issue and @ppknap for writing a failing test)
+* inotify: fix race in test
+* enable race detection for continuous integration (Linux, Mac, Windows)
+
+## [1.2.5] - 2015-10-17
+
+* inotify: use epoll_create1 for arm64 support (requires Linux 2.6.27 or later) [#100](https://github.com/fsnotify/fsnotify/pull/100) (thanks @suihkulokki)
+* inotify: fix path leaks [#73](https://github.com/fsnotify/fsnotify/pull/73) (thanks @chamaken)
+* kqueue: watch for rename events on subdirectories [#83](https://github.com/fsnotify/fsnotify/pull/83) (thanks @guotie)
+* kqueue: avoid infinite loops from symlinks cycles [#101](https://github.com/fsnotify/fsnotify/pull/101) (thanks @illicitonion)
+
+## [1.2.1] - 2015-10-14
+
+* kqueue: don't watch named pipes [#98](https://github.com/fsnotify/fsnotify/pull/98) (thanks @evanphx)
+
+## [1.2.0] - 2015-02-08
+
+* inotify: use epoll to wake up readEvents [#66](https://github.com/fsnotify/fsnotify/pull/66) (thanks @PieterD)
+* inotify: closing watcher should now always shut down goroutine [#63](https://github.com/fsnotify/fsnotify/pull/63) (thanks @PieterD)
+* kqueue: close kqueue after removing watches, fixes [#59](https://github.com/fsnotify/fsnotify/issues/59)
+
+## [1.1.1] - 2015-02-05
+
+* inotify: Retry read on EINTR [#61](https://github.com/fsnotify/fsnotify/issues/61) (thanks @PieterD)
+
+## [1.1.0] - 2014-12-12
+
+* kqueue: rework internals [#43](https://github.com/fsnotify/fsnotify/pull/43)
+    * add low-level functions
+    * only need to store flags on directories
+    * less mutexes [#13](https://github.com/fsnotify/fsnotify/issues/13)
+    * done can be an unbuffered channel
+    * remove calls to os.NewSyscallError
+* More efficient string concatenation for Event.String() [#52](https://github.com/fsnotify/fsnotify/pull/52) (thanks @mdlayher)
+* kqueue: fix regression in  rework causing subdirectories to be watched [#48](https://github.com/fsnotify/fsnotify/issues/48)
+* kqueue: cleanup internal watch before sending remove event [#51](https://github.com/fsnotify/fsnotify/issues/51)
+
+## [1.0.4] - 2014-09-07
+
+* kqueue: add dragonfly to the build tags.
+* Rename source code files, rearrange code so exported APIs are at the top.
+* Add done channel to example code. [#37](https://github.com/fsnotify/fsnotify/pull/37) (thanks @chenyukang)
+
+## [1.0.3] - 2014-08-19
+
+* [Fix] Windows MOVED_TO now translates to Create like on BSD and Linux. [#36](https://github.com/fsnotify/fsnotify/issues/36)
+
+## [1.0.2] - 2014-08-17
+
+* [Fix] Missing create events on macOS. [#14](https://github.com/fsnotify/fsnotify/issues/14) (thanks @zhsso)
+* [Fix] Make ./path and path equivalent. (thanks @zhsso)
+
+## [1.0.0] - 2014-08-15
+
+* [API] Remove AddWatch on Windows, use Add.
+* Improve documentation for exported identifiers. [#30](https://github.com/fsnotify/fsnotify/issues/30)
+* Minor updates based on feedback from golint.
+
+## dev / 2014-07-09
+
+* Moved to [github.com/fsnotify/fsnotify](https://github.com/fsnotify/fsnotify).
+* Use os.NewSyscallError instead of returning errno (thanks @hariharan-uno)
+
+## dev / 2014-07-04
+
+* kqueue: fix incorrect mutex used in Close()
+* Update example to demonstrate usage of Op.
+
+## dev / 2014-06-28
+
+* [API] Don't set the Write Op for attribute notifications [#4](https://github.com/fsnotify/fsnotify/issues/4)
+* Fix for String() method on Event (thanks Alex Brainman)
+* Don't build on Plan 9 or Solaris (thanks @4ad)
+
+## dev / 2014-06-21
+
+* Events channel of type Event rather than *Event.
+* [internal] use syscall constants directly for inotify and kqueue.
+* [internal] kqueue: rename events to kevents and fileEvent to event.
+
+## dev / 2014-06-19
+
+* Go 1.3+ required on Windows (uses syscall.ERROR_MORE_DATA internally).
+* [internal] remove cookie from Event struct (unused).
+* [internal] Event struct has the same definition across every OS.
+* [internal] remove internal watch and removeWatch methods.
+
+## dev / 2014-06-12
+
+* [API] Renamed Watch() to Add() and RemoveWatch() to Remove().
+* [API] Pluralized channel names: Events and Errors.
+* [API] Renamed FileEvent struct to Event.
+* [API] Op constants replace methods like IsCreate().
+
+## dev / 2014-06-12
+
+* Fix data race on kevent buffer (thanks @tilaks) [#98](https://github.com/howeyc/fsnotify/pull/98)
+
+## dev / 2014-05-23
+
+* [API] Remove current implementation of WatchFlags.
+    * current implementation doesn't take advantage of OS for efficiency
+    * provides little benefit over filtering events as they are received, but has  extra bookkeeping and mutexes
+    * no tests for the current implementation
+    * not fully implemented on Windows [#93](https://github.com/howeyc/fsnotify/issues/93#issuecomment-39285195)
+
+## [0.9.3] - 2014-12-31
+
+* kqueue: cleanup internal watch before sending remove event [#51](https://github.com/fsnotify/fsnotify/issues/51)
+
+## [0.9.2] - 2014-08-17
+
+* [Backport] Fix missing create events on macOS. [#14](https://github.com/fsnotify/fsnotify/issues/14) (thanks @zhsso)
+
+## [0.9.1] - 2014-06-12
+
+* Fix data race on kevent buffer (thanks @tilaks) [#98](https://github.com/howeyc/fsnotify/pull/98)
+
+## [0.9.0] - 2014-01-17
+
+* IsAttrib() for events that only concern a file's metadata [#79][] (thanks @abustany)
+* [Fix] kqueue: fix deadlock [#77][] (thanks @cespare)
+* [NOTICE] Development has moved to `code.google.com/p/go.exp/fsnotify` in preparation for inclusion in the Go standard library.
+
+## [0.8.12] - 2013-11-13
+
+* [API] Remove FD_SET and friends from Linux adapter
+
+## [0.8.11] - 2013-11-02
+
+* [Doc] Add Changelog [#72][] (thanks @nathany)
+* [Doc] Spotlight and double modify events on macOS [#62][] (reported by @paulhammond)
+
+## [0.8.10] - 2013-10-19
+
+* [Fix] kqueue: remove file watches when parent directory is removed [#71][] (reported by @mdwhatcott)
+* [Fix] kqueue: race between Close and readEvents [#70][] (reported by @bernerdschaefer)
+* [Doc] specify OS-specific limits in README (thanks @debrando)
+
+## [0.8.9] - 2013-09-08
+
+* [Doc] Contributing (thanks @nathany)
+* [Doc] update package path in example code [#63][] (thanks @paulhammond)
+* [Doc] GoCI badge in README (Linux only) [#60][]
+* [Doc] Cross-platform testing with Vagrant  [#59][] (thanks @nathany)
+
+## [0.8.8] - 2013-06-17
+
+* [Fix] Windows: handle `ERROR_MORE_DATA` on Windows [#49][] (thanks @jbowtie)
+
+## [0.8.7] - 2013-06-03
+
+* [API] Make syscall flags internal
+* [Fix] inotify: ignore event changes
+* [Fix] race in symlink test [#45][] (reported by @srid)
+* [Fix] tests on Windows
+* lower case error messages
+
+## [0.8.6] - 2013-05-23
+
+* kqueue: Use EVT_ONLY flag on Darwin
+* [Doc] Update README with full example
+
+## [0.8.5] - 2013-05-09
+
+* [Fix] inotify: allow monitoring of "broken" symlinks (thanks @tsg)
+
+## [0.8.4] - 2013-04-07
+
+* [Fix] kqueue: watch all file events [#40][] (thanks @ChrisBuchholz)
+
+## [0.8.3] - 2013-03-13
+
+* [Fix] inoitfy/kqueue memory leak [#36][] (reported by @nbkolchin)
+* [Fix] kqueue: use fsnFlags for watching a directory [#33][] (reported by @nbkolchin)
+
+## [0.8.2] - 2013-02-07
+
+* [Doc] add Authors
+* [Fix] fix data races for map access [#29][] (thanks @fsouza)
+
+## [0.8.1] - 2013-01-09
+
+* [Fix] Windows path separators
+* [Doc] BSD License
+
+## [0.8.0] - 2012-11-09
+
+* kqueue: directory watching improvements (thanks @vmirage)
+* inotify: add `IN_MOVED_TO` [#25][] (requested by @cpisto)
+* [Fix] kqueue: deleting watched directory [#24][] (reported by @jakerr)
+
+## [0.7.4] - 2012-10-09
+
+* [Fix] inotify: fixes from https://codereview.appspot.com/5418045/ (ugorji)
+* [Fix] kqueue: preserve watch flags when watching for delete [#21][] (reported by @robfig)
+* [Fix] kqueue: watch the directory even if it isn't a new watch (thanks @robfig)
+* [Fix] kqueue: modify after recreation of file
+
+## [0.7.3] - 2012-09-27
+
+* [Fix] kqueue: watch with an existing folder inside the watched folder (thanks @vmirage)
+* [Fix] kqueue: no longer get duplicate CREATE events
+
+## [0.7.2] - 2012-09-01
+
+* kqueue: events for created directories
+
+## [0.7.1] - 2012-07-14
+
+* [Fix] for renaming files
+
+## [0.7.0] - 2012-07-02
+
+* [Feature] FSNotify flags
+* [Fix] inotify: Added file name back to event path
+
+## [0.6.0] - 2012-06-06
+
+* kqueue: watch files after directory created (thanks @tmc)
+
+## [0.5.1] - 2012-05-22
+
+* [Fix] inotify: remove all watches before Close()
+
+## [0.5.0] - 2012-05-03
+
+* [API] kqueue: return errors during watch instead of sending over channel
+* kqueue: match symlink behavior on Linux
+* inotify: add `DELETE_SELF` (requested by @taralx)
+* [Fix] kqueue: handle EINTR (reported by @robfig)
+* [Doc] Godoc example [#1][] (thanks @davecheney)
+
+## [0.4.0] - 2012-03-30
+
+* Go 1 released: build with go tool
+* [Feature] Windows support using winfsnotify
+* Windows does not have attribute change notifications
+* Roll attribute notifications into IsModify
+
+## [0.3.0] - 2012-02-19
+
+* kqueue: add files when watch directory
+
+## [0.2.0] - 2011-12-30
+
+* update to latest Go weekly code
+
+## [0.1.0] - 2011-10-19
+
+* kqueue: add watch on file creation to match inotify
+* kqueue: create file event
+* inotify: ignore `IN_IGNORED` events
+* event String()
+* linux: common FileEvent functions
+* initial commit
+
+[#79]: https://github.com/howeyc/fsnotify/pull/79
+[#77]: https://github.com/howeyc/fsnotify/pull/77
+[#72]: https://github.com/howeyc/fsnotify/issues/72
+[#71]: https://github.com/howeyc/fsnotify/issues/71
+[#70]: https://github.com/howeyc/fsnotify/issues/70
+[#63]: https://github.com/howeyc/fsnotify/issues/63
+[#62]: https://github.com/howeyc/fsnotify/issues/62
+[#60]: https://github.com/howeyc/fsnotify/issues/60
+[#59]: https://github.com/howeyc/fsnotify/issues/59
+[#49]: https://github.com/howeyc/fsnotify/issues/49
+[#45]: https://github.com/howeyc/fsnotify/issues/45
+[#40]: https://github.com/howeyc/fsnotify/issues/40
+[#36]: https://github.com/howeyc/fsnotify/issues/36
+[#33]: https://github.com/howeyc/fsnotify/issues/33
+[#29]: https://github.com/howeyc/fsnotify/issues/29
+[#25]: https://github.com/howeyc/fsnotify/issues/25
+[#24]: https://github.com/howeyc/fsnotify/issues/24
+[#21]: https://github.com/howeyc/fsnotify/issues/21

vendor/github.com/fsnotify/fsnotify/CONTRIBUTING.md

+Thank you for your interest in contributing to fsnotify! We try to review and
+merge PRs in a reasonable timeframe, but please be aware that:
+
+- To avoid "wasted" work, please discus changes on the issue tracker first. You
+  can just send PRs, but they may end up being rejected for one reason or the
+  other.
+
+- fsnotify is a cross-platform library, and changes must work reasonably well on
+  all supported platforms.
+
+- Changes will need to be compatible; old code should still compile, and the
+  runtime behaviour can't change in ways that are likely to lead to problems for
+  users.
+
+Testing
+-------
+Just `go test ./...` runs all the tests; the CI runs this on all supported
+platforms. Testing different platforms locally can be done with something like
+[goon] or [Vagrant], but this isn't super-easy to set up at the moment.
+
+Use the `-short` flag to make the "stress test" run faster.
+
+
+[goon]: https://github.com/arp242/goon
+[Vagrant]: https://www.vagrantup.com/
+[integration_test.go]: /integration_test.go

vendor/github.com/fsnotify/fsnotify/LICENSE

+Copyright © 2012 The Go Authors. All rights reserved.
+Copyright © fsnotify Authors. All rights reserved.
+
+Redistribution and use in source and binary forms, with or without modification,
+are permitted provided that the following conditions are met:
+
+* Redistributions of source code must retain the above copyright notice, this
+  list of conditions and the following disclaimer.
+* Redistributions in binary form must reproduce the above copyright notice, this
+  list of conditions and the following disclaimer in the documentation and/or
+  other materials provided with the distribution.
+* Neither the name of Google Inc. nor the names of its contributors may be used
+  to endorse or promote products derived from this software without specific
+  prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

vendor/github.com/fsnotify/fsnotify/README.md

+fsnotify is a Go library to provide cross-platform filesystem notifications on
+Windows, Linux, macOS, BSD, and illumos.
+
+Go 1.17 or newer is required; the full documentation is at
+https://pkg.go.dev/github.com/fsnotify/fsnotify
+
+---
+
+Platform support:
+
+| Backend               | OS         | Status                                                                    |
+| :-------------------- | :--------- | :------------------------------------------------------------------------ |
+| inotify               | Linux      | Supported                                                                 |
+| kqueue                | BSD, macOS | Supported                                                                 |
+| ReadDirectoryChangesW | Windows    | Supported                                                                 |
+| FEN                   | illumos    | Supported                                                                 |
+| fanotify              | Linux 5.9+ | [Not yet](https://github.com/fsnotify/fsnotify/issues/114)                |
+| AHAFS                 | AIX        | [aix branch]; experimental due to lack of maintainer and test environment |
+| FSEvents              | macOS      | [Needs support in x/sys/unix][fsevents]                                   |
+| USN Journals          | Windows    | [Needs support in x/sys/windows][usn]                                     |
+| Polling               | *All*      | [Not yet](https://github.com/fsnotify/fsnotify/issues/9)                  |
+
+Linux and illumos should include Android and Solaris, but these are currently
+untested.
+
+[fsevents]:   https://github.com/fsnotify/fsnotify/issues/11#issuecomment-1279133120
+[usn]:        https://github.com/fsnotify/fsnotify/issues/53#issuecomment-1279829847
+[aix branch]: https://github.com/fsnotify/fsnotify/issues/353#issuecomment-1284590129
+
+Usage
+-----
+A basic example:
+
+```go
+package main
+
+import (
+    "log"
+
+    "github.com/fsnotify/fsnotify"
+)
+
+func main() {
+    // Create new watcher.
+    watcher, err := fsnotify.NewWatcher()
+    if err != nil {
+        log.Fatal(err)
+    }
+    defer watcher.Close()
+
+    // Start listening for events.
+    go func() {
+        for {
+            select {
+            case event, ok := <-watcher.Events:
+                if !ok {
+                    return
+                }
+                log.Println("event:", event)
+                if event.Has(fsnotify.Write) {
+                    log.Println("modified file:", event.Name)
+                }
+            case err, ok := <-watcher.Errors:
+                if !ok {
+                    return
+                }
+                log.Println("error:", err)
+            }
+        }
+    }()
+
+    // Add a path.
+    err = watcher.Add("/tmp")
+    if err != nil {
+        log.Fatal(err)
+    }
+
+    // Block main goroutine forever.
+    <-make(chan struct{})
+}
+```
+
+Some more examples can be found in [cmd/fsnotify](cmd/fsnotify), which can be
+run with:
+
+    % go run ./cmd/fsnotify
+
+Further detailed documentation can be found in godoc:
+https://pkg.go.dev/github.com/fsnotify/fsnotify
+
+FAQ
+---
+### Will a file still be watched when it's moved to another directory?
+No, not unless you are watching the location it was moved to.
+
+### Are subdirectories watched?
+No, you must add watches for any directory you want to watch (a recursive
+watcher is on the roadmap: [#18]).
+
+[#18]: https://github.com/fsnotify/fsnotify/issues/18
+
+### Do I have to watch the Error and Event channels in a goroutine?
+Yes. You can read both channels in the same goroutine using `select` (you don't
+need a separate goroutine for both channels; see the example).
+
+### Why don't notifications work with NFS, SMB, FUSE, /proc, or /sys?
+fsnotify requires support from underlying OS to work. The current NFS and SMB
+protocols does not provide network level support for file notifications, and
+neither do the /proc and /sys virtual filesystems.
+
+This could be fixed with a polling watcher ([#9]), but it's not yet implemented.
+
+[#9]: https://github.com/fsnotify/fsnotify/issues/9
+
+### Why do I get many Chmod events?
+Some programs may generate a lot of attribute changes; for example Spotlight on
+macOS, anti-virus programs, backup applications, and some others are known to do
+this. As a rule, it's typically best to ignore Chmod events. They're often not
+useful, and tend to cause problems.
+
+Spotlight indexing on macOS can result in multiple events (see [#15]). A
+temporary workaround is to add your folder(s) to the *Spotlight Privacy
+settings* until we have a native FSEvents implementation (see [#11]).
+
+[#11]: https://github.com/fsnotify/fsnotify/issues/11
+[#15]: https://github.com/fsnotify/fsnotify/issues/15
+
+### Watching a file doesn't work well
+Watching individual files (rather than directories) is generally not recommended
+as many programs (especially editors) update files atomically: it will write to
+a temporary file which is then moved to to destination, overwriting the original
+(or some variant thereof). The watcher on the original file is now lost, as that
+no longer exists.
+
+The upshot of this is that a power failure or crash won't leave a half-written
+file.
+
+Watch the parent directory and use `Event.Name` to filter out files you're not
+interested in. There is an example of this in `cmd/fsnotify/file.go`.
+
+Platform-specific notes
+-----------------------
+### Linux
+When a file is removed a REMOVE event won't be emitted until all file
+descriptors are closed; it will emit a CHMOD instead:
+
+    fp := os.Open("file")
+    os.Remove("file")        // CHMOD
+    fp.Close()               // REMOVE
+
+This is the event that inotify sends, so not much can be changed about this.
+
+The `fs.inotify.max_user_watches` sysctl variable specifies the upper limit for
+the number of watches per user, and `fs.inotify.max_user_instances` specifies
+the maximum number of inotify instances per user. Every Watcher you create is an
+"instance", and every path you add is a "watch".
+
+These are also exposed in `/proc` as `/proc/sys/fs/inotify/max_user_watches` and
+`/proc/sys/fs/inotify/max_user_instances`
+
+To increase them you can use `sysctl` or write the value to proc file:
+
+    # The default values on Linux 5.18
+    sysctl fs.inotify.max_user_watches=124983
+    sysctl fs.inotify.max_user_instances=128
+
+To make the changes persist on reboot edit `/etc/sysctl.conf` or
+`/usr/lib/sysctl.d/50-default.conf` (details differ per Linux distro; check your
+distro's documentation):
+
+    fs.inotify.max_user_watches=124983
+    fs.inotify.max_user_instances=128
+
+Reaching the limit will result in a "no space left on device" or "too many open
+files" error.
+
+### kqueue (macOS, all BSD systems)
+kqueue requires opening a file descriptor for every file that's being watched;
+so if you're watching a directory with five files then that's six file
+descriptors. You will run in to your system's "max open files" limit faster on
+these platforms.
+
+The sysctl variables `kern.maxfiles` and `kern.maxfilesperproc` can be used to
+control the maximum number of open files.

