go-pulse/vendor/github.com/rjeczalik/notify/watcher_fsevents.go
Péter Szilágyi 289b30715d Godeps, vendor: convert dependency management to trash (#3198)
This commit converts the dependency management from Godeps to the vendor
folder, also switching the tool from godep to trash. Since the upstream tool
lacks a few features proposed via a few PRs, until those PRs are merged in
(if), use github.com/karalabe/trash.

You can update dependencies via trash --update.

All dependencies have been updated to their latest version.

Parts of the build system are reworked to drop old notions of Godeps and
invocation of the go vet command so that it doesn't run against the vendor
folder, as that will just blow up during vetting.

The conversion drops OpenCL (and hence GPU mining support) from ethash and our
codebase. The short reasoning is that there's noone to maintain and having
opencl libs in our deps messes up builds as go install ./... tries to build
them, failing with unsatisfied link errors for the C OpenCL deps.

golang.org/x/net/context is not vendored in. We expect it to be fetched by the
user (i.e. using go get). To keep ci.go builds reproducible the package is
"vendored" in build/_vendor.
2016-10-28 19:05:01 +02:00

320 lines
9.0 KiB
Go

// Copyright (c) 2014-2015 The Notify Authors. All rights reserved.
// Use of this source code is governed by the MIT license that can be
// found in the LICENSE file.
// +build darwin,!kqueue
package notify
import (
"errors"
"strings"
"sync/atomic"
)
// TODO(rjeczalik): get rid of calls to canonical, it's tree responsibility
const (
failure = uint32(FSEventsMustScanSubDirs | FSEventsUserDropped | FSEventsKernelDropped)
filter = uint32(FSEventsCreated | FSEventsRemoved | FSEventsRenamed |
FSEventsModified | FSEventsInodeMetaMod)
)
// FSEvent represents single file event. It is created out of values passed by
// FSEvents to FSEventStreamCallback function.
type FSEvent struct {
Path string // real path of the file or directory
ID uint64 // ID of the event (FSEventStreamEventId)
Flags uint32 // joint FSEvents* flags (FSEventStreamEventFlags)
}
// splitflags separates event flags from single set into slice of flags.
func splitflags(set uint32) (e []uint32) {
for i := uint32(1); set != 0; i, set = i<<1, set>>1 {
if (set & 1) != 0 {
e = append(e, i)
}
}
return
}
// watch represents a filesystem watchpoint. It is a higher level abstraction
// over FSEvents' stream, which implements filtering of file events based
// on path and event set. It emulates non-recursive watch-point by filtering out
// events which paths are more than 1 level deeper than the watched path.
type watch struct {
// prev stores last event set per path in order to filter out old flags
// for new events, which appratenly FSEvents likes to retain. It's a disgusting
// hack, it should be researched how to get rid of it.
prev map[string]uint32
c chan<- EventInfo
stream *stream
path string
events uint32
isrec int32
flushed bool
}
// Example format:
//
// ~ $ (trigger command) # (event set) -> (effective event set)
//
// Heuristics:
//
// 1. Create event is removed when it was present in previous event set.
// Example:
//
// ~ $ echo > file # Create|Write -> Create|Write
// ~ $ echo > file # Create|Write|InodeMetaMod -> Write|InodeMetaMod
//
// 2. Remove event is removed if it was present in previouse event set.
// Example:
//
// ~ $ touch file # Create -> Create
// ~ $ rm file # Create|Remove -> Remove
// ~ $ touch file # Create|Remove -> Create
//
// 3. Write event is removed if not followed by InodeMetaMod on existing
// file. Example:
//
// ~ $ echo > file # Create|Write -> Create|Write
// ~ $ chmod +x file # Create|Write|ChangeOwner -> ChangeOwner
//
// 4. Write&InodeMetaMod is removed when effective event set contain Remove event.
// Example:
//
// ~ $ echo > file # Write|InodeMetaMod -> Write|InodeMetaMod
// ~ $ rm file # Remove|Write|InodeMetaMod -> Remove
//
func (w *watch) strip(base string, set uint32) uint32 {
const (
write = FSEventsModified | FSEventsInodeMetaMod
both = FSEventsCreated | FSEventsRemoved
)
switch w.prev[base] {
case FSEventsCreated:
set &^= FSEventsCreated
if set&FSEventsRemoved != 0 {
w.prev[base] = FSEventsRemoved
set &^= write
}
case FSEventsRemoved:
set &^= FSEventsRemoved
if set&FSEventsCreated != 0 {
w.prev[base] = FSEventsCreated
}
default:
switch set & both {
case FSEventsCreated:
w.prev[base] = FSEventsCreated
case FSEventsRemoved:
w.prev[base] = FSEventsRemoved
set &^= write
}
}
dbgprintf("split()=%v\n", Event(set))
return set
}
// Dispatch is a stream function which forwards given file events for the watched
// path to underlying FileInfo channel.
func (w *watch) Dispatch(ev []FSEvent) {
events := atomic.LoadUint32(&w.events)
isrec := (atomic.LoadInt32(&w.isrec) == 1)
for i := range ev {
if ev[i].Flags&FSEventsHistoryDone != 0 {
w.flushed = true
continue
}
if !w.flushed {
continue
}
dbgprintf("%v (0x%x) (%s, i=%d, ID=%d, len=%d)\n", Event(ev[i].Flags),
ev[i].Flags, ev[i].Path, i, ev[i].ID, len(ev))
