test/sandbox: treat ESRCH as temporary failure

This is an ugly fix that makes various assumptions guaranteed to hold true in the testing vm. The test package is filtered by the build system so some ugliness is tolerable here.

Signed-off-by: Ophestra <cat@gensokyo.uk>

test/sandbox/assert.go

@@ -15,6 +15,7 @@ import (
 	"log"
 	"os"
 	"syscall"
+	"time"
 )
 
 var (
@@ -163,15 +164,36 @@ func CheckFilter(pid int, want string) error {
 		}
 	}()
 
+	h := sha512.New()
+	{
+	getFilter:
 		buf, err := getFilter[[8]byte](pid, 0)
+		/* this is not how ESRCH should be handled: the manpage advises the
+		use of waitpid, however that is not applicable for attaching to an
+		arbitrary process, and spawning target process here is not easily
+		possible under the current testing framework;
+
+		despite checking for /proc/pid/status indicating state t (tracing stop),
+		it does not appear to be directly related to the internal state used to
+		determine whether a process is ready to accept ptrace operations, it also
+		introduces a TOCTOU that is irrelevant in the testing vm; this behaviour
+		is kept anyway as it reduces the average iterations required here;
+
+		since this code is only ever compiled into the test program, whatever
+		implications this ugliness might have should not hurt anyone */
+		if errors.Is(err, syscall.ESRCH) {
+			time.Sleep(100 * time.Millisecond)
+			goto getFilter
+		}
+
 		if err != nil {
 			return err
 		}
 
-	h := sha512.New()
 		for _, b := range buf {
 			h.Write(b[:])
 		}
+	}
 
 	if got := hex.EncodeToString(h.Sum(nil)); got != want {
 		printf("[FAIL] %s", got)