vendor/github.com/fsnotify/fsnotify/backend_fen.go

+//go:build solaris
+// +build solaris
+
+// Note: the documentation on the Watcher type and methods is generated from
+// mkdoc.zsh
+
+package fsnotify
+
+import (
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"sync"
+
+	"golang.org/x/sys/unix"
+)
+
+// Watcher watches a set of paths, delivering events on a channel.
+//
+// A watcher should not be copied (e.g. pass it by pointer, rather than by
+// value).
+//
+// # Linux notes
+//
+// When a file is removed a Remove event won't be emitted until all file
+// descriptors are closed, and deletes will always emit a Chmod. For example:
+//
+//	fp := os.Open("file")
+//	os.Remove("file")        // Triggers Chmod
+//	fp.Close()               // Triggers Remove
+//
+// This is the event that inotify sends, so not much can be changed about this.
+//
+// The fs.inotify.max_user_watches sysctl variable specifies the upper limit
+// for the number of watches per user, and fs.inotify.max_user_instances
+// specifies the maximum number of inotify instances per user. Every Watcher you
+// create is an "instance", and every path you add is a "watch".
+//
+// These are also exposed in /proc as /proc/sys/fs/inotify/max_user_watches and
+// /proc/sys/fs/inotify/max_user_instances
+//
+// To increase them you can use sysctl or write the value to the /proc file:
+//
+//	# Default values on Linux 5.18
+//	sysctl fs.inotify.max_user_watches=124983
+//	sysctl fs.inotify.max_user_instances=128
+//
+// To make the changes persist on reboot edit /etc/sysctl.conf or
+// /usr/lib/sysctl.d/50-default.conf (details differ per Linux distro; check
+// your distro's documentation):
+//
+//	fs.inotify.max_user_watches=124983
+//	fs.inotify.max_user_instances=128
+//
+// Reaching the limit will result in a "no space left on device" or "too many open
+// files" error.
+//
+// # kqueue notes (macOS, BSD)
+//
+// kqueue requires opening a file descriptor for every file that's being watched;
+// so if you're watching a directory with five files then that's six file
+// descriptors. You will run in to your system's "max open files" limit faster on
+// these platforms.
+//
+// The sysctl variables kern.maxfiles and kern.maxfilesperproc can be used to
+// control the maximum number of open files, as well as /etc/login.conf on BSD
+// systems.
+//
+// # Windows notes
+//
+// Paths can be added as "C:\path\to\dir", but forward slashes
+// ("C:/path/to/dir") will also work.
+//
+// When a watched directory is removed it will always send an event for the
+// directory itself, but may not send events for all files in that directory.
+// Sometimes it will send events for all times, sometimes it will send no
+// events, and often only for some files.
+//
+// The default ReadDirectoryChangesW() buffer size is 64K, which is the largest
+// value that is guaranteed to work with SMB filesystems. If you have many
+// events in quick succession this may not be enough, and you will have to use
+// [WithBufferSize] to increase the value.
+type Watcher struct {
+	// Events sends the filesystem change events.
+	//
+	// fsnotify can send the following events; a "path" here can refer to a
+	// file, directory, symbolic link, or special file like a FIFO.
+	//
+	//   fsnotify.Create    A new path was created; this may be followed by one
+	//                      or more Write events if data also gets written to a
+	//                      file.
+	//
+	//   fsnotify.Remove    A path was removed.
+	//
+	//   fsnotify.Rename    A path was renamed. A rename is always sent with the
+	//                      old path as Event.Name, and a Create event will be
+	//                      sent with the new name. Renames are only sent for
+	//                      paths that are currently watched; e.g. moving an
+	//                      unmonitored file into a monitored directory will
+	//                      show up as just a Create. Similarly, renaming a file
+	//                      to outside a monitored directory will show up as
+	//                      only a Rename.
+	//
+	//   fsnotify.Write     A file or named pipe was written to. A Truncate will
+	//                      also trigger a Write. A single "write action"
+	//                      initiated by the user may show up as one or multiple
+	//                      writes, depending on when the system syncs things to
+	//                      disk. For example when compiling a large Go program
+	//                      you may get hundreds of Write events, and you may
+	//                      want to wait until you've stopped receiving them
+	//                      (see the dedup example in cmd/fsnotify).
+	//
+	//                      Some systems may send Write event for directories
+	//                      when the directory content changes.
+	//
+	//   fsnotify.Chmod     Attributes were changed. On Linux this is also sent
+	//                      when a file is removed (or more accurately, when a
+	//                      link to an inode is removed). On kqueue it's sent
+	//                      when a file is truncated. On Windows it's never
+	//                      sent.
+	Events chan Event
+
+	// Errors sends any errors.
+	//
+	// ErrEventOverflow is used to indicate there are too many events:
+	//
+	//  - inotify:      There are too many queued events (fs.inotify.max_queued_events sysctl)
+	//  - windows:      The buffer size is too small; WithBufferSize() can be used to increase it.
+	//  - kqueue, fen:  Not used.
+	Errors chan error
+
+	mu      sync.Mutex
+	port    *unix.EventPort
+	done    chan struct{}       // Channel for sending a "quit message" to the reader goroutine
+	dirs    map[string]struct{} // Explicitly watched directories
+	watches map[string]struct{} // Explicitly watched non-directories
+}
+
+// NewWatcher creates a new Watcher.
+func NewWatcher() (*Watcher, error) {
+	return NewBufferedWatcher(0)
+}
+
+// NewBufferedWatcher creates a new Watcher with a buffered Watcher.Events
+// channel.
+//
+// The main use case for this is situations with a very large number of events
+// where the kernel buffer size can't be increased (e.g. due to lack of
+// permissions). An unbuffered Watcher will perform better for almost all use
+// cases, and whenever possible you will be better off increasing the kernel
+// buffers instead of adding a large userspace buffer.
+func NewBufferedWatcher(sz uint) (*Watcher, error) {
+	w := &Watcher{
+		Events:  make(chan Event, sz),
+		Errors:  make(chan error),
+		dirs:    make(map[string]struct{}),
+		watches: make(map[string]struct{}),
+		done:    make(chan struct{}),
+	}
+
+	var err error
+	w.port, err = unix.NewEventPort()
+	if err != nil {
+		return nil, fmt.Errorf("fsnotify.NewWatcher: %w", err)
+	}
+
+	go w.readEvents()
+	return w, nil
+}
+
+// sendEvent attempts to send an event to the user, returning true if the event
+// was put in the channel successfully and false if the watcher has been closed.
+func (w *Watcher) sendEvent(name string, op Op) (sent bool) {
+	select {
+	case w.Events <- Event{Name: name, Op: op}:
+		return true
+	case <-w.done:
+		return false
+	}
+}
+
+// sendError attempts to send an error to the user, returning true if the error
+// was put in the channel successfully and false if the watcher has been closed.
+func (w *Watcher) sendError(err error) (sent bool) {
+	select {
+	case w.Errors <- err:
+		return true
+	case <-w.done:
+		return false
+	}
+}
+
+func (w *Watcher) isClosed() bool {
+	select {
+	case <-w.done:
+		return true
+	default:
+		return false
+	}
+}
+
+// Close removes all watches and closes the Events channel.
+func (w *Watcher) Close() error {
+	// Take the lock used by associateFile to prevent lingering events from
+	// being processed after the close
+	w.mu.Lock()
+	defer w.mu.Unlock()
+	if w.isClosed() {
+		return nil
+	}
+	close(w.done)
+	return w.port.Close()
+}
+
+// Add starts monitoring the path for changes.
+//
+// A path can only be watched once; watching it more than once is a no-op and will
+// not return an error. Paths that do not yet exist on the filesystem cannot be
+// watched.
+//
+// A watch will be automatically removed if the watched path is deleted or
+// renamed. The exception is the Windows backend, which doesn't remove the
+// watcher on renames.
+//
+// Notifications on network filesystems (NFS, SMB, FUSE, etc.) or special
+// filesystems (/proc, /sys, etc.) generally don't work.
+//
+// Returns [ErrClosed] if [Watcher.Close] was called.
+//
+// See [Watcher.AddWith] for a version that allows adding options.
+//
+// # Watching directories
+//
+// All files in a directory are monitored, including new files that are created
+// after the watcher is started. Subdirectories are not watched (i.e. it's
+// non-recursive).
+//
+// # Watching files
+//
+// Watching individual files (rather than directories) is generally not
+// recommended as many programs (especially editors) update files atomically: it
+// will write to a temporary file which is then moved to to destination,
+// overwriting the original (or some variant thereof). The watcher on the
+// original file is now lost, as that no longer exists.
+//
+// The upshot of this is that a power failure or crash won't leave a
+// half-written file.
+//
+// Watch the parent directory and use Event.Name to filter out files you're not
+// interested in. There is an example of this in cmd/fsnotify/file.go.
+func (w *Watcher) Add(name string) error { return w.AddWith(name) }
+
+// AddWith is like [Watcher.Add], but allows adding options. When using Add()
+// the defaults described below are used.
+//
+// Possible options are:
+//
+//   - [WithBufferSize] sets the buffer size for the Windows backend; no-op on
+//     other platforms. The default is 64K (65536 bytes).
+func (w *Watcher) AddWith(name string, opts ...addOpt) error {
+	if w.isClosed() {
+		return ErrClosed
+	}
+	if w.port.PathIsWatched(name) {
+		return nil
+	}
+
+	_ = getOptions(opts...)
+
+	// Currently we resolve symlinks that were explicitly requested to be
+	// watched. Otherwise we would use LStat here.
+	stat, err := os.Stat(name)
+	if err != nil {
+		return err
+	}
+
+	// Associate all files in the directory.
+	if stat.IsDir() {
+		err := w.handleDirectory(name, stat, true, w.associateFile)
+		if err != nil {
+			return err
+		}
+
+		w.mu.Lock()
+		w.dirs[name] = struct{}{}
+		w.mu.Unlock()
+		return nil
+	}
+
+	err = w.associateFile(name, stat, true)
+	if err != nil {
+		return err
+	}
+
+	w.mu.Lock()
+	w.watches[name] = struct{}{}
+	w.mu.Unlock()
+	return nil
+}
+
+// Remove stops monitoring the path for changes.
+//
+// Directories are always removed non-recursively. For example, if you added
+// /tmp/dir and /tmp/dir/subdir then you will need to remove both.
+//
+// Removing a path that has not yet been added returns [ErrNonExistentWatch].
+//
+// Returns nil if [Watcher.Close] was called.
+func (w *Watcher) Remove(name string) error {
+	if w.isClosed() {
+		return nil
+	}
+	if !w.port.PathIsWatched(name) {
+		return fmt.Errorf("%w: %s", ErrNonExistentWatch, name)
+	}
+
+	// The user has expressed an intent. Immediately remove this name from
+	// whichever watch list it might be in. If it's not in there the delete
+	// doesn't cause harm.
+	w.mu.Lock()
+	delete(w.watches, name)
+	delete(w.dirs, name)
+	w.mu.Unlock()
+
+	stat, err := os.Stat(name)
+	if err != nil {
+		return err
+	}
+
+	// Remove associations for every file in the directory.
+	if stat.IsDir() {
+		err := w.handleDirectory(name, stat, false, w.dissociateFile)
+		if err != nil {
+			return err
+		}
+		return nil
+	}
+
+	err = w.port.DissociatePath(name)
+	if err != nil {
+		return err
+	}
+
+	return nil
+}
+
+// readEvents contains the main loop that runs in a goroutine watching for events.
+func (w *Watcher) readEvents() {
+	// If this function returns, the watcher has been closed and we can close
+	// these channels
+	defer func() {
+		close(w.Errors)
+		close(w.Events)
+	}()
+
+	pevents := make([]unix.PortEvent, 8)
+	for {
+		count, err := w.port.Get(pevents, 1, nil)
+		if err != nil && err != unix.ETIME {
+			// Interrupted system call (count should be 0) ignore and continue
+			if errors.Is(err, unix.EINTR) && count == 0 {
+				continue
+			}
+			// Get failed because we called w.Close()
+			if errors.Is(err, unix.EBADF) && w.isClosed() {
+				return
+			}
+			// There was an error not caused by calling w.Close()
+			if !w.sendError(err) {
+				return
+			}
+		}
+
+		p := pevents[:count]
+		for _, pevent := range p {
+			if pevent.Source != unix.PORT_SOURCE_FILE {
+				// Event from unexpected source received; should never happen.
+				if !w.sendError(errors.New("Event from unexpected source received")) {
+					return
+				}
+				continue
+			}
+
+			err = w.handleEvent(&pevent)
+			if err != nil {
+				if !w.sendError(err) {
+					return
+				}
+			}
+		}
+	}
+}
+
+func (w *Watcher) handleDirectory(path string, stat os.FileInfo, follow bool, handler func(string, os.FileInfo, bool) error) error {
+	files, err := os.ReadDir(path)
+	if err != nil {
+		return err
+	}
+
+	// Handle all children of the directory.
+	for _, entry := range files {
+		finfo, err := entry.Info()
+		if err != nil {
+			return err
+		}
+		err = handler(filepath.Join(path, finfo.Name()), finfo, false)
+		if err != nil {
+			return err
+		}
+	}
+
+	// And finally handle the directory itself.
+	return handler(path, stat, follow)
+}
+
+// handleEvent might need to emit more than one fsnotify event if the events
+// bitmap matches more than one event type (e.g. the file was both modified and
+// had the attributes changed between when the association was created and the
+// when event was returned)
+func (w *Watcher) handleEvent(event *unix.PortEvent) error {
+	var (
+		events     = event.Events
+		path       = event.Path
+		fmode      = event.Cookie.(os.FileMode)
+		reRegister = true
+	)
+
+	w.mu.Lock()
+	_, watchedDir := w.dirs[path]
+	_, watchedPath := w.watches[path]
+	w.mu.Unlock()
+	isWatched := watchedDir || watchedPath
+
+	if events&unix.FILE_DELETE != 0 {
+		if !w.sendEvent(path, Remove) {
+			return nil
+		}
+		reRegister = false
+	}
+	if events&unix.FILE_RENAME_FROM != 0 {
+		if !w.sendEvent(path, Rename) {
+			return nil
+		}
+		// Don't keep watching the new file name
+		reRegister = false
+	}
+	if events&unix.FILE_RENAME_TO != 0 {
+		// We don't report a Rename event for this case, because Rename events
+		// are interpreted as referring to the _old_ name of the file, and in
+		// this case the event would refer to the new name of the file. This
+		// type of rename event is not supported by fsnotify.
+
+		// inotify reports a Remove event in this case, so we simulate this
+		// here.
+		if !w.sendEvent(path, Remove) {
+			return nil
+		}
+		// Don't keep watching the file that was removed
+		reRegister = false
+	}
+
+	// The file is gone, nothing left to do.
+	if !reRegister {
+		if watchedDir {
+			w.mu.Lock()
+			delete(w.dirs, path)
+			w.mu.Unlock()
+		}
+		if watchedPath {
+			w.mu.Lock()
+			delete(w.watches, path)
+			w.mu.Unlock()
+		}
+		return nil
+	}
+
+	// If we didn't get a deletion the file still exists and we're going to have
+	// to watch it again. Let's Stat it now so that we can compare permissions
+	// and have what we need to continue watching the file
+
+	stat, err := os.Lstat(path)
+	if err != nil {
+		// This is unexpected, but we should still emit an event. This happens
+		// most often on "rm -r" of a subdirectory inside a watched directory We
+		// get a modify event of something happening inside, but by the time we
+		// get here, the sudirectory is already gone. Clearly we were watching
+		// this path but now it is gone. Let's tell the user that it was
+		// removed.
+		if !w.sendEvent(path, Remove) {
+			return nil
+		}
+		// Suppress extra write events on removed directories; they are not
+		// informative and can be confusing.
+		return nil
+	}
+
+	// resolve symlinks that were explicitly watched as we would have at Add()
+	// time. this helps suppress spurious Chmod events on watched symlinks
+	if isWatched {
+		stat, err = os.Stat(path)
+		if err != nil {
+			// The symlink still exists, but the target is gone. Report the
+			// Remove similar to above.
+			if !w.sendEvent(path, Remove) {
+				return nil
+			}
+			// Don't return the error
+		}
+	}
+
+	if events&unix.FILE_MODIFIED != 0 {
+		if fmode.IsDir() {
+			if watchedDir {
+				if err := w.updateDirectory(path); err != nil {
+					return err
+				}
+			} else {
+				if !w.sendEvent(path, Write) {
+					return nil
+				}
+			}
+		} else {
+			if !w.sendEvent(path, Write) {
+				return nil
+			}
+		}
+	}
+	if events&unix.FILE_ATTRIB != 0 && stat != nil {
+		// Only send Chmod if perms changed
+		if stat.Mode().Perm() != fmode.Perm() {
+			if !w.sendEvent(path, Chmod) {
+				return nil
+			}
+		}
+	}
+
+	if stat != nil {
+		// If we get here, it means we've hit an event above that requires us to
+		// continue watching the file or directory
+		return w.associateFile(path, stat, isWatched)
+	}
+	return nil
+}
+
+func (w *Watcher) updateDirectory(path string) error {
+	// The directory was modified, so we must find unwatched entities and watch
+	// them. If something was removed from the directory, nothing will happen,
+	// as everything else should still be watched.
+	files, err := os.ReadDir(path)
+	if err != nil {
+		return err
+	}
+
+	for _, entry := range files {
+		path := filepath.Join(path, entry.Name())
+		if w.port.PathIsWatched(path) {
+			continue
+		}
+
+		finfo, err := entry.Info()
+		if err != nil {
+			return err
+		}
+		err = w.associateFile(path, finfo, false)
+		if err != nil {
+			if !w.sendError(err) {
+				return nil
+			}
+		}
+		if !w.sendEvent(path, Create) {
+			return nil
+		}
+	}
+	return nil
+}
+
+func (w *Watcher) associateFile(path string, stat os.FileInfo, follow bool) error {
+	if w.isClosed() {
+		return ErrClosed
+	}
+	// This is primarily protecting the call to AssociatePath but it is
+	// important and intentional that the call to PathIsWatched is also
+	// protected by this mutex. Without this mutex, AssociatePath has been seen
+	// to error out that the path is already associated.
+	w.mu.Lock()
+	defer w.mu.Unlock()
+
+	if w.port.PathIsWatched(path) {
+		// Remove the old association in favor of this one If we get ENOENT,
+		// then while the x/sys/unix wrapper still thought that this path was
+		// associated, the underlying event port did not. This call will have
+		// cleared up that discrepancy. The most likely cause is that the event
+		// has fired but we haven't processed it yet.
+		err := w.port.DissociatePath(path)
+		if err != nil && err != unix.ENOENT {
+			return err
+		}
+	}
+	// FILE_NOFOLLOW means we watch symlinks themselves rather than their
+	// targets.
+	events := unix.FILE_MODIFIED | unix.FILE_ATTRIB | unix.FILE_NOFOLLOW
+	if follow {
+		// We *DO* follow symlinks for explicitly watched entries.
+		events = unix.FILE_MODIFIED | unix.FILE_ATTRIB
+	}
+	return w.port.AssociatePath(path, stat,
+		events,
+		stat.Mode())
+}
+
+func (w *Watcher) dissociateFile(path string, stat os.FileInfo, unused bool) error {
+	if !w.port.PathIsWatched(path) {
+		return nil
+	}
+	return w.port.DissociatePath(path)
+}
+
+// WatchList returns all paths explicitly added with [Watcher.Add] (and are not
+// yet removed).
+//
+// Returns nil if [Watcher.Close] was called.
+func (w *Watcher) WatchList() []string {
+	if w.isClosed() {
+		return nil
+	}
+
+	w.mu.Lock()
+	defer w.mu.Unlock()
+
+	entries := make([]string, 0, len(w.watches)+len(w.dirs))
+	for pathname := range w.dirs {
+		entries = append(entries, pathname)
+	}
+	for pathname := range w.watches {
+		entries = append(entries, pathname)
+	}
+
+	return entries
+}

