December 2004 - Darwin-dev - lists.apple.com

instantiating classes globally in a startup item (was: Re: IOServiceMatching produces oddly inert dictionary for startup item)
by Pete Gontier 08 Dec '04

08 Dec '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Microsoft-Entourage/10.1.6.040913.0 circa 12/8/04 9:47 AM, "Garth Cummings" <gcummings(a)apple.com> wrote: >> The real problem is that IOServiceGetMatchingServices is failing with a >> return value of 0x10000003 and I can't figure out why. This error code >> apparently has something to do with Mach IPC, but I don't know much more >> than that even after searching the Darwin sources for clues. > If you search for 0x10000003 on <http://developer.apple.com>, you'll > get these two hits that should clear things up: > > Retain Counts of io_object_t Objects in IOKit.framework > <http://developer.apple.com/qa/qa2001/qa1195.html> Thanks for reading through that mess to find the real problem. I was about to start another thread. :-) The relevant passage from the Q&A seems to be: > If you get the famous 0x10000003 (MACH_SEND_INVALID_DEST) return code back > from an IOKit.framework function, that's referring to a stale io_object_t, > not the underlying kernel object. This is an excellent clue which cracks the case. The only io_object_t involved here is the master port passed to IOServiceGetMatchingServices. Assuming it's stale, the question becomes how and why. I am of course obtaining the master port from IOMasterPort. This call is wrapped into the constructor of a C++ class which I instantiate globally. Such objects are supposed to be constructed by the C++ runtime library before they are used. When I launch the app from the terminal, this works; I can tell because I added a log statement to the constructor. However, when the app is launched as a startup item, the constructor does NOT run (ever). So rather than a stale io_object_t, I'm passing an uninitialized io_object_t. The question then becomes why the constructor is not running. What about being launched as a startup item inhibits the C++ runtime library from doing its job? I don't have an acute need to know this because I can just pass NULL to IOServiceGetMatchingServices since it occurs to me this app requires Panther anyway. But I would still like to know what the problem is, because if I can't instantiate classes globally in general, I'm going to have to take a census of all my code to make sure I'm not unknowingly bitten by this problem elsewhere. -- Pete Gontier http://www.m-audio.com/ _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: IOServiceMatching produces oddly inert dictionary for startup item
by Garth Cummings 08 Dec '04

08 Dec '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Hi Pete, Responding to darwin-dev alone to avoid cross-posting... On Dec 7, 2004, at 9:35 PM, Pete Gontier wrote: Retain Counts of io_object_t Objects in IOKit.framework <http://developer.apple.com/qa/qa2001/qa1195.html> Making sense of IOKit error codes <http://developer.apple.com/qa/qa2001/qa1075.html> --gc __________________________________________________________________ Garth Cummings Apple Developer Technical Support gcummings(a)apple.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com The real problem is that IOServiceGetMatchingServices is failing with a return value of 0x10000003 and I can't figure out why. This error code apparently has something to do with Mach IPC, but I don't know much more than that even after searching the Darwin sources for clues. You're not the first person who's tried to figure out what 0x10000003 is, since it is defined in <mach/message.h> which has no obvious relationship to I/O Kit. (At least one engineer within Apple has asked me the same question.) If you search for 0x10000003 on <http://developer.apple.com>, you'll get these two hits that should clear things up: I've filed <rdar://problem/3910936> asking for a way to make that error more visible within the I/O Kit headers. http://developer.apple.com/technicalsupport smime.p7s

1 0

RE: static initializers do not get run in dylibs with gcc 3.3
by Jan Hardenbergh 08 Dec '04

08 Dec '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Thread-index: AcTcpwVO0X0rsXroQPidSdIC9pb/DAAfbUVg Thread-topic: static initializers do not get run in dylibs with gcc 3.3 We have a winner!!! Geoff suggestion of using -single_module works perfectly in my case. T H A N K Y O U ! ! ! YON - Jan C. Hardenbergh, hardenbergh(a)terarecon.com, 978-369-6500x272 Director of VolumePro Software Engineering, TeraRecon, Inc. Concord Office -----Original Message----- From: Geoffrey Schmit [mailto:geoffrey.schmit@ni.com] Sent: Tuesday, December 07, 2004 4:51 PM To: darwin-dev(a)lists.apple.com Cc: Jan Hardenbergh Subject: RE: static initializers do not get run in dylibs with gcc 3.3 On 07/12/2004 at 8:59 AM, Jan Hardenbergh <hardenbergh(a)terarecon.com> wrote: > That's a good idea, but, that is not what is happening. > > This is a dylib that is registering procedure pointers to be called > externally, so, understanding who is calliing them is hard. > > But, no matter what optimization level I use (-g, -O0, -O2) the > code is there. If I call hackMyClass(), then magically, the > constructor for myClass is run. So, the code is there. > > The optimization levels do make a difference. The dylib has about > 25 classes in it. With -O2, no static initializers run. With -O0 > and -g, 9 of the static initializers run. > > So, currently, I compile -O0 and call 16 hackMyClass() routines and > it works. > > I'd like to get rid of those 16 calls. It seems like many people > have run into this problem and there seem to be many piecewise > solutions. Try the single_module linker option. It will combine all of your modules into one when they are linked. As a result, when any function in your dynamic library is invoked, all static initializers should be invoked first. geoff -- Geoffrey Schmit Senior Software Engineer National Instruments geoffrey.schmit(a)ni.com www.ni.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: IOServiceMatching produces oddly inert dictionary for startup item
by Pete Gontier 08 Dec '04

