[debugserver] Add one additional sleep before attaching after waiting

[bulbazord]

Summary

It's possible for debugserver to attach to a process during the handoff
between /usr/lib/dyld and the dyld in the shared cache. When that
happens, we may end up in a state where there is no dyld in the process
and our debugging session is doomed. To make that scenario a lot less
likely, we can insert a sleep right before attaching after waiting to
find the right pid.

rdar://105513180

[jasonmolenda]

Ah, sorry I missed this one, LGTM. Attaching to a process very early in startup during this transition between dyld's is especially difficult to recover from; letting the process run a little extra to get past this is a good workaround.

lldb/tools/debugserver/source/DNB.cpp

if (waitfor_pid == INVALID_NUB_PROCESS) {
 
       ::usleep(waitfor_interval); // Sleep for WAITFOR_INTERVAL, then poll again
     }
   }
 
   if (waitfor_pid != INVALID_NUB_PROCESS) {
     DNBLogThreadedIf(LOG_PROCESS, "Attaching to %s with pid %i...\n",
                      waitfor_process_name, waitfor_pid);
+    // In some cases, we attempt to attach during the transition from
+    // /usr/lib/dyld to the dyld in the shared cache. If that happens, we may
+    // end up in a state where there is no dyld in the process and from there
+    // the debugging session is doomed.
+    // In an attempt to make this scenario much less likely, we sleep
+    // for an additional `waitfor_interval` number of microseconds before
+    // attaching.
+    ::usleep(waitfor_interval);
     waitfor_pid = DNBProcessAttach(waitfor_pid, timeout_abstime,
                                    ctx->GetIgnoredExceptions(), err_str, 
                                    err_len);
   }
 
   bool success = waitfor_pid != INVALID_NUB_PROCESS;
   MachProcess::CleanupAfterAttach(attach_token, launch_flavor, success,
                                   prepare_error);