vendor/github.com/fsnotify/fsnotify/backend_inotify.go

+//go:build linux && !appengine
+// +build linux,!appengine
+
+// Note: the documentation on the Watcher type and methods is generated from
+// mkdoc.zsh
+
+package fsnotify
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync"
+	"unsafe"
+
+	"golang.org/x/sys/unix"
+)
+
+// Watcher watches a set of paths, delivering events on a channel.
+//
+// A watcher should not be copied (e.g. pass it by pointer, rather than by
+// value).
+//
+// # Linux notes
+//
+// When a file is removed a Remove event won't be emitted until all file
+// descriptors are closed, and deletes will always emit a Chmod. For example:
+//
+//	fp := os.Open("file")
+//	os.Remove("file")        // Triggers Chmod
+//	fp.Close()               // Triggers Remove
+//
+// This is the event that inotify sends, so not much can be changed about this.
+//
+// The fs.inotify.max_user_watches sysctl variable specifies the upper limit
+// for the number of watches per user, and fs.inotify.max_user_instances
+// specifies the maximum number of inotify instances per user. Every Watcher you
+// create is an "instance", and every path you add is a "watch".
+//
+// These are also exposed in /proc as /proc/sys/fs/inotify/max_user_watches and
+// /proc/sys/fs/inotify/max_user_instances
+//
+// To increase them you can use sysctl or write the value to the /proc file:
+//
+//	# Default values on Linux 5.18
+//	sysctl fs.inotify.max_user_watches=124983
+//	sysctl fs.inotify.max_user_instances=128
+//
+// To make the changes persist on reboot edit /etc/sysctl.conf or
+// /usr/lib/sysctl.d/50-default.conf (details differ per Linux distro; check
+// your distro's documentation):
+//
+//	fs.inotify.max_user_watches=124983
+//	fs.inotify.max_user_instances=128
+//
+// Reaching the limit will result in a "no space left on device" or "too many open
+// files" error.
+//
+// # kqueue notes (macOS, BSD)
+//
+// kqueue requires opening a file descriptor for every file that's being watched;
+// so if you're watching a directory with five files then that's six file
+// descriptors. You will run in to your system's "max open files" limit faster on
+// these platforms.
+//
+// The sysctl variables kern.maxfiles and kern.maxfilesperproc can be used to
+// control the maximum number of open files, as well as /etc/login.conf on BSD
+// systems.
+//
+// # Windows notes
+//
+// Paths can be added as "C:\path\to\dir", but forward slashes
+// ("C:/path/to/dir") will also work.
+//
+// When a watched directory is removed it will always send an event for the
+// directory itself, but may not send events for all files in that directory.
+// Sometimes it will send events for all times, sometimes it will send no
+// events, and often only for some files.
+//
+// The default ReadDirectoryChangesW() buffer size is 64K, which is the largest
+// value that is guaranteed to work with SMB filesystems. If you have many
+// events in quick succession this may not be enough, and you will have to use
+// [WithBufferSize] to increase the value.
+type Watcher struct {
+	// Events sends the filesystem change events.
+	//
+	// fsnotify can send the following events; a "path" here can refer to a
+	// file, directory, symbolic link, or special file like a FIFO.
+	//
+	//   fsnotify.Create    A new path was created; this may be followed by one
+	//                      or more Write events if data also gets written to a
+	//                      file.
+	//
+	//   fsnotify.Remove    A path was removed.
+	//
+	//   fsnotify.Rename    A path was renamed. A rename is always sent with the
+	//                      old path as Event.Name, and a Create event will be
+	//                      sent with the new name. Renames are only sent for
+	//                      paths that are currently watched; e.g. moving an
+	//                      unmonitored file into a monitored directory will
+	//                      show up as just a Create. Similarly, renaming a file
+	//                      to outside a monitored directory will show up as
+	//                      only a Rename.
+	//
+	//   fsnotify.Write     A file or named pipe was written to. A Truncate will
+	//                      also trigger a Write. A single "write action"
+	//                      initiated by the user may show up as one or multiple
+	//                      writes, depending on when the system syncs things to
+	//                      disk. For example when compiling a large Go program
+	//                      you may get hundreds of Write events, and you may
+	//                      want to wait until you've stopped receiving them
+	//                      (see the dedup example in cmd/fsnotify).
+	//
+	//                      Some systems may send Write event for directories
+	//                      when the directory content changes.
+	//
+	//   fsnotify.Chmod     Attributes were changed. On Linux this is also sent
+	//                      when a file is removed (or more accurately, when a
+	//                      link to an inode is removed). On kqueue it's sent
+	//                      when a file is truncated. On Windows it's never
+	//                      sent.
+	Events chan Event
+
+	// Errors sends any errors.
+	//
+	// ErrEventOverflow is used to indicate there are too many events:
+	//
+	//  - inotify:      There are too many queued events (fs.inotify.max_queued_events sysctl)
+	//  - windows:      The buffer size is too small; WithBufferSize() can be used to increase it.
+	//  - kqueue, fen:  Not used.
+	Errors chan error
+
+	// Store fd here as os.File.Read() will no longer return on close after
+	// calling Fd(). See: https://github.com/golang/go/issues/26439
+	fd          int
+	inotifyFile *os.File
+	watches     *watches
+	done        chan struct{} // Channel for sending a "quit message" to the reader goroutine
+	closeMu     sync.Mutex
+	doneResp    chan struct{} // Channel to respond to Close
+}
+
+type (
+	watches struct {
+		mu   sync.RWMutex
+		wd   map[uint32]*watch // wd → watch
+		path map[string]uint32 // pathname → wd
+	}
+	watch struct {
+		wd    uint32 // Watch descriptor (as returned by the inotify_add_watch() syscall)
+		flags uint32 // inotify flags of this watch (see inotify(7) for the list of valid flags)
+		path  string // Watch path.
+	}
+)
+
+func newWatches() *watches {
+	return &watches{
+		wd:   make(map[uint32]*watch),
+		path: make(map[string]uint32),
+	}
+}
+
+func (w *watches) len() int {
+	w.mu.RLock()
+	defer w.mu.RUnlock()
+	return len(w.wd)
+}
+
+func (w *watches) add(ww *watch) {
+	w.mu.Lock()
+	defer w.mu.Unlock()
+	w.wd[ww.wd] = ww
+	w.path[ww.path] = ww.wd
+}
+
+func (w *watches) remove(wd uint32) {
+	w.mu.Lock()
+	defer w.mu.Unlock()
+	delete(w.path, w.wd[wd].path)
+	delete(w.wd, wd)
+}
+
+func (w *watches) removePath(path string) (uint32, bool) {
+	w.mu.Lock()
+	defer w.mu.Unlock()
+
+	wd, ok := w.path[path]
+	if !ok {
+		return 0, false
+	}
+
+	delete(w.path, path)
+	delete(w.wd, wd)
+
+	return wd, true
+}
+
+func (w *watches) byPath(path string) *watch {
+	w.mu.RLock()
+	defer w.mu.RUnlock()
+	return w.wd[w.path[path]]
+}
+
+func (w *watches) byWd(wd uint32) *watch {
+	w.mu.RLock()
+	defer w.mu.RUnlock()
+	return w.wd[wd]
+}
+
+func (w *watches) updatePath(path string, f func(*watch) (*watch, error)) error {
+	w.mu.Lock()
+	defer w.mu.Unlock()
+
+	var existing *watch
+	wd, ok := w.path[path]
+	if ok {
+		existing = w.wd[wd]
+	}
+
+	upd, err := f(existing)
+	if err != nil {
+		return err
+	}
+	if upd != nil {
+		w.wd[upd.wd] = upd
+		w.path[upd.path] = upd.wd
+
+		if upd.wd != wd {
+			delete(w.wd, wd)
+		}
+	}
+
+	return nil
+}
+
+// NewWatcher creates a new Watcher.
+func NewWatcher() (*Watcher, error) {
+	return NewBufferedWatcher(0)
+}
+
+// NewBufferedWatcher creates a new Watcher with a buffered Watcher.Events
+// channel.
+//
+// The main use case for this is situations with a very large number of events
+// where the kernel buffer size can't be increased (e.g. due to lack of
+// permissions). An unbuffered Watcher will perform better for almost all use
+// cases, and whenever possible you will be better off increasing the kernel
+// buffers instead of adding a large userspace buffer.
+func NewBufferedWatcher(sz uint) (*Watcher, error) {
+	// Need to set nonblocking mode for SetDeadline to work, otherwise blocking
+	// I/O operations won't terminate on close.
+	fd, errno := unix.InotifyInit1(unix.IN_CLOEXEC | unix.IN_NONBLOCK)
+	if fd == -1 {
+		return nil, errno
+	}
+
+	w := &Watcher{
+		fd:          fd,
+		inotifyFile: os.NewFile(uintptr(fd), ""),
+		watches:     newWatches(),
+		Events:      make(chan Event, sz),
+		Errors:      make(chan error),
+		done:        make(chan struct{}),
+		doneResp:    make(chan struct{}),
+	}
+
+	go w.readEvents()
+	return w, nil
+}
+
+// Returns true if the event was sent, or false if watcher is closed.
+func (w *Watcher) sendEvent(e Event) bool {
+	select {
+	case w.Events <- e:
+		return true
+	case <-w.done:
+		return false
+	}
+}
+
+// Returns true if the error was sent, or false if watcher is closed.
+func (w *Watcher) sendError(err error) bool {
+	select {
+	case w.Errors <- err:
+		return true
+	case <-w.done:
+		return false
+	}
+}
+
+func (w *Watcher) isClosed() bool {
+	select {
+	case <-w.done:
+		return true
+	default:
+		return false
+	}
+}
+
+// Close removes all watches and closes the Events channel.
+func (w *Watcher) Close() error {
+	w.closeMu.Lock()
+	if w.isClosed() {
+		w.closeMu.Unlock()
+		return nil
+	}
+	close(w.done)
+	w.closeMu.Unlock()
+
+	// Causes any blocking reads to return with an error, provided the file
+	// still supports deadline operations.
+	err := w.inotifyFile.Close()
+	if err != nil {
+		return err
+	}
+
+	// Wait for goroutine to close
+	<-w.doneResp
+
+	return nil
+}
+
+// Add starts monitoring the path for changes.
+//
+// A path can only be watched once; watching it more than once is a no-op and will
+// not return an error. Paths that do not yet exist on the filesystem cannot be
+// watched.
+//
+// A watch will be automatically removed if the watched path is deleted or
+// renamed. The exception is the Windows backend, which doesn't remove the
+// watcher on renames.
+//
+// Notifications on network filesystems (NFS, SMB, FUSE, etc.) or special
+// filesystems (/proc, /sys, etc.) generally don't work.
+//
+// Returns [ErrClosed] if [Watcher.Close] was called.
+//
+// See [Watcher.AddWith] for a version that allows adding options.
+//
+// # Watching directories
+//
+// All files in a directory are monitored, including new files that are created
+// after the watcher is started. Subdirectories are not watched (i.e. it's
+// non-recursive).
+//
+// # Watching files
+//
+// Watching individual files (rather than directories) is generally not
+// recommended as many programs (especially editors) update files atomically: it
+// will write to a temporary file which is then moved to to destination,
+// overwriting the original (or some variant thereof). The watcher on the
+// original file is now lost, as that no longer exists.
+//
+// The upshot of this is that a power failure or crash won't leave a
+// half-written file.
+//
+// Watch the parent directory and use Event.Name to filter out files you're not
+// interested in. There is an example of this in cmd/fsnotify/file.go.
+func (w *Watcher) Add(name string) error { return w.AddWith(name) }
+
+// AddWith is like [Watcher.Add], but allows adding options. When using Add()
+// the defaults described below are used.
+//
+// Possible options are:
+//
+//   - [WithBufferSize] sets the buffer size for the Windows backend; no-op on
+//     other platforms. The default is 64K (65536 bytes).
+func (w *Watcher) AddWith(name string, opts ...addOpt) error {
+	if w.isClosed() {
+		return ErrClosed
+	}
+
+	name = filepath.Clean(name)
+	_ = getOptions(opts...)
+
+	var flags uint32 = unix.IN_MOVED_TO | unix.IN_MOVED_FROM |
+		unix.IN_CREATE | unix.IN_ATTRIB | unix.IN_MODIFY |
+		unix.IN_MOVE_SELF | unix.IN_DELETE | unix.IN_DELETE_SELF
+
+	return w.watches.updatePath(name, func(existing *watch) (*watch, error) {
+		if existing != nil {
+			flags |= existing.flags | unix.IN_MASK_ADD
+		}
+
+		wd, err := unix.InotifyAddWatch(w.fd, name, flags)
+		if wd == -1 {
+			return nil, err
+		}
+
+		if existing == nil {
+			return &watch{
+				wd:    uint32(wd),
+				path:  name,
+				flags: flags,
+			}, nil
+		}
+
+		existing.wd = uint32(wd)
+		existing.flags = flags
+		return existing, nil
+	})
+}
+
+// Remove stops monitoring the path for changes.
+//
+// Directories are always removed non-recursively. For example, if you added
+// /tmp/dir and /tmp/dir/subdir then you will need to remove both.
+//
+// Removing a path that has not yet been added returns [ErrNonExistentWatch].
+//
+// Returns nil if [Watcher.Close] was called.
+func (w *Watcher) Remove(name string) error {
+	if w.isClosed() {
+		return nil
+	}
+	return w.remove(filepath.Clean(name))
+}
+
+func (w *Watcher) remove(name string) error {
+	wd, ok := w.watches.removePath(name)
+	if !ok {
+		return fmt.Errorf("%w: %s", ErrNonExistentWatch, name)
+	}
+
+	success, errno := unix.InotifyRmWatch(w.fd, wd)
+	if success == -1 {
+		// TODO: Perhaps it's not helpful to return an error here in every case;
+		//       The only two possible errors are:
+		//
+		//       - EBADF, which happens when w.fd is not a valid file descriptor
+		//         of any kind.
+		//       - EINVAL, which is when fd is not an inotify descriptor or wd
+		//         is not a valid watch descriptor. Watch descriptors are
+		//         invalidated when they are removed explicitly or implicitly;
+		//         explicitly by inotify_rm_watch, implicitly when the file they
+		//         are watching is deleted.
+		return errno
+	}
+	return nil
+}
+
+// WatchList returns all paths explicitly added with [Watcher.Add] (and are not
+// yet removed).
+//
+// Returns nil if [Watcher.Close] was called.
+func (w *Watcher) WatchList() []string {
+	if w.isClosed() {
+		return nil
+	}
+
+	entries := make([]string, 0, w.watches.len())
+	w.watches.mu.RLock()
+	for pathname := range w.watches.path {
+		entries = append(entries, pathname)
+	}
+	w.watches.mu.RUnlock()
+
+	return entries
+}
+
+// readEvents reads from the inotify file descriptor, converts the
+// received events into Event objects and sends them via the Events channel
+func (w *Watcher) readEvents() {
+	defer func() {
+		close(w.doneResp)
+		close(w.Errors)
+		close(w.Events)
+	}()
+
+	var (
+		buf   [unix.SizeofInotifyEvent * 4096]byte // Buffer for a maximum of 4096 raw events
+		errno error                                // Syscall errno
+	)
+	for {
+		// See if we have been closed.
+		if w.isClosed() {
+			return
+		}
+
+		n, err := w.inotifyFile.Read(buf[:])
+		switch {
+		case errors.Unwrap(err) == os.ErrClosed:
+			return
+		case err != nil:
+			if !w.sendError(err) {
+				return
+			}
+			continue
+		}
+
+		if n < unix.SizeofInotifyEvent {
+			var err error
+			if n == 0 {
+				err = io.EOF // If EOF is received. This should really never happen.
+			} else if n < 0 {
+				err = errno // If an error occurred while reading.
+			} else {
+				err = errors.New("notify: short read in readEvents()") // Read was too short.
+			}
+			if !w.sendError(err) {
+				return
+			}
+			continue
+		}
+
+		var offset uint32
+		// We don't know how many events we just read into the buffer
+		// While the offset points to at least one whole event...
+		for offset <= uint32(n-unix.SizeofInotifyEvent) {
+			var (
+				// Point "raw" to the event in the buffer
+				raw     = (*unix.InotifyEvent)(unsafe.Pointer(&buf[offset]))
+				mask    = uint32(raw.Mask)
+				nameLen = uint32(raw.Len)
+			)
+
+			if mask&unix.IN_Q_OVERFLOW != 0 {
+				if !w.sendError(ErrEventOverflow) {
+					return
+				}
+			}
+
+			// If the event happened to the watched directory or the watched file, the kernel
+			// doesn't append the filename to the event, but we would like to always fill the
+			// the "Name" field with a valid filename. We retrieve the path of the watch from
+			// the "paths" map.
+			watch := w.watches.byWd(uint32(raw.Wd))
+
+			// inotify will automatically remove the watch on deletes; just need
+			// to clean our state here.
+			if watch != nil && mask&unix.IN_DELETE_SELF == unix.IN_DELETE_SELF {
+				w.watches.remove(watch.wd)
+			}
+			// We can't really update the state when a watched path is moved;
+			// only IN_MOVE_SELF is sent and not IN_MOVED_{FROM,TO}. So remove
+			// the watch.
+			if watch != nil && mask&unix.IN_MOVE_SELF == unix.IN_MOVE_SELF {
+				err := w.remove(watch.path)
+				if err != nil && !errors.Is(err, ErrNonExistentWatch) {
+					if !w.sendError(err) {
+						return
+					}
+				}
+			}
+
+			var name string
+			if watch != nil {
+				name = watch.path
+			}
+			if nameLen > 0 {
+				// Point "bytes" at the first byte of the filename
+				bytes := (*[unix.PathMax]byte)(unsafe.Pointer(&buf[offset+unix.SizeofInotifyEvent]))[:nameLen:nameLen]
+				// The filename is padded with NULL bytes. TrimRight() gets rid of those.
+				name += "/" + strings.TrimRight(string(bytes[0:nameLen]), "\000")
+			}
+
+			event := w.newEvent(name, mask)
+
+			// Send the events that are not ignored on the events channel
+			if mask&unix.IN_IGNORED == 0 {
+				if !w.sendEvent(event) {
+					return
+				}
+			}
+
+			// Move to the next event in the buffer
+			offset += unix.SizeofInotifyEvent + nameLen
+		}
+	}
+}
+
+// newEvent returns an platform-independent Event based on an inotify mask.
+func (w *Watcher) newEvent(name string, mask uint32) Event {
+	e := Event{Name: name}
+	if mask&unix.IN_CREATE == unix.IN_CREATE || mask&unix.IN_MOVED_TO == unix.IN_MOVED_TO {
+		e.Op |= Create
+	}
+	if mask&unix.IN_DELETE_SELF == unix.IN_DELETE_SELF || mask&unix.IN_DELETE == unix.IN_DELETE {
+		e.Op |= Remove
+	}
+	if mask&unix.IN_MODIFY == unix.IN_MODIFY {
+		e.Op |= Write
+	}
+	if mask&unix.IN_MOVE_SELF == unix.IN_MOVE_SELF || mask&unix.IN_MOVED_FROM == unix.IN_MOVED_FROM {
+		e.Op |= Rename
+	}
+	if mask&unix.IN_ATTRIB == unix.IN_ATTRIB {
+		e.Op |= Chmod
+	}
+	return e
+}