if ev[i].Flags&failure != 0 {
// TODO(rjeczalik): missing error handling
continue
}
if !strings.HasPrefix(ev[i].Path, w.path) {
continue
}
n := len(w.path)
base := ""
if len(ev[i].Path) > n {
if ev[i].Path[n] != '/' {
continue
}
base = ev[i].Path[n+1:]
if !isrec && strings.IndexByte(base, '/') != -1 {
continue
}
}
// TODO(rjeczalik): get diff only from filtered events?
e := w.strip(string(base), ev[i].Flags) & events
if e == 0 {
continue
}
for _, e := range splitflags(e) {
dbgprintf("%d: single event: %v", ev[i].ID, Event(e))
w.c <- &event{
fse: ev[i],
event: Event(e),
}
}
}
}
// Stop closes underlying FSEvents stream and stops dispatching events.
func (w *watch) Stop() {
w.stream.Stop()
// TODO(rjeczalik): make (*stream).Stop flush synchronously undelivered events,
// so the following hack can be removed. It should flush all the streams
// concurrently as we care not to block too much here.
atomic.StoreUint32(&w.events, 0)
atomic.StoreInt32(&w.isrec, 0)
}
// fsevents implements Watcher and RecursiveWatcher interfaces backed by FSEvents
// framework.
type fsevents struct {
watches map[string]*watch
c chan<- EventInfo
}
func newWatcher(c chan<- EventInfo) watcher {
return &fsevents{
watches: make(map[string]*watch),
c: c,
}
}
func (fse *fsevents) watch(path string, event Event, isrec int32) (err error) {
if path, err = canonical(path); err != nil {
return err
}
if _, ok := fse.watches[path]; ok {
return errAlreadyWatched
}
w := &watch{
prev: make(map[string]uint32),
c: fse.c,
path: path,
events: uint32(event),
isrec: isrec,
}
w.stream = newStream(path, w.Dispatch)
if err = w.stream.Start(); err != nil {
return err
}
fse.watches[path] = w
return nil
}
func (fse *fsevents) unwatch(path string) (err error) {
if path, err = canonical(path); err != nil {
return
}
w, ok := fse.watches[path]
if !ok {
return errNotWatched
}
w.stream.Stop()
delete(fse.watches, path)
return nil
}
// Watch implements Watcher interface. It fails with non-nil error when setting
// the watch-point by FSEvents fails or with errAlreadyWatched error when
// the given path is already watched.
func (fse *fsevents) Watch(path string, event Event) error {
return fse.watch(path, event, 0)
}
// Unwatch implements Watcher interface. It fails with errNotWatched when
// the given path is not being watched.
func (fse *fsevents) Unwatch(path string) error {
return fse.unwatch(path)
}
// Rewatch implements Watcher interface. It fails with errNotWatched when
// the given path is not being watched or with errInvalidEventSet when oldevent
// does not match event set the watch-point currently holds.
func (fse *fsevents) Rewatch(path string, oldevent, newevent Event) error {
w, ok := fse.watches[path]
if !ok {
return errNotWatched
}
if !atomic.CompareAndSwapUint32(&w.events, uint32(oldevent), uint32(newevent)) {
return errInvalidEventSet
}
atomic.StoreInt32(&w.isrec, 0)
return nil
}
// RecursiveWatch implements RecursiveWatcher interface. It fails with non-nil
// error when setting the watch-point by FSEvents fails or with errAlreadyWatched
// error when the given path is already watched.
func (fse *fsevents) RecursiveWatch(path string, event Event) error {
return fse.watch(path, event, 1)
}
// RecursiveUnwatch implements RecursiveWatcher interface. It fails with
// errNotWatched when the given path is not being watched.
//
// TODO(rjeczalik): fail if w.isrec == 0?
func (fse *fsevents) RecursiveUnwatch(path string) error {
return fse.unwatch(path)
}
// RecrusiveRewatch implements RecursiveWatcher interface. It fails:
//
// * with errNotWatched when the given path is not being watched
// * with errInvalidEventSet when oldevent does not match the current event set
// * with errAlreadyWatched when watch-point given by the oldpath was meant to
// be relocated to newpath, but the newpath is already watched
// * a non-nil error when setting the watch-point with FSEvents fails
//
// TODO(rjeczalik): Improve handling of watch-point relocation? See two TODOs
// that follows.
func (fse *fsevents) RecursiveRewatch(oldpath, newpath string, oldevent, newevent Event) error {
switch [2]bool{oldpath == newpath, oldevent == newevent} {
case [2]bool{true, true}:
w, ok := fse.watches[oldpath]
if !ok {
return errNotWatched
}
atomic.StoreInt32(&w.isrec, 1)
return nil
case [2]bool{true, false}:
w, ok := fse.watches[oldpath]
if !ok {
return errNotWatched
}
if !atomic.CompareAndSwapUint32(&w.events, uint32(oldevent), uint32(newevent)) {
return errors.New("invalid event state diff")
}
atomic.StoreInt32(&w.isrec, 1)
return nil
default:
// TODO(rjeczalik): rewatch newpath only if exists?
// TODO(rjeczalik): migrate w.prev to new watch?
if _, ok := fse.watches[newpath]; ok {
return errAlreadyWatched
}
if err := fse.Unwatch(oldpath); err != nil {
return err
}
// TODO(rjeczalik): revert unwatch if watch fails?
return fse.watch(newpath, newevent, 1)
}
}
// Close unwatches all watch-points.
func (fse *fsevents) Close() error {
for _, w := range fse.watches {
w.Stop()
}
fse.watches = nil
return nil
}