Merge pull request #185 from reicheltp/fix-windows-build

Fix #29: Windows Build
This commit is contained in:
Dawid Dziurla 2019-11-11 12:31:00 +01:00 committed by GitHub
commit 221f2b2490
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
5 changed files with 154 additions and 24 deletions

View file

@ -33,7 +33,7 @@ build-shared: &build-shared
- run:
name: Compile project on every platform
command: |
gox -parallel 10 -os "linux freebsd" -osarch "darwin/i386 darwin/amd64"
gox -parallel 10 -os "linux freebsd" -osarch "darwin/i386 darwin/amd64 windows/amd64"
- save_cache:
key: pkg-cache-{{ checksum "go.sum" }}-v5
paths:

View file

@ -9,7 +9,7 @@ builds:
- id: binary
goos:
# - freebsd
# - windows # may reenable later
- windows
- darwin
- linux
goarch:

View file

@ -8,7 +8,6 @@ import (
"path/filepath"
"strings"
"sync"
"syscall"
"time"
"github.com/go-errors/errors"
@ -330,24 +329,3 @@ func (c *OSCommand) PipeCommands(commandStrings ...string) error {
}
return nil
}
// Kill kills a process. If the process has Setpgid == true, then we have anticipated that it might spawn its own child processes, so we've given it a process group ID (PGID) equal to its process id (PID) and given its child processes will inherit the PGID, we can kill that group, rather than killing the process itself.
func (c *OSCommand) Kill(cmd *exec.Cmd) error {
if cmd.Process == nil {
// somebody got to it before we were able to, poor bastard
return nil
}
if cmd.SysProcAttr != nil && cmd.SysProcAttr.Setpgid {
// minus sign means we're talking about a PGID as opposed to a PID
return syscall.Kill(-cmd.Process.Pid, syscall.SIGKILL)
}
return cmd.Process.Kill()
}
// PrepareForChildren sets Setpgid to true on the cmd, so that when we run it as a sideproject, we can kill its group rather than the process itself. This is because some commands, like `docker-compose logs` spawn multiple children processes, and killing the parent process isn't sufficient for killing those child processes. We set the group id here, and then in subprocess.go we check if the group id is set and if so, we kill the whole group rather than just the one process.
func (c *OSCommand) PrepareForChildren(cmd *exec.Cmd) {
cmd.SysProcAttr = &syscall.SysProcAttr{
Setpgid: true,
}
}

View file

@ -3,7 +3,9 @@
package commands
import (
"os/exec"
"runtime"
"syscall"
)
func getPlatform() *Platform {
@ -17,3 +19,24 @@ func getPlatform() *Platform {
fallbackEscapedQuote: "\"",
}
}
// Kill kills a process. If the process has Setpgid == true, then we have anticipated that it might spawn its own child processes, so we've given it a process group ID (PGID) equal to its process id (PID) and given its child processes will inherit the PGID, we can kill that group, rather than killing the process itself.
func (c *OSCommand) Kill(cmd *exec.Cmd) error {
if cmd.Process == nil {
// somebody got to it before we were able to, poor bastard
return nil
}
if cmd.SysProcAttr != nil && cmd.SysProcAttr.Setpgid {
// minus sign means we're talking about a PGID as opposed to a PID
return syscall.Kill(-cmd.Process.Pid, syscall.SIGKILL)
}
return cmd.Process.Kill()
}
// PrepareForChildren sets Setpgid to true on the cmd, so that when we run it as a sideproject, we can kill its group rather than the process itself. This is because some commands, like `docker-compose logs` spawn multiple children processes, and killing the parent process isn't sufficient for killing those child processes. We set the group id here, and then in subprocess.go we check if the group id is set and if so, we kill the whole group rather than just the one process.
func (c *OSCommand) PrepareForChildren(cmd *exec.Cmd) {
cmd.SysProcAttr = &syscall.SysProcAttr{
Setpgid: true,
}
}