vendor/github.com/fsnotify/fsnotify/backend_kqueue.go

+//go:build freebsd || openbsd || netbsd || dragonfly || darwin
+// +build freebsd openbsd netbsd dragonfly darwin
+
+// Note: the documentation on the Watcher type and methods is generated from
+// mkdoc.zsh
+
+package fsnotify
+
+import (
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"sync"
+
+	"golang.org/x/sys/unix"
+)
+
+// Watcher watches a set of paths, delivering events on a channel.
+//
+// A watcher should not be copied (e.g. pass it by pointer, rather than by
+// value).
+//
+// # Linux notes
+//
+// When a file is removed a Remove event won't be emitted until all file
+// descriptors are closed, and deletes will always emit a Chmod. For example:
+//
+//	fp := os.Open("file")
+//	os.Remove("file")        // Triggers Chmod
+//	fp.Close()               // Triggers Remove
+//
+// This is the event that inotify sends, so not much can be changed about this.
+//
+// The fs.inotify.max_user_watches sysctl variable specifies the upper limit
+// for the number of watches per user, and fs.inotify.max_user_instances
+// specifies the maximum number of inotify instances per user. Every Watcher you
+// create is an "instance", and every path you add is a "watch".
+//
+// These are also exposed in /proc as /proc/sys/fs/inotify/max_user_watches and
+// /proc/sys/fs/inotify/max_user_instances
+//
+// To increase them you can use sysctl or write the value to the /proc file:
+//
+//	# Default values on Linux 5.18
+//	sysctl fs.inotify.max_user_watches=124983
+//	sysctl fs.inotify.max_user_instances=128
+//
+// To make the changes persist on reboot edit /etc/sysctl.conf or
+// /usr/lib/sysctl.d/50-default.conf (details differ per Linux distro; check
+// your distro's documentation):
+//
+//	fs.inotify.max_user_watches=124983
+//	fs.inotify.max_user_instances=128
+//
+// Reaching the limit will result in a "no space left on device" or "too many open
+// files" error.
+//
+// # kqueue notes (macOS, BSD)
+//
+// kqueue requires opening a file descriptor for every file that's being watched;
+// so if you're watching a directory with five files then that's six file
+// descriptors. You will run in to your system's "max open files" limit faster on
+// these platforms.
+//
+// The sysctl variables kern.maxfiles and kern.maxfilesperproc can be used to
+// control the maximum number of open files, as well as /etc/login.conf on BSD
+// systems.
+//
+// # Windows notes
+//
+// Paths can be added as "C:\path\to\dir", but forward slashes
+// ("C:/path/to/dir") will also work.
+//
+// When a watched directory is removed it will always send an event for the
+// directory itself, but may not send events for all files in that directory.
+// Sometimes it will send events for all times, sometimes it will send no
+// events, and often only for some files.
+//
+// The default ReadDirectoryChangesW() buffer size is 64K, which is the largest
+// value that is guaranteed to work with SMB filesystems. If you have many
+// events in quick succession this may not be enough, and you will have to use
+// [WithBufferSize] to increase the value.
+type Watcher struct {
+	// Events sends the filesystem change events.
+	//
+	// fsnotify can send the following events; a "path" here can refer to a
+	// file, directory, symbolic link, or special file like a FIFO.
+	//
+	//   fsnotify.Create    A new path was created; this may be followed by one
+	//                      or more Write events if data also gets written to a
+	//                      file.
+	//
+	//   fsnotify.Remove    A path was removed.
+	//
+	//   fsnotify.Rename    A path was renamed. A rename is always sent with the
+	//                      old path as Event.Name, and a Create event will be
+	//                      sent with the new name. Renames are only sent for
+	//                      paths that are currently watched; e.g. moving an
+	//                      unmonitored file into a monitored directory will
+	//                      show up as just a Create. Similarly, renaming a file
+	//                      to outside a monitored directory will show up as
+	//                      only a Rename.
+	//
+	//   fsnotify.Write     A file or named pipe was written to. A Truncate will
+	//                      also trigger a Write. A single "write action"
+	//                      initiated by the user may show up as one or multiple
+	//                      writes, depending on when the system syncs things to
+	//                      disk. For example when compiling a large Go program
+	//                      you may get hundreds of Write events, and you may
+	//                      want to wait until you've stopped receiving them
+	//                      (see the dedup example in cmd/fsnotify).
+	//
+	//                      Some systems may send Write event for directories
+	//                      when the directory content changes.
+	//
+	//   fsnotify.Chmod     Attributes were changed. On Linux this is also sent
+	//                      when a file is removed (or more accurately, when a
+	//                      link to an inode is removed). On kqueue it's sent
+	//                      when a file is truncated. On Windows it's never
+	//                      sent.
+	Events chan Event
+
+	// Errors sends any errors.
+	//
+	// ErrEventOverflow is used to indicate there are too many events:
+	//
+	//  - inotify:      There are too many queued events (fs.inotify.max_queued_events sysctl)
+	//  - windows:      The buffer size is too small; WithBufferSize() can be used to increase it.
+	//  - kqueue, fen:  Not used.
+	Errors chan error
+
+	done         chan struct{}
+	kq           int                         // File descriptor (as returned by the kqueue() syscall).
+	closepipe    [2]int                      // Pipe used for closing.
+	mu           sync.Mutex                  // Protects access to watcher data
+	watches      map[string]int              // Watched file descriptors (key: path).
+	watchesByDir map[string]map[int]struct{} // Watched file descriptors indexed by the parent directory (key: dirname(path)).
+	userWatches  map[string]struct{}         // Watches added with Watcher.Add()
+	dirFlags     map[string]uint32           // Watched directories to fflags used in kqueue.
+	paths        map[int]pathInfo            // File descriptors to path names for processing kqueue events.
+	fileExists   map[string]struct{}         // Keep track of if we know this file exists (to stop duplicate create events).
+	isClosed     bool                        // Set to true when Close() is first called
+}
+
+type pathInfo struct {
+	name  string
+	isDir bool
+}
+
+// NewWatcher creates a new Watcher.
+func NewWatcher() (*Watcher, error) {
+	return NewBufferedWatcher(0)
+}
+
+// NewBufferedWatcher creates a new Watcher with a buffered Watcher.Events
+// channel.
+//
+// The main use case for this is situations with a very large number of events
+// where the kernel buffer size can't be increased (e.g. due to lack of
+// permissions). An unbuffered Watcher will perform better for almost all use
+// cases, and whenever possible you will be better off increasing the kernel
+// buffers instead of adding a large userspace buffer.
+func NewBufferedWatcher(sz uint) (*Watcher, error) {
+	kq, closepipe, err := newKqueue()
+	if err != nil {
+		return nil, err
+	}
+
+	w := &Watcher{
+		kq:           kq,
+		closepipe:    closepipe,
+		watches:      make(map[string]int),
+		watchesByDir: make(map[string]map[int]struct{}),
+		dirFlags:     make(map[string]uint32),
+		paths:        make(map[int]pathInfo),
+		fileExists:   make(map[string]struct{}),
+		userWatches:  make(map[string]struct{}),
+		Events:       make(chan Event, sz),
+		Errors:       make(chan error),
+		done:         make(chan struct{}),
+	}
+
+	go w.readEvents()
+	return w, nil
+}
+
+// newKqueue creates a new kernel event queue and returns a descriptor.
+//
+// This registers a new event on closepipe, which will trigger an event when
+// it's closed. This way we can use kevent() without timeout/polling; without
+// the closepipe, it would block forever and we wouldn't be able to stop it at
+// all.
+func newKqueue() (kq int, closepipe [2]int, err error) {
+	kq, err = unix.Kqueue()
+	if kq == -1 {
+		return kq, closepipe, err
+	}
+
+	// Register the close pipe.
+	err = unix.Pipe(closepipe[:])
+	if err != nil {
+		unix.Close(kq)
+		return kq, closepipe, err
+	}
+
+	// Register changes to listen on the closepipe.
+	changes := make([]unix.Kevent_t, 1)
+	// SetKevent converts int to the platform-specific types.
+	unix.SetKevent(&changes[0], closepipe[0], unix.EVFILT_READ,
+		unix.EV_ADD|unix.EV_ENABLE|unix.EV_ONESHOT)
+
+	ok, err := unix.Kevent(kq, changes, nil, nil)
+	if ok == -1 {
+		unix.Close(kq)
+		unix.Close(closepipe[0])
+		unix.Close(closepipe[1])
+		return kq, closepipe, err
+	}
+	return kq, closepipe, nil
+}
+
+// Returns true if the event was sent, or false if watcher is closed.
+func (w *Watcher) sendEvent(e Event) bool {
+	select {
+	case w.Events <- e:
+		return true
+	case <-w.done:
+		return false
+	}
+}
+
+// Returns true if the error was sent, or false if watcher is closed.
+func (w *Watcher) sendError(err error) bool {
+	select {
+	case w.Errors <- err:
+		return true
+	case <-w.done:
+		return false
+	}
+}
+
+// Close removes all watches and closes the Events channel.
+func (w *Watcher) Close() error {
+	w.mu.Lock()
+	if w.isClosed {
+		w.mu.Unlock()
+		return nil
+	}
+	w.isClosed = true
+
+	// copy paths to remove while locked
+	pathsToRemove := make([]string, 0, len(w.watches))
+	for name := range w.watches {
+		pathsToRemove = append(pathsToRemove, name)
+	}
+	w.mu.Unlock() // Unlock before calling Remove, which also locks
+	for _, name := range pathsToRemove {
+		w.Remove(name)
+	}
+
+	// Send "quit" message to the reader goroutine.
+	unix.Close(w.closepipe[1])
+	close(w.done)
+
+	return nil
+}
+
+// Add starts monitoring the path for changes.
+//
+// A path can only be watched once; watching it more than once is a no-op and will
+// not return an error. Paths that do not yet exist on the filesystem cannot be
+// watched.
+//
+// A watch will be automatically removed if the watched path is deleted or
+// renamed. The exception is the Windows backend, which doesn't remove the
+// watcher on renames.
+//
+// Notifications on network filesystems (NFS, SMB, FUSE, etc.) or special
+// filesystems (/proc, /sys, etc.) generally don't work.
+//
+// Returns [ErrClosed] if [Watcher.Close] was called.
+//
+// See [Watcher.AddWith] for a version that allows adding options.
+//
+// # Watching directories
+//
+// All files in a directory are monitored, including new files that are created
+// after the watcher is started. Subdirectories are not watched (i.e. it's
+// non-recursive).
+//
+// # Watching files
+//
+// Watching individual files (rather than directories) is generally not
+// recommended as many programs (especially editors) update files atomically: it
+// will write to a temporary file which is then moved to to destination,
+// overwriting the original (or some variant thereof). The watcher on the
+// original file is now lost, as that no longer exists.
+//
+// The upshot of this is that a power failure or crash won't leave a
+// half-written file.
+//
+// Watch the parent directory and use Event.Name to filter out files you're not
+// interested in. There is an example of this in cmd/fsnotify/file.go.
+func (w *Watcher) Add(name string) error { return w.AddWith(name) }
+
+// AddWith is like [Watcher.Add], but allows adding options. When using Add()
+// the defaults described below are used.
+//
+// Possible options are:
+//
+//   - [WithBufferSize] sets the buffer size for the Windows backend; no-op on
+//     other platforms. The default is 64K (65536 bytes).
+func (w *Watcher) AddWith(name string, opts ...addOpt) error {
+	_ = getOptions(opts...)
+
+	w.mu.Lock()
+	w.userWatches[name] = struct{}{}
+	w.mu.Unlock()
+	_, err := w.addWatch(name, noteAllEvents)
+	return err
+}
+
+// Remove stops monitoring the path for changes.
+//
+// Directories are always removed non-recursively. For example, if you added
+// /tmp/dir and /tmp/dir/subdir then you will need to remove both.
+//
+// Removing a path that has not yet been added returns [ErrNonExistentWatch].
+//
+// Returns nil if [Watcher.Close] was called.
+func (w *Watcher) Remove(name string) error {
+	return w.remove(name, true)
+}
+
+func (w *Watcher) remove(name string, unwatchFiles bool) error {
+	name = filepath.Clean(name)
+	w.mu.Lock()
+	if w.isClosed {
+		w.mu.Unlock()
+		return nil
+	}
+	watchfd, ok := w.watches[name]
+	w.mu.Unlock()
+	if !ok {
+		return fmt.Errorf("%w: %s", ErrNonExistentWatch, name)
+	}
+
+	err := w.register([]int{watchfd}, unix.EV_DELETE, 0)
+	if err != nil {
+		return err
+	}
+
+	unix.Close(watchfd)
+
+	w.mu.Lock()
+	isDir := w.paths[watchfd].isDir
+	delete(w.watches, name)
+	delete(w.userWatches, name)
+
+	parentName := filepath.Dir(name)
+	delete(w.watchesByDir[parentName], watchfd)
+
+	if len(w.watchesByDir[parentName]) == 0 {
+		delete(w.watchesByDir, parentName)
+	}
+
+	delete(w.paths, watchfd)
+	delete(w.dirFlags, name)
+	delete(w.fileExists, name)
+	w.mu.Unlock()
+
+	// Find all watched paths that are in this directory that are not external.
+	if unwatchFiles && isDir {
+		var pathsToRemove []string
+		w.mu.Lock()
+		for fd := range w.watchesByDir[name] {
+			path := w.paths[fd]
+			if _, ok := w.userWatches[path.name]; !ok {
+				pathsToRemove = append(pathsToRemove, path.name)
+			}
+		}
+		w.mu.Unlock()
+		for _, name := range pathsToRemove {
+			// Since these are internal, not much sense in propagating error to
+			// the user, as that will just confuse them with an error about a
+			// path they did not explicitly watch themselves.
+			w.Remove(name)
+		}
+	}
+	return nil
+}
+
+// WatchList returns all paths explicitly added with [Watcher.Add] (and are not
+// yet removed).
+//
+// Returns nil if [Watcher.Close] was called.
+func (w *Watcher) WatchList() []string {
+	w.mu.Lock()
+	defer w.mu.Unlock()
+	if w.isClosed {
+		return nil
+	}
+
+	entries := make([]string, 0, len(w.userWatches))
+	for pathname := range w.userWatches {
+		entries = append(entries, pathname)
+	}
+
+	return entries
+}
+
+// Watch all events (except NOTE_EXTEND, NOTE_LINK, NOTE_REVOKE)
+const noteAllEvents = unix.NOTE_DELETE | unix.NOTE_WRITE | unix.NOTE_ATTRIB | unix.NOTE_RENAME
+
+// addWatch adds name to the watched file set; the flags are interpreted as
+// described in kevent(2).
+//
+// Returns the real path to the file which was added, with symlinks resolved.
+func (w *Watcher) addWatch(name string, flags uint32) (string, error) {
+	var isDir bool
+	name = filepath.Clean(name)
+
+	w.mu.Lock()
+	if w.isClosed {
+		w.mu.Unlock()
+		return "", ErrClosed
+	}
+	watchfd, alreadyWatching := w.watches[name]
+	// We already have a watch, but we can still override flags.
+	if alreadyWatching {
+		isDir = w.paths[watchfd].isDir
+	}
+	w.mu.Unlock()
+
+	if !alreadyWatching {
+		fi, err := os.Lstat(name)
+		if err != nil {
+			return "", err
+		}
+
+		// Don't watch sockets or named pipes
+		if (fi.Mode()&os.ModeSocket == os.ModeSocket) || (fi.Mode()&os.ModeNamedPipe == os.ModeNamedPipe) {
+			return "", nil
+		}
+
+		// Follow Symlinks.
+		if fi.Mode()&os.ModeSymlink == os.ModeSymlink {
+			link, err := os.Readlink(name)
+			if err != nil {
+				// Return nil because Linux can add unresolvable symlinks to the
+				// watch list without problems, so maintain consistency with
+				// that. There will be no file events for broken symlinks.
+				// TODO: more specific check; returns os.PathError; ENOENT?
+				return "", nil
+			}
+
+			w.mu.Lock()
+			_, alreadyWatching = w.watches[link]
+			w.mu.Unlock()
+
+			if alreadyWatching {
+				// Add to watches so we don't get spurious Create events later
+				// on when we diff the directories.
+				w.watches[name] = 0
+				w.fileExists[name] = struct{}{}
+				return link, nil
+			}
+
+			name = link
+			fi, err = os.Lstat(name)
+			if err != nil {
+				return "", nil
+			}
+		}
+
+		// Retry on EINTR; open() can return EINTR in practice on macOS.
+		// See #354, and Go issues 11180 and 39237.
+		for {
+			watchfd, err = unix.Open(name, openMode, 0)
+			if err == nil {
+				break
+			}
+			if errors.Is(err, unix.EINTR) {
+				continue
+			}
+
+			return "", err
+		}
+
+		isDir = fi.IsDir()
+	}
+
+	err := w.register([]int{watchfd}, unix.EV_ADD|unix.EV_CLEAR|unix.EV_ENABLE, flags)
+	if err != nil {
+		unix.Close(watchfd)
+		return "", err
+	}
+
+	if !alreadyWatching {
+		w.mu.Lock()
+		parentName := filepath.Dir(name)
+		w.watches[name] = watchfd
+
+		watchesByDir, ok := w.watchesByDir[parentName]
+		if !ok {
+			watchesByDir = make(map[int]struct{}, 1)
+			w.watchesByDir[parentName] = watchesByDir
+		}
+		watchesByDir[watchfd] = struct{}{}
+		w.paths[watchfd] = pathInfo{name: name, isDir: isDir}
+		w.mu.Unlock()
+	}
+
+	if isDir {
+		// Watch the directory if it has not been watched before, or if it was
+		// watched before, but perhaps only a NOTE_DELETE (watchDirectoryFiles)
+		w.mu.Lock()
+
+		watchDir := (flags&unix.NOTE_WRITE) == unix.NOTE_WRITE &&
+			(!alreadyWatching || (w.dirFlags[name]&unix.NOTE_WRITE) != unix.NOTE_WRITE)
+		// Store flags so this watch can be updated later
+		w.dirFlags[name] = flags
+		w.mu.Unlock()
+
+		if watchDir {
+			if err := w.watchDirectoryFiles(name); err != nil {
+				return "", err
+			}
+		}
+	}
+	return name, nil
+}
+
+// readEvents reads from kqueue and converts the received kevents into
+// Event values that it sends down the Events channel.
+func (w *Watcher) readEvents() {
+	defer func() {
+		close(w.Events)
+		close(w.Errors)
+		_ = unix.Close(w.kq)
+		unix.Close(w.closepipe[0])
+	}()
+
+	eventBuffer := make([]unix.Kevent_t, 10)
+	for closed := false; !closed; {
+		kevents, err := w.read(eventBuffer)
+		// EINTR is okay, the syscall was interrupted before timeout expired.
+		if err != nil && err != unix.EINTR {
+			if !w.sendError(fmt.Errorf("fsnotify.readEvents: %w", err)) {
+				closed = true
+			}
+			continue
+		}
+
+		// Flush the events we received to the Events channel
+		for _, kevent := range kevents {
+			var (
+				watchfd = int(kevent.Ident)
+				mask    = uint32(kevent.Fflags)
+			)
+
+			// Shut down the loop when the pipe is closed, but only after all
+			// other events have been processed.
+			if watchfd == w.closepipe[0] {
+				closed = true
+				continue
+			}
+
+			w.mu.Lock()
+			path := w.paths[watchfd]
+			w.mu.Unlock()
+
+			event := w.newEvent(path.name, mask)
+
+			if event.Has(Rename) || event.Has(Remove) {
+				w.remove(event.Name, false)
+				w.mu.Lock()
+				delete(w.fileExists, event.Name)
+				w.mu.Unlock()
+			}
+
+			if path.isDir && event.Has(Write) && !event.Has(Remove) {
+				w.sendDirectoryChangeEvents(event.Name)
+			} else {
+				if !w.sendEvent(event) {
+					closed = true
+					continue
+				}
+			}
+
+			if event.Has(Remove) {
+				// Look for a file that may have overwritten this; for example,
+				// mv f1 f2 will delete f2, then create f2.
+				if path.isDir {
+					fileDir := filepath.Clean(event.Name)
+					w.mu.Lock()
+					_, found := w.watches[fileDir]
+					w.mu.Unlock()
+					if found {
+						err := w.sendDirectoryChangeEvents(fileDir)
+						if err != nil {
+							if !w.sendError(err) {
+								closed = true
+							}
+						}
+					}
+				} else {
+					filePath := filepath.Clean(event.Name)
+					if fi, err := os.Lstat(filePath); err == nil {
+						err := w.sendFileCreatedEventIfNew(filePath, fi)
+						if err != nil {
+							if !w.sendError(err) {
+								closed = true
+							}
+						}
+					}
+				}
+			}
+		}
+	}
+}
+
+// newEvent returns an platform-independent Event based on kqueue Fflags.
+func (w *Watcher) newEvent(name string, mask uint32) Event {
+	e := Event{Name: name}
+	if mask&unix.NOTE_DELETE == unix.NOTE_DELETE {
+		e.Op |= Remove
+	}
+	if mask&unix.NOTE_WRITE == unix.NOTE_WRITE {
+		e.Op |= Write
+	}
+	if mask&unix.NOTE_RENAME == unix.NOTE_RENAME {
+		e.Op |= Rename
+	}
+	if mask&unix.NOTE_ATTRIB == unix.NOTE_ATTRIB {
+		e.Op |= Chmod
+	}
+	// No point sending a write and delete event at the same time: if it's gone,
+	// then it's gone.
+	if e.Op.Has(Write) && e.Op.Has(Remove) {
+		e.Op &^= Write
+	}
+	return e
+}
+
+// watchDirectoryFiles to mimic inotify when adding a watch on a directory
+func (w *Watcher) watchDirectoryFiles(dirPath string) error {
+	// Get all files
+	files, err := os.ReadDir(dirPath)
+	if err != nil {
+		return err
+	}
+
+	for _, f := range files {
+		path := filepath.Join(dirPath, f.Name())
+
+		fi, err := f.Info()
+		if err != nil {
+			return fmt.Errorf("%q: %w", path, err)
+		}
+
+		cleanPath, err := w.internalWatch(path, fi)
+		if err != nil {
+			// No permission to read the file; that's not a problem: just skip.
+			// But do add it to w.fileExists to prevent it from being picked up
+			// as a "new" file later (it still shows up in the directory
+			// listing).
+			switch {
+			case errors.Is(err, unix.EACCES) || errors.Is(err, unix.EPERM):
+				cleanPath = filepath.Clean(path)
+			default:
+				return fmt.Errorf("%q: %w", path, err)
+			}
+		}
+
+		w.mu.Lock()
+		w.fileExists[cleanPath] = struct{}{}
+		w.mu.Unlock()
+	}
+
+	return nil
+}
+
+// Search the directory for new files and send an event for them.
+//
+// This functionality is to have the BSD watcher match the inotify, which sends
+// a create event for files created in a watched directory.
+func (w *Watcher) sendDirectoryChangeEvents(dir string) error {
+	files, err := os.ReadDir(dir)
+	if err != nil {
+		// Directory no longer exists: we can ignore this safely. kqueue will
+		// still give us the correct events.
+		if errors.Is(err, os.ErrNotExist) {
+			return nil
+		}
+		return fmt.Errorf("fsnotify.sendDirectoryChangeEvents: %w", err)
+	}
+
+	for _, f := range files {
+		fi, err := f.Info()
+		if err != nil {
+			return fmt.Errorf("fsnotify.sendDirectoryChangeEvents: %w", err)
+		}
+
+		err = w.sendFileCreatedEventIfNew(filepath.Join(dir, fi.Name()), fi)
+		if err != nil {
+			// Don't need to send an error if this file isn't readable.
+			if errors.Is(err, unix.EACCES) || errors.Is(err, unix.EPERM) {
+				return nil
+			}
+			return fmt.Errorf("fsnotify.sendDirectoryChangeEvents: %w", err)
+		}
+	}
+	return nil
+}
+
+// sendFileCreatedEvent sends a create event if the file isn't already being tracked.
+func (w *Watcher) sendFileCreatedEventIfNew(filePath string, fi os.FileInfo) (err error) {
+	w.mu.Lock()
+	_, doesExist := w.fileExists[filePath]
+	w.mu.Unlock()
+	if !doesExist {
+		if !w.sendEvent(Event{Name: filePath, Op: Create}) {
+			return
+		}
+	}
+
+	// like watchDirectoryFiles (but without doing another ReadDir)
+	filePath, err = w.internalWatch(filePath, fi)
+	if err != nil {
+		return err
+	}
+
+	w.mu.Lock()
+	w.fileExists[filePath] = struct{}{}
+	w.mu.Unlock()
+
+	return nil
+}
+
+func (w *Watcher) internalWatch(name string, fi os.FileInfo) (string, error) {
+	if fi.IsDir() {
+		// mimic Linux providing delete events for subdirectories, but preserve
+		// the flags used if currently watching subdirectory
+		w.mu.Lock()
+		flags := w.dirFlags[name]
+		w.mu.Unlock()
+
+		flags |= unix.NOTE_DELETE | unix.NOTE_RENAME
+		return w.addWatch(name, flags)
+	}
+
+	// watch file to mimic Linux inotify
+	return w.addWatch(name, noteAllEvents)
+}
+
+// Register events with the queue.
+func (w *Watcher) register(fds []int, flags int, fflags uint32) error {
+	changes := make([]unix.Kevent_t, len(fds))
+	for i, fd := range fds {
+		// SetKevent converts int to the platform-specific types.
+		unix.SetKevent(&changes[i], fd, unix.EVFILT_VNODE, flags)
+		changes[i].Fflags = fflags
+	}
+
+	// Register the events.
+	success, err := unix.Kevent(w.kq, changes, nil, nil)
+	if success == -1 {
+		return err
+	}
+	return nil
+}
+
+// read retrieves pending events, or waits until an event occurs.
+func (w *Watcher) read(events []unix.Kevent_t) ([]unix.Kevent_t, error) {
+	n, err := unix.Kevent(w.kq, nil, events, nil)
+	if err != nil {
+		return nil, err
+	}
+	return events[0:n], nil
+}

