[Zope-Checkins] SVN: Zope/branches/2.9/ Collector #1904: On Mac OS X avoid a spurious OSError when zopectl

[Florent Guillaume]

Log message for revision 40547:
  Collector #1904: On Mac OS X avoid a spurious OSError when zopectl
  exits.
  

Changed:
  U   Zope/branches/2.9/doc/CHANGES.txt
  U   Zope/branches/2.9/lib/python/Zope2/Startup/zopectl.py

-=-
Modified: Zope/branches/2.9/doc/CHANGES.txt
===================================================================
--- Zope/branches/2.9/doc/CHANGES.txt	2005-12-05 16:34:38 UTC (rev 40546)
+++ Zope/branches/2.9/doc/CHANGES.txt	2005-12-05 16:50:14 UTC (rev 40547)
@@ -104,3 +104,6 @@
 
       - Switched to the new improved test runner from Zope 3.  Run
         test.py with -h to find out more.
+
+      - Collector #1904: On Mac OS X avoid a spurious OSError when
+        zopectl exits.

Modified: Zope/branches/2.9/lib/python/Zope2/Startup/zopectl.py
===================================================================
--- Zope/branches/2.9/lib/python/Zope2/Startup/zopectl.py	2005-12-05 16:34:38 UTC (rev 40546)
+++ Zope/branches/2.9/lib/python/Zope2/Startup/zopectl.py	2005-12-05 16:50:14 UTC (rev 40547)
@@ -304,5 +304,13 @@
     #   If it is not reset, 'os.wait[pid]' can non-deterministically fail.
     #   Thus, use a way such that "SIGCHLD" is definitely reset in children.
     #signal.signal(signal.SIGCHLD, signal.SIG_IGN)
-    signal.signal(signal.SIGCHLD, _ignoreSIGCHLD)
+    if os.uname()[0] != 'Darwin':
+        # On Mac OS X, setting up a signal handler causes waitpid to
+        # raise EINTR, which is not preventable via the Python signal
+        # handler API and can't be dealt with properly as we can't pass
+        # the SA_RESTART to the signal API. Since Mac OS X doesn't
+        # appear to clutter up the process table with zombies if
+        # SIGCHILD is unset, just don't bother registering a SIGCHILD
+        # signal handler at all.
+        signal.signal(signal.SIGCHLD, _ignoreSIGCHLD)
     main()