View file

@ -1,5 +1,12 @@
package commands
import (
"os"
"os/exec"
"syscall"
"unsafe"
)
func getPlatform() *Platform {
return &Platform{
os: "windows",
@ -9,3 +16,125 @@ func getPlatform() *Platform {
fallbackEscapedQuote: "\\'",
}
}
// Kill kills a process. If the process has Setpgid == true, then we have anticipated that it might spawn its own child processes, so we've given it a process group ID (PGID) equal to its process id (PID) and given its child processes will inherit the PGID, we can kill that group, rather than killing the process itself.
func (c *OSCommand) Kill(cmd *exec.Cmd) error {
if cmd.Process == nil {
// somebody got to it before we were able to, poor bastard
return nil
}
pids := Getppids(uint32(cmd.Process.Pid))
for _, pid := range pids {
pro, err := os.FindProcess(int(pid))
if err != nil {
continue
}
pro.Kill()
}
return nil
}
// PrepareForChildren sets Setpgid to true on the cmd, so that when we run it as a sideproject, we can kill its group rather than the process itself. This is because some commands, like `docker-compose logs` spawn multiple children processes, and killing the parent process isn't sufficient for killing those child processes. We set the group id here, and then in subprocess.go we check if the group id is set and if so, we kill the whole group rather than just the one process.
func (c *OSCommand) PrepareForChildren(cmd *exec.Cmd) {}
const (
MAX_PATH = 260
TH32CS_SNAPPROCESS = 0x00000002
)
type ProcessInfo struct {
Name string
Pid uint32
PPid uint32
}
type PROCESSENTRY32 struct {
DwSize uint32
CntUsage uint32
Th32ProcessID uint32
Th32DefaultHeapID uintptr
Th32ModuleID uint32
CntThreads uint32
Th32ParentProcessID uint32
PcPriClassBase int32
DwFlags uint32
SzExeFile [MAX_PATH]uint16
}
type HANDLE uintptr
var (
modkernel32 = syscall.NewLazyDLL("kernel32.dll")
procCreateToolhelp32Snapshot = modkernel32.NewProc("CreateToolhelp32Snapshot")
procProcess32First = modkernel32.NewProc("Process32FirstW")
procProcess32Next = modkernel32.NewProc("Process32NextW")
procCloseHandle = modkernel32.NewProc("CloseHandle")
)
func Getppids(pid uint32) []uint32 {
infos, err := GetProcs()
if err != nil {
return []uint32{pid}
}
var pids []uint32 = make([]uint32, 0, len(infos))
var index int = 0
pids = append(pids, pid)
var length int = len(pids)
for index < length {
for _, info := range infos {
if info.PPid == pids[index] {
pids = append(pids, info.Pid)
}
}
index += 1
length = len(pids)
}
return pids
}
func GetProcs() (procs []ProcessInfo, err error) {
snap := createToolhelp32Snapshot(TH32CS_SNAPPROCESS, uint32(0))
if snap == 0 {
err = syscall.GetLastError()
return
}
defer closeHandle(snap)
var pe32 PROCESSENTRY32
pe32.DwSize = uint32(unsafe.Sizeof(pe32))
if process32First(snap, &pe32) == false {
err = syscall.GetLastError()
return
}
procs = append(procs, ProcessInfo{syscall.UTF16ToString(pe32.SzExeFile[:260]), pe32.Th32ProcessID, pe32.Th32ParentProcessID})
for process32Next(snap, &pe32) {
procs = append(procs, ProcessInfo{syscall.UTF16ToString(pe32.SzExeFile[:260]), pe32.Th32ProcessID, pe32.Th32ParentProcessID})
}
return
}
func createToolhelp32Snapshot(flags, processId uint32) HANDLE {
ret, _, _ := procCreateToolhelp32Snapshot.Call(uintptr(flags), uintptr(processId))
if ret <= 0 {
return HANDLE(0)
}
return HANDLE(ret)
}
func process32First(snapshot HANDLE, pe *PROCESSENTRY32) bool {
ret, _, _ := procProcess32First.Call(uintptr(snapshot), uintptr(unsafe.Pointer(pe)))
return ret != 0
}
func process32Next(snapshot HANDLE, pe *PROCESSENTRY32) bool {
ret, _, _ := procProcess32Next.Call(uintptr(snapshot), uintptr(unsafe.Pointer(pe)))
return ret != 0
}
func closeHandle(object HANDLE) bool {
ret, _, _ := procCloseHandle.Call(uintptr(object))
return ret != 0
}