vendor/github.com/fsnotify/fsnotify/backend_other.go

+//go:build appengine || (!darwin && !dragonfly && !freebsd && !openbsd && !linux && !netbsd && !solaris && !windows)
+// +build appengine !darwin,!dragonfly,!freebsd,!openbsd,!linux,!netbsd,!solaris,!windows
+
+// Note: the documentation on the Watcher type and methods is generated from
+// mkdoc.zsh
+
+package fsnotify
+
+import "errors"
+
+// Watcher watches a set of paths, delivering events on a channel.
+//
+// A watcher should not be copied (e.g. pass it by pointer, rather than by
+// value).
+//
+// # Linux notes
+//
+// When a file is removed a Remove event won't be emitted until all file
+// descriptors are closed, and deletes will always emit a Chmod. For example:
+//
+//	fp := os.Open("file")
+//	os.Remove("file")        // Triggers Chmod
+//	fp.Close()               // Triggers Remove
+//
+// This is the event that inotify sends, so not much can be changed about this.
+//
+// The fs.inotify.max_user_watches sysctl variable specifies the upper limit
+// for the number of watches per user, and fs.inotify.max_user_instances
+// specifies the maximum number of inotify instances per user. Every Watcher you
+// create is an "instance", and every path you add is a "watch".
+//
+// These are also exposed in /proc as /proc/sys/fs/inotify/max_user_watches and
+// /proc/sys/fs/inotify/max_user_instances
+//
+// To increase them you can use sysctl or write the value to the /proc file:
+//
+//	# Default values on Linux 5.18
+//	sysctl fs.inotify.max_user_watches=124983
+//	sysctl fs.inotify.max_user_instances=128
+//
+// To make the changes persist on reboot edit /etc/sysctl.conf or
+// /usr/lib/sysctl.d/50-default.conf (details differ per Linux distro; check
+// your distro's documentation):
+//
+//	fs.inotify.max_user_watches=124983
+//	fs.inotify.max_user_instances=128
+//
+// Reaching the limit will result in a "no space left on device" or "too many open
+// files" error.
+//
+// # kqueue notes (macOS, BSD)
+//
+// kqueue requires opening a file descriptor for every file that's being watched;
+// so if you're watching a directory with five files then that's six file
+// descriptors. You will run in to your system's "max open files" limit faster on
+// these platforms.
+//
+// The sysctl variables kern.maxfiles and kern.maxfilesperproc can be used to
+// control the maximum number of open files, as well as /etc/login.conf on BSD
+// systems.
+//
+// # Windows notes
+//
+// Paths can be added as "C:\path\to\dir", but forward slashes
+// ("C:/path/to/dir") will also work.
+//
+// When a watched directory is removed it will always send an event for the
+// directory itself, but may not send events for all files in that directory.
+// Sometimes it will send events for all times, sometimes it will send no
+// events, and often only for some files.
+//
+// The default ReadDirectoryChangesW() buffer size is 64K, which is the largest
+// value that is guaranteed to work with SMB filesystems. If you have many
+// events in quick succession this may not be enough, and you will have to use
+// [WithBufferSize] to increase the value.
+type Watcher struct {
+	// Events sends the filesystem change events.
+	//
+	// fsnotify can send the following events; a "path" here can refer to a
+	// file, directory, symbolic link, or special file like a FIFO.
+	//
+	//   fsnotify.Create    A new path was created; this may be followed by one
+	//                      or more Write events if data also gets written to a
+	//                      file.
+	//
+	//   fsnotify.Remove    A path was removed.
+	//
+	//   fsnotify.Rename    A path was renamed. A rename is always sent with the
+	//                      old path as Event.Name, and a Create event will be
+	//                      sent with the new name. Renames are only sent for
+	//                      paths that are currently watched; e.g. moving an
+	//                      unmonitored file into a monitored directory will
+	//                      show up as just a Create. Similarly, renaming a file
+	//                      to outside a monitored directory will show up as
+	//                      only a Rename.
+	//
+	//   fsnotify.Write     A file or named pipe was written to. A Truncate will
+	//                      also trigger a Write. A single "write action"
+	//                      initiated by the user may show up as one or multiple
+	//                      writes, depending on when the system syncs things to
+	//                      disk. For example when compiling a large Go program
+	//                      you may get hundreds of Write events, and you may
+	//                      want to wait until you've stopped receiving them
+	//                      (see the dedup example in cmd/fsnotify).
+	//
+	//                      Some systems may send Write event for directories
+	//                      when the directory content changes.
+	//
+	//   fsnotify.Chmod     Attributes were changed. On Linux this is also sent
+	//                      when a file is removed (or more accurately, when a
+	//                      link to an inode is removed). On kqueue it's sent
+	//                      when a file is truncated. On Windows it's never
+	//                      sent.
+	Events chan Event
+
+	// Errors sends any errors.
+	//
+	// ErrEventOverflow is used to indicate there are too many events:
+	//
+	//  - inotify:      There are too many queued events (fs.inotify.max_queued_events sysctl)
+	//  - windows:      The buffer size is too small; WithBufferSize() can be used to increase it.
+	//  - kqueue, fen:  Not used.
+	Errors chan error
+}
+
+// NewWatcher creates a new Watcher.
+func NewWatcher() (*Watcher, error) {
+	return nil, errors.New("fsnotify not supported on the current platform")
+}
+
+// NewBufferedWatcher creates a new Watcher with a buffered Watcher.Events
+// channel.
+//
+// The main use case for this is situations with a very large number of events
+// where the kernel buffer size can't be increased (e.g. due to lack of
+// permissions). An unbuffered Watcher will perform better for almost all use
+// cases, and whenever possible you will be better off increasing the kernel
+// buffers instead of adding a large userspace buffer.
+func NewBufferedWatcher(sz uint) (*Watcher, error) { return NewWatcher() }
+
+// Close removes all watches and closes the Events channel.
+func (w *Watcher) Close() error { return nil }
+
+// WatchList returns all paths explicitly added with [Watcher.Add] (and are not
+// yet removed).
+//
+// Returns nil if [Watcher.Close] was called.
+func (w *Watcher) WatchList() []string { return nil }
+
+// Add starts monitoring the path for changes.
+//
+// A path can only be watched once; watching it more than once is a no-op and will
+// not return an error. Paths that do not yet exist on the filesystem cannot be
+// watched.
+//
+// A watch will be automatically removed if the watched path is deleted or
+// renamed. The exception is the Windows backend, which doesn't remove the
+// watcher on renames.
+//
+// Notifications on network filesystems (NFS, SMB, FUSE, etc.) or special
+// filesystems (/proc, /sys, etc.) generally don't work.
+//
+// Returns [ErrClosed] if [Watcher.Close] was called.
+//
+// See [Watcher.AddWith] for a version that allows adding options.
+//
+// # Watching directories
+//
+// All files in a directory are monitored, including new files that are created
+// after the watcher is started. Subdirectories are not watched (i.e. it's
+// non-recursive).
+//
+// # Watching files
+//
+// Watching individual files (rather than directories) is generally not
+// recommended as many programs (especially editors) update files atomically: it
+// will write to a temporary file which is then moved to to destination,
+// overwriting the original (or some variant thereof). The watcher on the
+// original file is now lost, as that no longer exists.
+//
+// The upshot of this is that a power failure or crash won't leave a
+// half-written file.
+//
+// Watch the parent directory and use Event.Name to filter out files you're not
+// interested in. There is an example of this in cmd/fsnotify/file.go.
+func (w *Watcher) Add(name string) error { return nil }
+
+// AddWith is like [Watcher.Add], but allows adding options. When using Add()
+// the defaults described below are used.
+//
+// Possible options are:
+//
+//   - [WithBufferSize] sets the buffer size for the Windows backend; no-op on
+//     other platforms. The default is 64K (65536 bytes).
+func (w *Watcher) AddWith(name string, opts ...addOpt) error { return nil }
+
+// Remove stops monitoring the path for changes.
+//
+// Directories are always removed non-recursively. For example, if you added
+// /tmp/dir and /tmp/dir/subdir then you will need to remove both.
+//
+// Removing a path that has not yet been added returns [ErrNonExistentWatch].
+//
+// Returns nil if [Watcher.Close] was called.
+func (w *Watcher) Remove(name string) error { return nil }

