src/backend/port/darwin/README - hawq - Git at Google

 $PostgreSQL: pgsql/src/backend/port/darwin/README,v 1.5 2008/03/21 13:23:28 momjian Exp $

 Darwin
 ======

 The file system.c included herein is taken directly from Apple's Darwin
 open-source CVS archives, and is redistributed under the BSD copyright
 notice it bears.  (According to Apple's CVS logs, their version is
 identical to the FreeBSD original.)  It provides our own implementation of
 the system(3) function, which ought by all rights to be identical to the
 one provided in libc on Darwin machines.  Nonetheless, this version works,
 whereas the one that actually ships with Mac OS X 10.1 doesn't.  The
 shipped version appears to disconnect the calling process from any shared
 memory segments it is attached to.  (The symptom seen in PostgreSQL is
 that a backend attempting to execute CREATE DATABASE core-dumps.)  I would
 love to know why there is a discrepancy between the published source and
 the actual behavior --- tgl 7-Nov-2001.

 Appropriate bug reports have been filed with Apple --- see
 Radar Bug#s 2767956, 2683531, 2805147.  One hopes we can retire this
 kluge in the not too distant future.


 As of PostgreSQL 7.3 and Mac OS X 10.1, one should expect warnings
 like these while linking the backend:

 /usr/bin/ld: warning unused multiple definitions of symbol _system
 port/SUBSYS.o definition of _system in section (__TEXT,__text)
 /usr/lib/libm.dylib(system.o) unused definition of _system

 These are due to overriding system() per the above-described hack.


 The bug appears to be repaired in OS X 10.2.6 and later (possibly in
 earlier 10.2.* as well, but no systems handy to check).  We #ifdef out
 the substitute system() definition on 10.3 and later.
	$PostgreSQL: pgsql/src/backend/port/darwin/README,v 1.5 2008/03/21 13:23:28 momjian Exp $

	Darwin
	======

	The file system.c included herein is taken directly from Apple's Darwin
	open-source CVS archives, and is redistributed under the BSD copyright
	notice it bears. (According to Apple's CVS logs, their version is
	identical to the FreeBSD original.) It provides our own implementation of
	the system(3) function, which ought by all rights to be identical to the
	one provided in libc on Darwin machines. Nonetheless, this version works,
	whereas the one that actually ships with Mac OS X 10.1 doesn't. The
	shipped version appears to disconnect the calling process from any shared
	memory segments it is attached to. (The symptom seen in PostgreSQL is
	that a backend attempting to execute CREATE DATABASE core-dumps.) I would
	love to know why there is a discrepancy between the published source and
	the actual behavior --- tgl 7-Nov-2001.

	Appropriate bug reports have been filed with Apple --- see
	Radar Bug#s 2767956, 2683531, 2805147. One hopes we can retire this
	kluge in the not too distant future.


	As of PostgreSQL 7.3 and Mac OS X 10.1, one should expect warnings
	like these while linking the backend:

	/usr/bin/ld: warning unused multiple definitions of symbol _system
	port/SUBSYS.o definition of _system in section (__TEXT,__text)
	/usr/lib/libm.dylib(system.o) unused definition of _system

	These are due to overriding system() per the above-described hack.


	The bug appears to be repaired in OS X 10.2.6 and later (possibly in
	earlier 10.2.* as well, but no systems handy to check). We #ifdef out
	the substitute system() definition on 10.3 and later.