delve
delve copied to clipboard
go-delve
Function call injection failures on Travis-CI
Travis-CI has updated its macOS builders to 10.13.6 (from ???) and now call injection tests are failing with the error:
Call - could not restore registers: protocol error E32 during register write for packet $P7b=03000000;thread...
The cause of this error is possibly inside debugserver but it could also be that we are doing something wrong. Somebody with access to a system that can replicate this issue should investigate this problem.
We also get a similar error during GoLand integration tests on linux. The test debugs the following program and once breakpoint is reached calls SliceResult():
package main
type T struct {
f string
}
type T2 struct {
T
}
func (t T) MethodNoResults() {
println("foo")
}
func (t T) MethodOneResult() string {
return "ok" //position MethodOneResult
}
func (t T) MethodManyUnnamedResults() (string, int, bool) {
return "ok", 42, false
}
func (t T) MethodManyNamedResults() (one string, two int, three bool) {
return "ok", 42, false
}
func (t T) UnusedMethod() int {
return 0
}
func UnusedFunc() int {
return 0
}
func FuncOneResult() string {
return "ok"
}
func Factorial(n int) int {
if n == 0 {
return 1 //position fact1
} else {
return n * Factorial(n - 1) //position fact2
}
}
func Panic(b bool) {
if b {
panic("panic value")
}
}
func SliceResult() []string {
return []string{"one", "two"}
}
func BigSliceResult() []int {
var result []int
for i := 0; i < 1000; i++ {
result = append(result, i)
}
return result
}
func main() {
t := T{"ok"}
t2 := T2{}
_ = FuncOneResult()
_ = t.MethodOneResult()
_, _, _ = t.MethodManyUnnamedResults()
_, _, _ = t.MethodManyNamedResults()
_ = SliceResult()
_ = BigSliceResult()
_ = t2.MethodOneResult()
t.MethodNoResults() //break
_ = Factorial(1)
Panic(false)
}
Full rpc log is here: https://gist.github.com/nd/6da1b41d356beefe3746c5aa652929ba. Can any other logs be helpful?
That error is actually very different, it's coming from a different backend. We should use a different issue for this.
- is this error systematic or sporadic?
- I've never seen it locally or on Travis-CI so it must be something influenced by kernel version or virtualization environment
- The code that's producing that error is pkg/proc/native/threads_linux_amd64.go, the first step would be to decorate restoreRegistersErr to figure out which of the three syscalls is returning EFAULT.
Looking at that function and its sister ptraceGetRegset (in ptrace_linux_amd64.go) I would guess that it's PTRACE_SETFPREGS, but it would be a systematic error, not something that happens once in a while.
Thanks will file a dedicated issue. The error seems to be sporadic and so far I've seen it only on virtual machines.