vendor/github.com/fsnotify/fsnotify/backend_windows.go

+//go:build windows
+// +build windows
+
+// Windows backend based on ReadDirectoryChangesW()
+//
+// https://learn.microsoft.com/en-us/windows/win32/api/winbase/nf-winbase-readdirectorychangesw
+//
+// Note: the documentation on the Watcher type and methods is generated from
+// mkdoc.zsh
+
+package fsnotify
+
+import (
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"reflect"
+	"runtime"
+	"strings"
+	"sync"
+	"unsafe"
+
+	"golang.org/x/sys/windows"
+)
+
+// Watcher watches a set of paths, delivering events on a channel.
+//
+// A watcher should not be copied (e.g. pass it by pointer, rather than by
+// value).
+//
+// # Linux notes
+//
+// When a file is removed a Remove event won't be emitted until all file
+// descriptors are closed, and deletes will always emit a Chmod. For example:
+//
+//	fp := os.Open("file")
+//	os.Remove("file")        // Triggers Chmod
+//	fp.Close()               // Triggers Remove
+//
+// This is the event that inotify sends, so not much can be changed about this.
+//
+// The fs.inotify.max_user_watches sysctl variable specifies the upper limit
+// for the number of watches per user, and fs.inotify.max_user_instances
+// specifies the maximum number of inotify instances per user. Every Watcher you
+// create is an "instance", and every path you add is a "watch".
+//
+// These are also exposed in /proc as /proc/sys/fs/inotify/max_user_watches and
+// /proc/sys/fs/inotify/max_user_instances
+//
+// To increase them you can use sysctl or write the value to the /proc file:
+//
+//	# Default values on Linux 5.18
+//	sysctl fs.inotify.max_user_watches=124983
+//	sysctl fs.inotify.max_user_instances=128
+//
+// To make the changes persist on reboot edit /etc/sysctl.conf or
+// /usr/lib/sysctl.d/50-default.conf (details differ per Linux distro; check
+// your distro's documentation):
+//
+//	fs.inotify.max_user_watches=124983
+//	fs.inotify.max_user_instances=128
+//
+// Reaching the limit will result in a "no space left on device" or "too many open
+// files" error.
+//
+// # kqueue notes (macOS, BSD)
+//
+// kqueue requires opening a file descriptor for every file that's being watched;
+// so if you're watching a directory with five files then that's six file
+// descriptors. You will run in to your system's "max open files" limit faster on
+// these platforms.
+//
+// The sysctl variables kern.maxfiles and kern.maxfilesperproc can be used to
+// control the maximum number of open files, as well as /etc/login.conf on BSD
+// systems.
+//
+// # Windows notes
+//
+// Paths can be added as "C:\path\to\dir", but forward slashes
+// ("C:/path/to/dir") will also work.
+//
+// When a watched directory is removed it will always send an event for the
+// directory itself, but may not send events for all files in that directory.
+// Sometimes it will send events for all times, sometimes it will send no
+// events, and often only for some files.
+//
+// The default ReadDirectoryChangesW() buffer size is 64K, which is the largest
+// value that is guaranteed to work with SMB filesystems. If you have many
+// events in quick succession this may not be enough, and you will have to use
+// [WithBufferSize] to increase the value.
+type Watcher struct {
+	// Events sends the filesystem change events.
+	//
+	// fsnotify can send the following events; a "path" here can refer to a
+	// file, directory, symbolic link, or special file like a FIFO.
+	//
+	//   fsnotify.Create    A new path was created; this may be followed by one
+	//                      or more Write events if data also gets written to a
+	//                      file.
+	//
+	//   fsnotify.Remove    A path was removed.
+	//
+	//   fsnotify.Rename    A path was renamed. A rename is always sent with the
+	//                      old path as Event.Name, and a Create event will be
+	//                      sent with the new name. Renames are only sent for
+	//                      paths that are currently watched; e.g. moving an
+	//                      unmonitored file into a monitored directory will
+	//                      show up as just a Create. Similarly, renaming a file
+	//                      to outside a monitored directory will show up as
+	//                      only a Rename.
+	//
+	//   fsnotify.Write     A file or named pipe was written to. A Truncate will
+	//                      also trigger a Write. A single "write action"
+	//                      initiated by the user may show up as one or multiple
+	//                      writes, depending on when the system syncs things to
+	//                      disk. For example when compiling a large Go program
+	//                      you may get hundreds of Write events, and you may
+	//                      want to wait until you've stopped receiving them
+	//                      (see the dedup example in cmd/fsnotify).
+	//
+	//                      Some systems may send Write event for directories
+	//                      when the directory content changes.
+	//
+	//   fsnotify.Chmod     Attributes were changed. On Linux this is also sent
+	//                      when a file is removed (or more accurately, when a
+	//                      link to an inode is removed). On kqueue it's sent
+	//                      when a file is truncated. On Windows it's never
+	//                      sent.
+	Events chan Event
+
+	// Errors sends any errors.
+	//
+	// ErrEventOverflow is used to indicate there are too many events:
+	//
+	//  - inotify:      There are too many queued events (fs.inotify.max_queued_events sysctl)
+	//  - windows:      The buffer size is too small; WithBufferSize() can be used to increase it.
+	//  - kqueue, fen:  Not used.
+	Errors chan error
+
+	port  windows.Handle // Handle to completion port
+	input chan *input    // Inputs to the reader are sent on this channel
+	quit  chan chan<- error
+
+	mu      sync.Mutex // Protects access to watches, closed
+	watches watchMap   // Map of watches (key: i-number)
+	closed  bool       // Set to true when Close() is first called
+}
+
+// NewWatcher creates a new Watcher.
+func NewWatcher() (*Watcher, error) {
+	return NewBufferedWatcher(50)
+}
+
+// NewBufferedWatcher creates a new Watcher with a buffered Watcher.Events
+// channel.
+//
+// The main use case for this is situations with a very large number of events
+// where the kernel buffer size can't be increased (e.g. due to lack of
+// permissions). An unbuffered Watcher will perform better for almost all use
+// cases, and whenever possible you will be better off increasing the kernel
+// buffers instead of adding a large userspace buffer.
+func NewBufferedWatcher(sz uint) (*Watcher, error) {
+	port, err := windows.CreateIoCompletionPort(windows.InvalidHandle, 0, 0, 0)
+	if err != nil {
+		return nil, os.NewSyscallError("CreateIoCompletionPort", err)
+	}
+	w := &Watcher{
+		port:    port,
+		watches: make(watchMap),
+		input:   make(chan *input, 1),
+		Events:  make(chan Event, sz),
+		Errors:  make(chan error),
+		quit:    make(chan chan<- error, 1),
+	}
+	go w.readEvents()
+	return w, nil
+}
+
+func (w *Watcher) isClosed() bool {
+	w.mu.Lock()
+	defer w.mu.Unlock()
+	return w.closed
+}
+
+func (w *Watcher) sendEvent(name string, mask uint64) bool {
+	if mask == 0 {
+		return false
+	}
+
+	event := w.newEvent(name, uint32(mask))
+	select {
+	case ch := <-w.quit:
+		w.quit <- ch
+	case w.Events <- event:
+	}
+	return true
+}
+
+// Returns true if the error was sent, or false if watcher is closed.
+func (w *Watcher) sendError(err error) bool {
+	select {
+	case w.Errors <- err:
+		return true
+	case <-w.quit:
+	}
+	return false
+}
+
+// Close removes all watches and closes the Events channel.
+func (w *Watcher) Close() error {
+	if w.isClosed() {
+		return nil
+	}
+
+	w.mu.Lock()
+	w.closed = true
+	w.mu.Unlock()
+
+	// Send "quit" message to the reader goroutine
+	ch := make(chan error)
+	w.quit <- ch
+	if err := w.wakeupReader(); err != nil {
+		return err
+	}
+	return <-ch
+}
+
+// Add starts monitoring the path for changes.
+//
+// A path can only be watched once; watching it more than once is a no-op and will
+// not return an error. Paths that do not yet exist on the filesystem cannot be
+// watched.
+//
+// A watch will be automatically removed if the watched path is deleted or
+// renamed. The exception is the Windows backend, which doesn't remove the
+// watcher on renames.
+//
+// Notifications on network filesystems (NFS, SMB, FUSE, etc.) or special
+// filesystems (/proc, /sys, etc.) generally don't work.
+//
+// Returns [ErrClosed] if [Watcher.Close] was called.
+//
+// See [Watcher.AddWith] for a version that allows adding options.
+//
+// # Watching directories
+//
+// All files in a directory are monitored, including new files that are created
+// after the watcher is started. Subdirectories are not watched (i.e. it's
+// non-recursive).
+//
+// # Watching files
+//
+// Watching individual files (rather than directories) is generally not
+// recommended as many programs (especially editors) update files atomically: it
+// will write to a temporary file which is then moved to to destination,
+// overwriting the original (or some variant thereof). The watcher on the
+// original file is now lost, as that no longer exists.
+//
+// The upshot of this is that a power failure or crash won't leave a
+// half-written file.
+//
+// Watch the parent directory and use Event.Name to filter out files you're not
+// interested in. There is an example of this in cmd/fsnotify/file.go.
+func (w *Watcher) Add(name string) error { return w.AddWith(name) }
+
+// AddWith is like [Watcher.Add], but allows adding options. When using Add()
+// the defaults described below are used.
+//
+// Possible options are:
+//
+//   - [WithBufferSize] sets the buffer size for the Windows backend; no-op on
+//     other platforms. The default is 64K (65536 bytes).
+func (w *Watcher) AddWith(name string, opts ...addOpt) error {
+	if w.isClosed() {
+		return ErrClosed
+	}
+
+	with := getOptions(opts...)
+	if with.bufsize < 4096 {
+		return fmt.Errorf("fsnotify.WithBufferSize: buffer size cannot be smaller than 4096 bytes")
+	}
+
+	in := &input{
+		op:      opAddWatch,
+		path:    filepath.Clean(name),
+		flags:   sysFSALLEVENTS,
+		reply:   make(chan error),
+		bufsize: with.bufsize,
+	}
+	w.input <- in
+	if err := w.wakeupReader(); err != nil {
+		return err
+	}
+	return <-in.reply
+}
+
+// Remove stops monitoring the path for changes.
+//
+// Directories are always removed non-recursively. For example, if you added
+// /tmp/dir and /tmp/dir/subdir then you will need to remove both.
+//
+// Removing a path that has not yet been added returns [ErrNonExistentWatch].
+//
+// Returns nil if [Watcher.Close] was called.
+func (w *Watcher) Remove(name string) error {
+	if w.isClosed() {
+		return nil
+	}
+
+	in := &input{
+		op:    opRemoveWatch,
+		path:  filepath.Clean(name),
+		reply: make(chan error),
+	}
+	w.input <- in
+	if err := w.wakeupReader(); err != nil {
+		return err
+	}
+	return <-in.reply
+}
+
+// WatchList returns all paths explicitly added with [Watcher.Add] (and are not
+// yet removed).
+//
+// Returns nil if [Watcher.Close] was called.
+func (w *Watcher) WatchList() []string {
+	if w.isClosed() {
+		return nil
+	}
+
+	w.mu.Lock()
+	defer w.mu.Unlock()
+
+	entries := make([]string, 0, len(w.watches))
+	for _, entry := range w.watches {
+		for _, watchEntry := range entry {
+			entries = append(entries, watchEntry.path)
+		}
+	}
+
+	return entries
+}
+
+// These options are from the old golang.org/x/exp/winfsnotify, where you could
+// add various options to the watch. This has long since been removed.
+//
+// The "sys" in the name is misleading as they're not part of any "system".
+//
+// This should all be removed at some point, and just use windows.FILE_NOTIFY_*
+const (
+	sysFSALLEVENTS  = 0xfff
+	sysFSCREATE     = 0x100
+	sysFSDELETE     = 0x200
+	sysFSDELETESELF = 0x400
+	sysFSMODIFY     = 0x2
+	sysFSMOVE       = 0xc0
+	sysFSMOVEDFROM  = 0x40
+	sysFSMOVEDTO    = 0x80
+	sysFSMOVESELF   = 0x800
+	sysFSIGNORED    = 0x8000
+)
+
+func (w *Watcher) newEvent(name string, mask uint32) Event {
+	e := Event{Name: name}
+	if mask&sysFSCREATE == sysFSCREATE || mask&sysFSMOVEDTO == sysFSMOVEDTO {
+		e.Op |= Create
+	}
+	if mask&sysFSDELETE == sysFSDELETE || mask&sysFSDELETESELF == sysFSDELETESELF {
+		e.Op |= Remove
+	}
+	if mask&sysFSMODIFY == sysFSMODIFY {
+		e.Op |= Write
+	}
+	if mask&sysFSMOVE == sysFSMOVE || mask&sysFSMOVESELF == sysFSMOVESELF || mask&sysFSMOVEDFROM == sysFSMOVEDFROM {
+		e.Op |= Rename
+	}
+	return e
+}
+
+const (
+	opAddWatch = iota
+	opRemoveWatch
+)
+
+const (
+	provisional uint64 = 1 << (32 + iota)
+)
+
+type input struct {
+	op      int
+	path    string
+	flags   uint32
+	bufsize int
+	reply   chan error
+}
+
+type inode struct {
+	handle windows.Handle
+	volume uint32
+	index  uint64
+}
+
+type watch struct {
+	ov      windows.Overlapped
+	ino     *inode            // i-number
+	recurse bool              // Recursive watch?
+	path    string            // Directory path
+	mask    uint64            // Directory itself is being watched with these notify flags
+	names   map[string]uint64 // Map of names being watched and their notify flags
+	rename  string            // Remembers the old name while renaming a file
+	buf     []byte            // buffer, allocated later
+}
+
+type (
+	indexMap map[uint64]*watch
+	watchMap map[uint32]indexMap
+)
+
+func (w *Watcher) wakeupReader() error {
+	err := windows.PostQueuedCompletionStatus(w.port, 0, 0, nil)
+	if err != nil {
+		return os.NewSyscallError("PostQueuedCompletionStatus", err)
+	}
+	return nil
+}
+
+func (w *Watcher) getDir(pathname string) (dir string, err error) {
+	attr, err := windows.GetFileAttributes(windows.StringToUTF16Ptr(pathname))
+	if err != nil {
+		return "", os.NewSyscallError("GetFileAttributes", err)
+	}
+	if attr&windows.FILE_ATTRIBUTE_DIRECTORY != 0 {
+		dir = pathname
+	} else {
+		dir, _ = filepath.Split(pathname)
+		dir = filepath.Clean(dir)
+	}
+	return
+}
+
+func (w *Watcher) getIno(path string) (ino *inode, err error) {
+	h, err := windows.CreateFile(windows.StringToUTF16Ptr(path),
+		windows.FILE_LIST_DIRECTORY,
