From 628db65504a69cff1d04a0639ace4a9358504fff Mon Sep 17 00:00:00 2001
From: Mahmoud Al-Qudsi <mqudsi@neosmart.net>
Date: Sat, 29 Jul 2017 16:42:06 -0500
Subject: [PATCH] OS X EINVAL compatibility for waitpid

The return value on OS X is more along the lines of the documented
waitpid behavior; EINVAL is returned if the group no longer exists.
---
 src/exec.cpp |  9 ++++++++-
 src/proc.cpp | 23 +++++++++++++++++------
 2 files changed, 25 insertions(+), 7 deletions(-)

diff --git a/src/exec.cpp b/src/exec.cpp
index 23e604f48..89d24f06a 100644
--- a/src/exec.cpp
+++ b/src/exec.cpp
@@ -1163,7 +1163,14 @@ void exec_job(parser_t &parser, job_t *j) {
             //but we only need to call set_child_group for the first process in the group.
             //If needs_keepalive is set, this has already been called for the keepalive process
             pid_t pid_status{};
-            if (waitpid(p->pid, &pid_status, WUNTRACED) == -1) {
+            int result;
+            while ((result = waitpid(p->pid, &pid_status, WUNTRACED)) == -1 && errno == EINTR) {
+                //This could be a superfluous interrupt or Ctrl+C at the terminal
+                //In all cases, it is OK to retry since the forking code above is specifically designed
+                //to never, ever hang/block in a child process before the SIGSTOP call is reached.
+                continue;
+            }
+            if (result == -1) {
                 exec_error = true;
                 debug(1, L"waitpid(%d) call in unblock_pid failed:!\n", p->pid);
                 wperror(L"waitpid");
diff --git a/src/proc.cpp b/src/proc.cpp
index 37c267850..a4b577b68 100644
--- a/src/proc.cpp
+++ b/src/proc.cpp
@@ -819,22 +819,33 @@ bool terminal_give_to_job(job_t *j, int cont) {
         //thing is that we can guarantee the process isn't going to exit while we wait (which would cause us to
         //possibly block indefinitely).
 
+        auto pgroupTerminated = [&j]() {
+            //everyone in the process group has exited
+            //The only way that can happen is if the very last process in the group terminated, and didn't need
+            //to access the terminal, otherwise it would have hung waiting for terminal IO. We can ignore this.
+            debug(3, L"terminal_give_to_job(): tcsetpgrp called but process group %d has terminated.\n", j->pgid);
+        };
+
         while (tcsetpgrp(STDIN_FILENO, j->pgid) != 0) {
             if (errno == EINTR) {
                 //always retry on EINTR
             }
+            else if (errno == EINVAL) {
+                //OS X returns EINVAL if the process group no longer lives. Probably other OSes, too.
+                //See comments in pgroupTerminated() above.
+                pgroupTerminated();
+                break;
+            }
             else if (errno == EPERM) {
-                //so long as this isn't because the process group is dead
+                //retry so long as this isn't because the process group is dead
                 int wait_result = waitpid(-1 * j->pgid, &wait_result, WNOHANG);
                 if (wait_result == -1) {
-                    //everyone in the process group has exited
-                    //The only way that can happen is if the very last process in the group terminated, and didn't need
-                    //to access the terminal, otherwise it would have hung waiting for terminal IO. We can ignore this.
                     //Note that -1 is technically an "error" for waitpid in the sense that an invalid argument was specified
-                    //because no such process group exists any longer.
+                    //because no such process group exists any longer. This is the observed behavior on Linux 4.4.0.
                     //a "success" result would mean processes from the group still exist but is still running in some state
                     //or the other.
-                    debug(3, L"terminal_give_to_job(): tcsetpgrp called but process group %d has terminated.\n", j->pgid);
+                    //See comments in pgroupTerminated() above.
+                    pgroupTerminated();
                     break;
                 }
                 debug(2, L"terminal_give_to_job(): EPERM.\n", j->pgid);