From: Nicholas Clark <nick@ccl4.org>
Date: Tue, 25 Oct 2005 12:54:18 +0000 (+0000)
Subject: Code in wait4pid was calling hv_delete with the hash iterator
X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=commitdiff_plain;h=7b9a32411c4ec6251e6f3ab9bb69dd43222d7692;p=p5sagit%2Fp5-mst-13.2.git

Code in wait4pid was calling hv_delete with the hash iterator
currently on that entry. On aggregate this does more work, beacuse
the next call to hv_iterinit() would spot the flag, and have to call
the delete routine, while in the meantime any new entries can't
re-use that memory.

p4raw-id: //depot/perl@25848
---

diff --git a/util.c b/util.c
index 96da1fc..74facce 100644
--- a/util.c
+++ b/util.c
@@ -2661,6 +2661,13 @@ Perl_wait4pid(pTHX_ Pid_t pid, int *statusp, int flags)
 		pid = atoi(hv_iterkey(entry,(I32*)statusp));
 		*statusp = SvIVX(sv);
 		len = my_sprintf(spid, "%"IVdf, (IV)pid);
+		/* The hash iterator is currently on this entry, so simply
+		   calling hv_delete would trigger the lazy delete, which on
+		   aggregate does more work, beacuse next call to hv_iterinit()
+		   would spot the flag, and have to call the delete routine,
+		   while in the meantime any new entries can't re-use that
+		   memory.  */
+		hv_iterinit(PL_pidstatus);
 		(void)hv_delete(PL_pidstatus,spid,len,G_DISCARD);
 		return pid;
 	    }