+		windows.FILE_SHARE_READ|windows.FILE_SHARE_WRITE|windows.FILE_SHARE_DELETE,
+		nil, windows.OPEN_EXISTING,
+		windows.FILE_FLAG_BACKUP_SEMANTICS|windows.FILE_FLAG_OVERLAPPED, 0)
+	if err != nil {
+		return nil, os.NewSyscallError("CreateFile", err)
+	}
+
+	var fi windows.ByHandleFileInformation
+	err = windows.GetFileInformationByHandle(h, &fi)
+	if err != nil {
+		windows.CloseHandle(h)
+		return nil, os.NewSyscallError("GetFileInformationByHandle", err)
+	}
+	ino = &inode{
+		handle: h,
+		volume: fi.VolumeSerialNumber,
+		index:  uint64(fi.FileIndexHigh)<<32 | uint64(fi.FileIndexLow),
+	}
+	return ino, nil
+}
+
+// Must run within the I/O thread.
+func (m watchMap) get(ino *inode) *watch {
+	if i := m[ino.volume]; i != nil {
+		return i[ino.index]
+	}
+	return nil
+}
+
+// Must run within the I/O thread.
+func (m watchMap) set(ino *inode, watch *watch) {
+	i := m[ino.volume]
+	if i == nil {
+		i = make(indexMap)
+		m[ino.volume] = i
+	}
+	i[ino.index] = watch
+}
+
+// Must run within the I/O thread.
+func (w *Watcher) addWatch(pathname string, flags uint64, bufsize int) error {
+	//pathname, recurse := recursivePath(pathname)
+	recurse := false
+
+	dir, err := w.getDir(pathname)
+	if err != nil {
+		return err
+	}
+
+	ino, err := w.getIno(dir)
+	if err != nil {
+		return err
+	}
+	w.mu.Lock()
+	watchEntry := w.watches.get(ino)
+	w.mu.Unlock()
+	if watchEntry == nil {
+		_, err := windows.CreateIoCompletionPort(ino.handle, w.port, 0, 0)
+		if err != nil {
+			windows.CloseHandle(ino.handle)
+			return os.NewSyscallError("CreateIoCompletionPort", err)
+		}
+		watchEntry = &watch{
+			ino:     ino,
+			path:    dir,
+			names:   make(map[string]uint64),
+			recurse: recurse,
+			buf:     make([]byte, bufsize),
+		}
+		w.mu.Lock()
+		w.watches.set(ino, watchEntry)
+		w.mu.Unlock()
+		flags |= provisional
+	} else {
+		windows.CloseHandle(ino.handle)
+	}
+	if pathname == dir {
+		watchEntry.mask |= flags
+	} else {
+		watchEntry.names[filepath.Base(pathname)] |= flags
+	}
+
+	err = w.startRead(watchEntry)
+	if err != nil {
+		return err
+	}
+
+	if pathname == dir {
+		watchEntry.mask &= ^provisional
+	} else {
+		watchEntry.names[filepath.Base(pathname)] &= ^provisional
+	}
+	return nil
+}
+
+// Must run within the I/O thread.
+func (w *Watcher) remWatch(pathname string) error {
+	pathname, recurse := recursivePath(pathname)
+
+	dir, err := w.getDir(pathname)
+	if err != nil {
+		return err
+	}
+	ino, err := w.getIno(dir)
+	if err != nil {
+		return err
+	}
+
+	w.mu.Lock()
+	watch := w.watches.get(ino)
+	w.mu.Unlock()
+
+	if recurse && !watch.recurse {
+		return fmt.Errorf("can't use \\... with non-recursive watch %q", pathname)
+	}
+
+	err = windows.CloseHandle(ino.handle)
+	if err != nil {
+		w.sendError(os.NewSyscallError("CloseHandle", err))
+	}
+	if watch == nil {
+		return fmt.Errorf("%w: %s", ErrNonExistentWatch, pathname)
+	}
+	if pathname == dir {
+		w.sendEvent(watch.path, watch.mask&sysFSIGNORED)
+		watch.mask = 0
+	} else {
+		name := filepath.Base(pathname)
+		w.sendEvent(filepath.Join(watch.path, name), watch.names[name]&sysFSIGNORED)
+		delete(watch.names, name)
+	}
+
+	return w.startRead(watch)
+}
+
+// Must run within the I/O thread.
+func (w *Watcher) deleteWatch(watch *watch) {
+	for name, mask := range watch.names {
+		if mask&provisional == 0 {
+			w.sendEvent(filepath.Join(watch.path, name), mask&sysFSIGNORED)
+		}
+		delete(watch.names, name)
+	}
+	if watch.mask != 0 {
+		if watch.mask&provisional == 0 {
+			w.sendEvent(watch.path, watch.mask&sysFSIGNORED)
+		}
+		watch.mask = 0
+	}
+}
+
+// Must run within the I/O thread.
+func (w *Watcher) startRead(watch *watch) error {
+	err := windows.CancelIo(watch.ino.handle)
+	if err != nil {
+		w.sendError(os.NewSyscallError("CancelIo", err))
+		w.deleteWatch(watch)
+	}
+	mask := w.toWindowsFlags(watch.mask)
+	for _, m := range watch.names {
+		mask |= w.toWindowsFlags(m)
+	}
+	if mask == 0 {
+		err := windows.CloseHandle(watch.ino.handle)
+		if err != nil {
+			w.sendError(os.NewSyscallError("CloseHandle", err))
+		}
+		w.mu.Lock()
+		delete(w.watches[watch.ino.volume], watch.ino.index)
+		w.mu.Unlock()
+		return nil
+	}
+
+	// We need to pass the array, rather than the slice.
+	hdr := (*reflect.SliceHeader)(unsafe.Pointer(&watch.buf))
+	rdErr := windows.ReadDirectoryChanges(watch.ino.handle,
+		(*byte)(unsafe.Pointer(hdr.Data)), uint32(hdr.Len),
+		watch.recurse, mask, nil, &watch.ov, 0)
+	if rdErr != nil {
+		err := os.NewSyscallError("ReadDirectoryChanges", rdErr)
+		if rdErr == windows.ERROR_ACCESS_DENIED && watch.mask&provisional == 0 {
+			// Watched directory was probably removed
+			w.sendEvent(watch.path, watch.mask&sysFSDELETESELF)
+			err = nil
+		}
+		w.deleteWatch(watch)
+		w.startRead(watch)
+		return err
+	}
+	return nil
+}
+
+// readEvents reads from the I/O completion port, converts the
+// received events into Event objects and sends them via the Events channel.
+// Entry point to the I/O thread.
+func (w *Watcher) readEvents() {
+	var (
+		n   uint32
+		key uintptr
+		ov  *windows.Overlapped
+	)
+	runtime.LockOSThread()
+
+	for {
+		// This error is handled after the watch == nil check below.
+		qErr := windows.GetQueuedCompletionStatus(w.port, &n, &key, &ov, windows.INFINITE)
+
+		watch := (*watch)(unsafe.Pointer(ov))
+		if watch == nil {
+			select {
+			case ch := <-w.quit:
+				w.mu.Lock()
+				var indexes []indexMap
+				for _, index := range w.watches {
+					indexes = append(indexes, index)
+				}
+				w.mu.Unlock()
+				for _, index := range indexes {
+					for _, watch := range index {
+						w.deleteWatch(watch)
+						w.startRead(watch)
+					}
+				}
+
+				err := windows.CloseHandle(w.port)
+				if err != nil {
+					err = os.NewSyscallError("CloseHandle", err)
+				}
+				close(w.Events)
+				close(w.Errors)
+				ch <- err
+				return
+			case in := <-w.input:
+				switch in.op {
+				case opAddWatch:
+					in.reply <- w.addWatch(in.path, uint64(in.flags), in.bufsize)
+				case opRemoveWatch:
+					in.reply <- w.remWatch(in.path)
+				}
+			default:
+			}
+			continue
+		}
+
+		switch qErr {
+		case nil:
+			// No error
+		case windows.ERROR_MORE_DATA:
+			if watch == nil {
+				w.sendError(errors.New("ERROR_MORE_DATA has unexpectedly null lpOverlapped buffer"))
+			} else {
+				// The i/o succeeded but the buffer is full.
+				// In theory we should be building up a full packet.
+				// In practice we can get away with just carrying on.
+				n = uint32(unsafe.Sizeof(watch.buf))
+			}
+		case windows.ERROR_ACCESS_DENIED:
+			// Watched directory was probably removed
+			w.sendEvent(watch.path, watch.mask&sysFSDELETESELF)
+			w.deleteWatch(watch)
+			w.startRead(watch)
+			continue
+		case windows.ERROR_OPERATION_ABORTED:
+			// CancelIo was called on this handle
+			continue
+		default:
+			w.sendError(os.NewSyscallError("GetQueuedCompletionPort", qErr))
+			continue
+		}
+
+		var offset uint32
+		for {
+			if n == 0 {
+				w.sendError(ErrEventOverflow)
+				break
+			}
+
+			// Point "raw" to the event in the buffer
+			raw := (*windows.FileNotifyInformation)(unsafe.Pointer(&watch.buf[offset]))
+
+			// Create a buf that is the size of the path name
+			size := int(raw.FileNameLength / 2)
+			var buf []uint16
+			// TODO: Use unsafe.Slice in Go 1.17; https://stackoverflow.com/questions/51187973
+			sh := (*reflect.SliceHeader)(unsafe.Pointer(&buf))
+			sh.Data = uintptr(unsafe.Pointer(&raw.FileName))
+			sh.Len = size
+			sh.Cap = size
+			name := windows.UTF16ToString(buf)
+			fullname := filepath.Join(watch.path, name)
+
+			var mask uint64
+			switch raw.Action {
+			case windows.FILE_ACTION_REMOVED:
+				mask = sysFSDELETESELF
+			case windows.FILE_ACTION_MODIFIED:
+				mask = sysFSMODIFY
+			case windows.FILE_ACTION_RENAMED_OLD_NAME:
+				watch.rename = name
+			case windows.FILE_ACTION_RENAMED_NEW_NAME:
+				// Update saved path of all sub-watches.
+				old := filepath.Join(watch.path, watch.rename)
+				w.mu.Lock()
+				for _, watchMap := range w.watches {
+					for _, ww := range watchMap {
+						if strings.HasPrefix(ww.path, old) {
+							ww.path = filepath.Join(fullname, strings.TrimPrefix(ww.path, old))
+						}
+					}
+				}
+				w.mu.Unlock()
+
+				if watch.names[watch.rename] != 0 {
+					watch.names[name] |= watch.names[watch.rename]
+					delete(watch.names, watch.rename)
+					mask = sysFSMOVESELF
+				}
+			}
+
+			sendNameEvent := func() {
+				w.sendEvent(fullname, watch.names[name]&mask)
+			}
+			if raw.Action != windows.FILE_ACTION_RENAMED_NEW_NAME {
+				sendNameEvent()
+			}
+			if raw.Action == windows.FILE_ACTION_REMOVED {
+				w.sendEvent(fullname, watch.names[name]&sysFSIGNORED)
+				delete(watch.names, name)
+			}
+
+			w.sendEvent(fullname, watch.mask&w.toFSnotifyFlags(raw.Action))
+			if raw.Action == windows.FILE_ACTION_RENAMED_NEW_NAME {
+				fullname = filepath.Join(watch.path, watch.rename)
+				sendNameEvent()
+			}
+
+			// Move to the next event in the buffer
+			if raw.NextEntryOffset == 0 {
+				break
+			}
+			offset += raw.NextEntryOffset
+
+			// Error!
+			if offset >= n {
+				//lint:ignore ST1005 Windows should be capitalized
+				w.sendError(errors.New(
+					"Windows system assumed buffer larger than it is, events have likely been missed"))
+				break
+			}
+		}
+
+		if err := w.startRead(watch); err != nil {
+			w.sendError(err)
+		}
+	}
+}
+
+func (w *Watcher) toWindowsFlags(mask uint64) uint32 {
+	var m uint32
+	if mask&sysFSMODIFY != 0 {
+		m |= windows.FILE_NOTIFY_CHANGE_LAST_WRITE
+	}
+	if mask&(sysFSMOVE|sysFSCREATE|sysFSDELETE) != 0 {
+		m |= windows.FILE_NOTIFY_CHANGE_FILE_NAME | windows.FILE_NOTIFY_CHANGE_DIR_NAME
+	}
+	return m
+}
+
+func (w *Watcher) toFSnotifyFlags(action uint32) uint64 {
+	switch action {
+	case windows.FILE_ACTION_ADDED:
+		return sysFSCREATE
+	case windows.FILE_ACTION_REMOVED:
+		return sysFSDELETE
+	case windows.FILE_ACTION_MODIFIED:
+		return sysFSMODIFY
+	case windows.FILE_ACTION_RENAMED_OLD_NAME:
+		return sysFSMOVEDFROM
+	case windows.FILE_ACTION_RENAMED_NEW_NAME:
+		return sysFSMOVEDTO
+	}
+	return 0
+}

vendor/github.com/fsnotify/fsnotify/fsnotify.go

+// Package fsnotify provides a cross-platform interface for file system
+// notifications.
+//
+// Currently supported systems:
+//
+//	Linux 2.6.32+    via inotify
+//	BSD, macOS       via kqueue
+//	Windows          via ReadDirectoryChangesW
+//	illumos          via FEN
+package fsnotify
+
+import (
+	"errors"
+	"fmt"
+	"path/filepath"
+	"strings"
+)
+
+// Event represents a file system notification.
+type Event struct {
+	// Path to the file or directory.
+	//
+	// Paths are relative to the input; for example with Add("dir") the Name
+	// will be set to "dir/file" if you create that file, but if you use
+	// Add("/path/to/dir") it will be "/path/to/dir/file".
+	Name string
+
+	// File operation that triggered the event.
+	//
+	// This is a bitmask and some systems may send multiple operations at once.
+	// Use the Event.Has() method instead of comparing with ==.
+	Op Op
+}
+
+// Op describes a set of file operations.
+type Op uint32
+
+// The operations fsnotify can trigger; see the documentation on [Watcher] for a
+// full description, and check them with [Event.Has].
+const (
+	// A new pathname was created.
+	Create Op = 1 << iota
+
+	// The pathname was written to; this does *not* mean the write has finished,
+	// and a write can be followed by more writes.
+	Write
+
+	// The path was removed; any watches on it will be removed. Some "remove"
+	// operations may trigger a Rename if the file is actually moved (for
+	// example "remove to trash" is often a rename).
+	Remove
+
+	// The path was renamed to something else; any watched on it will be
+	// removed.
+	Rename
+
+	// File attributes were changed.
+	//
+	// It's generally not recommended to take action on this event, as it may
+	// get triggered very frequently by some software. For example, Spotlight
+	// indexing on macOS, anti-virus software, backup software, etc.
+	Chmod
+)
+
+// Common errors that can be reported.
+var (
+	ErrNonExistentWatch = errors.New("fsnotify: can't remove non-existent watch")
+	ErrEventOverflow    = errors.New("fsnotify: queue or buffer overflow")
+	ErrClosed           = errors.New("fsnotify: watcher already closed")
+)
+
+func (o Op) String() string {
+	var b strings.Builder
+	if o.Has(Create) {
+		b.WriteString("|CREATE")
+	}
+	if o.Has(Remove) {
+		b.WriteString("|REMOVE")
+	}
+	if o.Has(Write) {
+		b.WriteString("|WRITE")
+	}
+	if o.Has(Rename) {
+		b.WriteString("|RENAME")
+	}
+	if o.Has(Chmod) {
+		b.WriteString("|CHMOD")
+	}
+	if b.Len() == 0 {
+		return "[no events]"
+	}
+	return b.String()[1:]
+}
+
+// Has reports if this operation has the given operation.
+func (o Op) Has(h Op) bool { return o&h != 0 }
+
+// Has reports if this event has the given operation.
+func (e Event) Has(op Op) bool { return e.Op.Has(op) }
+
+// String returns a string representation of the event with their path.
+func (e Event) String() string {
+	return fmt.Sprintf("%-13s %q", e.Op.String(), e.Name)
+}
+
+type (
+	addOpt   func(opt *withOpts)
+	withOpts struct {
+		bufsize int
+	}
+)
+
+var defaultOpts = withOpts{
+	bufsize: 65536, // 64K
+}
+
+func getOptions(opts ...addOpt) withOpts {
+	with := defaultOpts
+	for _, o := range opts {
+		o(&with)
+	}
+	return with
+}
+
+// WithBufferSize sets the [ReadDirectoryChangesW] buffer size.
+//
+// This only has effect on Windows systems, and is a no-op for other backends.
+//
+// The default value is 64K (65536 bytes) which is the highest value that works
+// on all filesystems and should be enough for most applications, but if you
+// have a large burst of events it may not be enough. You can increase it if
+// you're hitting "queue or buffer overflow" errors ([ErrEventOverflow]).
+//
+// [ReadDirectoryChangesW]: https://learn.microsoft.com/en-gb/windows/win32/api/winbase/nf-winbase-readdirectorychangesw
+func WithBufferSize(bytes int) addOpt {
+	return func(opt *withOpts) { opt.bufsize = bytes }
+}
+
+// Check if this path is recursive (ends with "/..." or "\..."), and return the
+// path with the /... stripped.
+func recursivePath(path string) (string, bool) {
+	if filepath.Base(path) == "..." {
+		return filepath.Dir(path), true
+	}
+	return path, false
+}

vendor/github.com/fsnotify/fsnotify/mkdoc.zsh

+#!/usr/bin/env zsh
+[ "${ZSH_VERSION:-}" = "" ] && echo >&2 "Only works with zsh" && exit 1
+setopt err_exit no_unset pipefail extended_glob
+
+# Simple script to update the godoc comments on all watchers so you don't need
+# to update the same comment 5 times.
+
+watcher=$(<<EOF
+// Watcher watches a set of paths, delivering events on a channel.
+//
+// A watcher should not be copied (e.g. pass it by pointer, rather than by
+// value).
+//
+// # Linux notes
+//
+// When a file is removed a Remove event won't be emitted until all file
+// descriptors are closed, and deletes will always emit a Chmod. For example:
+//
+//	fp := os.Open("file")
+//	os.Remove("file")        // Triggers Chmod
+//	fp.Close()               // Triggers Remove
+//
+// This is the event that inotify sends, so not much can be changed about this.
+//
+// The fs.inotify.max_user_watches sysctl variable specifies the upper limit
+// for the number of watches per user, and fs.inotify.max_user_instances
+// specifies the maximum number of inotify instances per user. Every Watcher you
+// create is an "instance", and every path you add is a "watch".
+//
+// These are also exposed in /proc as /proc/sys/fs/inotify/max_user_watches and
+// /proc/sys/fs/inotify/max_user_instances
+//
+// To increase them you can use sysctl or write the value to the /proc file:
+//
+//	# Default values on Linux 5.18
+//	sysctl fs.inotify.max_user_watches=124983
+//	sysctl fs.inotify.max_user_instances=128
+//
+// To make the changes persist on reboot edit /etc/sysctl.conf or
+// /usr/lib/sysctl.d/50-default.conf (details differ per Linux distro; check
+// your distro's documentation):
+//
+//	fs.inotify.max_user_watches=124983
+//	fs.inotify.max_user_instances=128
+//
+// Reaching the limit will result in a "no space left on device" or "too many open
+// files" error.
+//
+// # kqueue notes (macOS, BSD)
+//
+// kqueue requires opening a file descriptor for every file that's being watched;
+// so if you're watching a directory with five files then that's six file
+// descriptors. You will run in to your system's "max open files" limit faster on
+// these platforms.
+//
+// The sysctl variables kern.maxfiles and kern.maxfilesperproc can be used to
+// control the maximum number of open files, as well as /etc/login.conf on BSD
+// systems.
+//
+// # Windows notes
+//
+// Paths can be added as "C:\\path\\to\\dir", but forward slashes
+// ("C:/path/to/dir") will also work.
+//
+// When a watched directory is removed it will always send an event for the
+// directory itself, but may not send events for all files in that directory.
+// Sometimes it will send events for all times, sometimes it will send no
+// events, and often only for some files.
+//
+// The default ReadDirectoryChangesW() buffer size is 64K, which is the largest
+// value that is guaranteed to work with SMB filesystems. If you have many
+// events in quick succession this may not be enough, and you will have to use
+// [WithBufferSize] to increase the value.
+EOF
+)
+
+new=$(<<EOF
+// NewWatcher creates a new Watcher.
+EOF
+)
+
+newbuffered=$(<<EOF
+// NewBufferedWatcher creates a new Watcher with a buffered Watcher.Events
+// channel.
+//
+// The main use case for this is situations with a very large number of events
+// where the kernel buffer size can't be increased (e.g. due to lack of
+// permissions). An unbuffered Watcher will perform better for almost all use
+// cases, and whenever possible you will be better off increasing the kernel
+// buffers instead of adding a large userspace buffer.
+EOF
+)
+
+add=$(<<EOF
+// Add starts monitoring the path for changes.
+//
+// A path can only be watched once; watching it more than once is a no-op and will
+// not return an error. Paths that do not yet exist on the filesystem cannot be
+// watched.
+//
+// A watch will be automatically removed if the watched path is deleted or
+// renamed. The exception is the Windows backend, which doesn't remove the
+// watcher on renames.
+//
+// Notifications on network filesystems (NFS, SMB, FUSE, etc.) or special
+// filesystems (/proc, /sys, etc.) generally don't work.
+//
+// Returns [ErrClosed] if [Watcher.Close] was called.
+//
+// See [Watcher.AddWith] for a version that allows adding options.
+//
+// # Watching directories
+//
+// All files in a directory are monitored, including new files that are created
+// after the watcher is started. Subdirectories are not watched (i.e. it's
+// non-recursive).
+//
+// # Watching files
+//
+// Watching individual files (rather than directories) is generally not
+// recommended as many programs (especially editors) update files atomically: it
+// will write to a temporary file which is then moved to to destination,
+// overwriting the original (or some variant thereof). The watcher on the
+// original file is now lost, as that no longer exists.
+//
+// The upshot of this is that a power failure or crash won't leave a
+// half-written file.
+//
+// Watch the parent directory and use Event.Name to filter out files you're not
+// interested in. There is an example of this in cmd/fsnotify/file.go.
+EOF
+)
+
+addwith=$(<<EOF
+// AddWith is like [Watcher.Add], but allows adding options. When using Add()
+// the defaults described below are used.
+//
+// Possible options are:
+//
+//   - [WithBufferSize] sets the buffer size for the Windows backend; no-op on
+//     other platforms. The default is 64K (65536 bytes).
+EOF
+)
+
+remove=$(<<EOF
+// Remove stops monitoring the path for changes.
+//
+// Directories are always removed non-recursively. For example, if you added
+// /tmp/dir and /tmp/dir/subdir then you will need to remove both.
+//
+// Removing a path that has not yet been added returns [ErrNonExistentWatch].
+//
+// Returns nil if [Watcher.Close] was called.
+EOF
+)
+
+close=$(<<EOF
+// Close removes all watches and closes the Events channel.
+EOF
+)
+
+watchlist=$(<<EOF
+// WatchList returns all paths explicitly added with [Watcher.Add] (and are not
+// yet removed).
+//
+// Returns nil if [Watcher.Close] was called.
+EOF
+)
+
+events=$(<<EOF
+	// Events sends the filesystem change events.
+	//
+	// fsnotify can send the following events; a "path" here can refer to a
+	// file, directory, symbolic link, or special file like a FIFO.
+	//
+	//   fsnotify.Create    A new path was created; this may be followed by one
+	//                      or more Write events if data also gets written to a
+	//                      file.
+	//
+	//   fsnotify.Remove    A path was removed.
+	//
+	//   fsnotify.Rename    A path was renamed. A rename is always sent with the
+	//                      old path as Event.Name, and a Create event will be
+	//                      sent with the new name. Renames are only sent for
+	//                      paths that are currently watched; e.g. moving an
+	//                      unmonitored file into a monitored directory will
+	//                      show up as just a Create. Similarly, renaming a file
+	//                      to outside a monitored directory will show up as
+	//                      only a Rename.
+	//
+	//   fsnotify.Write     A file or named pipe was written to. A Truncate will
+	//                      also trigger a Write. A single "write action"
+	//                      initiated by the user may show up as one or multiple
+	//                      writes, depending on when the system syncs things to
+	//                      disk. For example when compiling a large Go program
+	//                      you may get hundreds of Write events, and you may
+	//                      want to wait until you've stopped receiving them
+	//                      (see the dedup example in cmd/fsnotify).
+	//
+	//                      Some systems may send Write event for directories
+	//                      when the directory content changes.
+	//
+	//   fsnotify.Chmod     Attributes were changed. On Linux this is also sent
+	//                      when a file is removed (or more accurately, when a
+	//                      link to an inode is removed). On kqueue it's sent
+	//                      when a file is truncated. On Windows it's never
+	//                      sent.
+EOF
+)
+
+errors=$(<<EOF
+	// Errors sends any errors.
+	//
+	// ErrEventOverflow is used to indicate there are too many events:
+	//
+	//  - inotify:      There are too many queued events (fs.inotify.max_queued_events sysctl)
+	//  - windows:      The buffer size is too small; WithBufferSize() can be used to increase it.
+	//  - kqueue, fen:  Not used.
+EOF
+)
+
+set-cmt() {
+	local pat=$1
+	local cmt=$2
+
+	IFS=$'\n' local files=($(grep -n $pat backend_*~*_test.go))
+	for f in $files; do
+		IFS=':' local fields=($=f)
+		local file=$fields[1]
+		local end=$(( $fields[2] - 1 ))
+
+		# Find start of comment.
+		local start=0
+		IFS=$'\n' local lines=($(head -n$end $file))
+		for (( i = 1; i <= $#lines; i++ )); do
+			local line=$lines[-$i]
+			if ! grep -q '^[[:space:]]*//' <<<$line; then
+				start=$(( end - (i - 2) ))
+				break
+			fi
+		done
+
+		head -n $(( start - 1 )) $file  >/tmp/x
+		print -r -- $cmt                >>/tmp/x
+		tail -n+$(( end + 1 ))   $file  >>/tmp/x
+		mv /tmp/x $file
+	done
+}
+
+set-cmt '^type Watcher struct '             $watcher
+set-cmt '^func NewWatcher('                 $new
+set-cmt '^func NewBufferedWatcher('         $newbuffered
+set-cmt '^func (w \*Watcher) Add('          $add
+set-cmt '^func (w \*Watcher) AddWith('      $addwith
+set-cmt '^func (w \*Watcher) Remove('       $remove
+set-cmt '^func (w \*Watcher) Close('        $close
+set-cmt '^func (w \*Watcher) WatchList('    $watchlist
+set-cmt '^[[:space:]]*Events *chan Event$'  $events
+set-cmt '^[[:space:]]*Errors *chan error$'  $errors

vendor/github.com/fsnotify/fsnotify/system_bsd.go

+//go:build freebsd || openbsd || netbsd || dragonfly
+// +build freebsd openbsd netbsd dragonfly
+
+package fsnotify
+
+import "golang.org/x/sys/unix"
+
+const openMode = unix.O_NONBLOCK | unix.O_RDONLY | unix.O_CLOEXEC

vendor/github.com/fsnotify/fsnotify/system_darwin.go

+//go:build darwin
+// +build darwin
+
+package fsnotify
+
+import "golang.org/x/sys/unix"
+
+// note: this constant is not defined on BSD
+const openMode = unix.O_EVTONLY | unix.O_CLOEXEC

vendor/github.com/go-ole/go-ole/.travis.yml

+language: go
+sudo: false
+
+go:
+  - 1.9.x
+  - 1.10.x
+  - 1.11.x
+  - tip

vendor/github.com/go-ole/go-ole/ChangeLog.md

+# Version 1.x.x
+
+* **Add more test cases and reference new test COM server project.** (Placeholder for future additions)
+
+# Version 1.2.0-alphaX
+
+**Minimum supported version is now Go 1.4. Go 1.1 support is deprecated, but should still build.**
+
+ * Added CI configuration for Travis-CI and AppVeyor.
+ * Added test InterfaceID and ClassID for the COM Test Server project.
+ * Added more inline documentation (#83).
+ * Added IEnumVARIANT implementation (#88).
+ * Added IEnumVARIANT test cases (#99, #100, #101).
+ * Added support for retrieving `time.Time` from VARIANT (#92).
+ * Added test case for IUnknown (#64).
+ * Added test case for IDispatch (#64).
+ * Added test cases for scalar variants (#64, #76).
+
+# Version 1.1.1
+
+ * Fixes for Linux build.
+ * Fixes for Windows build.
+
+# Version 1.1.0
+
+The change to provide building on all platforms is a new feature. The increase in minor version reflects that and allows those who wish to stay on 1.0.x to continue to do so. Support for 1.0.x will be limited to bug fixes.
+
+ * Move GUID out of variables.go into its own file to make new documentation available.
+ * Move OleError out of ole.go into its own file to make new documentation available.
+ * Add documentation to utility functions.
+ * Add documentation to variant receiver functions.
+ * Add documentation to ole structures.
+ * Make variant available to other systems outside of Windows.
+ * Make OLE structures available to other systems outside of Windows.
+
+## New Features
+
+ * Library should now be built on all platforms supported by Go. Library will NOOP on any platform that is not Windows.
+ * More functions are now documented and available on godoc.org.
+
+# Version 1.0.1
+
+ 1. Fix package references from repository location change.
+
+# Version 1.0.0
+
+This version is stable enough for use. The COM API is still incomplete, but provides enough functionality for accessing COM servers using IDispatch interface.
+
+There is no changelog for this version. Check commits for history.

vendor/github.com/go-ole/go-ole/LICENSE

+The MIT License (MIT)
+
+Copyright © 2013-2017 Yasuhiro Matsumoto, <mattn.jp@gmail.com>
+
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the “Software”), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

vendor/github.com/go-ole/go-ole/README.md

+# Go OLE
+
+[![Build status](https://ci.appveyor.com/api/projects/status/qr0u2sf7q43us9fj?svg=true)](https://ci.appveyor.com/project/jacobsantos/go-ole-jgs28)
+[![Build Status](https://travis-ci.org/go-ole/go-ole.svg?branch=master)](https://travis-ci.org/go-ole/go-ole)
+[![GoDoc](https://godoc.org/github.com/go-ole/go-ole?status.svg)](https://godoc.org/github.com/go-ole/go-ole)
+
+Go bindings for Windows COM using shared libraries instead of cgo.
+
+By Yasuhiro Matsumoto.
+
+## Install
+
+To experiment with go-ole, you can just compile and run the example program:
+
+```
+go get github.com/go-ole/go-ole
+cd /path/to/go-ole/
+go test
+
+cd /path/to/go-ole/example/excel
+go run excel.go
+```
+
+## Continuous Integration
+
+Continuous integration configuration has been added for both Travis-CI and AppVeyor. You will have to add these to your own account for your fork in order for it to run.
+
+**Travis-CI**
+
+Travis-CI was added to check builds on Linux to ensure that `go get` works when cross building. Currently, Travis-CI is not used to test cross-building, but this may be changed in the future. It is also not currently possible to test the library on Linux, since COM API is specific to Windows and it is not currently possible to run a COM server on Linux or even connect to a remote COM server.
+
+**AppVeyor**
+
+AppVeyor is used to build on Windows using the (in-development) test COM server. It is currently only used to test the build and ensure that the code works on Windows. It will be used to register a COM server and then run the test cases based on the test COM server.
+
+The tests currently do run and do pass and this should be maintained with commits.
+
+## Versioning
+
+Go OLE uses [semantic versioning](http://semver.org) for version numbers, which is similar to the version contract of the Go language. Which means that the major version will always maintain backwards compatibility with minor versions. Minor versions will only add new additions and changes. Fixes will always be in patch. 
+
+This contract should allow you to upgrade to new minor and patch versions without breakage or modifications to your existing code. Leave a ticket, if there is breakage, so that it could be fixed.
+
+## LICENSE
+
+Under the MIT License: http://mattn.mit-license.org/2013

vendor/github.com/go-ole/go-ole/SECURITY.md

+# Security Policy
+
+## Supported Versions
+
+Security updates are applied only to the latest release.
+
+## Reporting a Vulnerability
+
+If you have discovered a security vulnerability in this project, please report it privately. **Do not disclose it as a public issue.** This gives us time to work with you to fix the issue before public exposure, reducing the chance that the exploit will be used before a patch is released.
+
+Please disclose it at [security advisory](https://github.com/go-ole/go-ole/security/advisories/new).
+
+This project is maintained by a team of volunteers on a reasonable-effort basis. As such, please give us at least 90 days to work on a fix before public exposure.

vendor/github.com/go-ole/go-ole/appveyor.yml

+# Notes:
+#   - Minimal appveyor.yml file is an empty file. All sections are optional.
+#   - Indent each level of configuration with 2 spaces. Do not use tabs!
+#   - All section names are case-sensitive.
+#   - Section names should be unique on each level.
+
+version: "1.3.0.{build}-alpha-{branch}"
+
+os: Visual Studio 2019
+
+build: off
+
+skip_tags: true
+
+clone_folder: c:\gopath\src\github.com\go-ole\go-ole
+
+environment:
+  GOPATH: c:\gopath
+  GOROOT: c:\go
+  DOWNLOADPLATFORM: "x64"
+
+before_test:
+  # - Download COM Server
+  - ps: Start-FileDownload "https://github.com/go-ole/test-com-server/releases/download/v1.0.2/test-com-server-${env:DOWNLOADPLATFORM}.zip"
+  - 7z e test-com-server-%DOWNLOADPLATFORM%.zip -oc:\gopath\src\github.com\go-ole\go-ole > NUL
+  - c:\gopath\src\github.com\go-ole\go-ole\build\register-assembly.bat
+
+test_script:
+  - go test -v -cover ./...
+  # go vet has false positives on unsafe.Pointer with windows/sys. Disabling since it is recommended to use go test instead.
+  # - go vet ./...
+
+branches:
+  only:
+    - master
+    - v1.2
+    - v1.1
+    - v1.0
+
+matrix:
+  allow_failures:
+    - environment:
+      GOROOT: C:\go-x86
+      DOWNLOADPLATFORM: "x86"
+    - environment:
+      GOROOT: C:\go118
+      DOWNLOADPLATFORM: "x64"
+    - environment:
+      GOROOT: C:\go118-x86
+      DOWNLOADPLATFORM: "x86"
+
+install:
+  - go version
+  - go env
+  - go get -u golang.org/x/tools/cmd/cover
+  - go get -u golang.org/x/tools/cmd/godoc
+  - go get -u golang.org/x/tools/cmd/stringer
+
+build_script:
+  - cd c:\gopath\src\github.com\go-ole\go-ole
+  - go get -v -t ./...
+  - go build
+
+# disable automatic tests
+test: on
+
+# disable deployment
+deploy: off