08 Dec '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Microsoft-Entourage/10.1.6.040913.0 Oops. I am certainly chasing a weird bug having to do with different behavior per launch context, but it probably has nothing to do with CFShow's silence or even the dictionary produced by IOServiceMatching. I forgot that CFShow sends its output to stderr, bypassing my own logging mechanism, which logs to multiple places, and I wasn't watching all of them in all cases. I implemented my own CFShow and now I see the dictionary is just fine. So the problem is something else entirely. Please ignore my original message and feel free to tease me mercilessly the next time you see me at WWDC. The real problem is that IOServiceGetMatchingServices is failing with a return value of 0x10000003 and I can't figure out why. This error code apparently has something to do with Mach IPC, but I don't know much more than that even after searching the Darwin sources for clues. circa 12/7/04 8:43 PM, "Pete Gontier" <pete(a)m-audio.com> wrote: > Under 10.3.6, I'm obtaining an instance of CFMutableDictionaryRef by passing > kIOUSBDeviceClassName to IOServiceMatching. When I launch my app from a > terminal, this works fine. I pass the dictionary to CFShow and get the output > I would expect. However, when my app is launched as a startup item, CFShow > outputs *nothing* (and the CFMutableDictionaryRef isn't good for anything; for > example, CFShow still outputs nothing after I call CFDictionarySetValue). So I > logged some pertinent data about the dictionary. > > launch context CFMutableDictionaryRef first 12 bytes > ============== ====================== ======================== > Terminal 6323872 006004700001039500000001 > Startup Item 4214160 004004400001039500000001 > > Does anybody have any insight into the significance of these differences or > any other insight into why IOServiceMatching might produce such an oddly inert > dictionary? -- Pete Gontier http://www.m-audio.com/ _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

IOServiceMatching produces oddly inert dictionary for startup item
by Pete Gontier 08 Dec '04

08 Dec '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Microsoft-Entourage/10.1.6.040913.0 Under 10.3.6, I'm obtaining an instance of CFMutableDictionaryRef by passing kIOUSBDeviceClassName to IOServiceMatching. When I launch my app from a terminal, this works fine. I pass the dictionary to CFShow and get the output I would expect. However, when my app is launched as a startup item, CFShow outputs *nothing* (and the CFMutableDictionaryRef isn't good for anything; for example, CFShow still outputs nothing after I call CFDictionarySetValue). So I logged some pertinent data about the dictionary. launch context CFMutableDictionaryRef first 12 bytes ============== ====================== ======================== Terminal 6323872 006004700001039500000001 Startup Item 4214160 004004400001039500000001 Does anybody have any insight into the significance of these differences or any other insight into why IOServiceMatching might produce such an oddly inert dictionary? -- Pete Gontier http://www.m-audio.com/ _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

RE: static initializers do not get run in dylibs with gcc 3.3
by Geoffrey Schmit 07 Dec '04

07 Dec '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On 07/12/2004 at 8:59 AM, Jan Hardenbergh <hardenbergh(a)terarecon.com> wrote: > That's a good idea, but, that is not what is happening. > > This is a dylib that is registering procedure pointers to be called > externally, so, understanding who is calliing them is hard. > > But, no matter what optimization level I use (-g, -O0, -O2) the > code is there. If I call hackMyClass(), then magically, the > constructor for myClass is run. So, the code is there. > > The optimization levels do make a difference. The dylib has about > 25 classes in it. With -O2, no static initializers run. With -O0 > and -g, 9 of the static initializers run. > > So, currently, I compile -O0 and call 16 hackMyClass() routines and > it works. > > I'd like to get rid of those 16 calls. It seems like many people > have run into this problem and there seem to be many piecewise > solutions. Try the single_module linker option. It will combine all of your modules into one when they are linked. As a result, when any function in your dynamic library is invoked, all static initializers should be invoked first. geoff -- Geoffrey Schmit Senior Software Engineer National Instruments geoffrey.schmit(a)ni.com www.ni.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Latest plug-ins version?
by Stéphane Letz 07 Dec '04

07 Dec '04

site_archiver(a)lists.apple.com Delivered-To: Darwin-dev(a)lists.apple.com Hi Johnny, Did you solved the AU validation issue for Jack AU? If yes can you send me the binary version of plug-ins? Thanks _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… Stephane This email sent to site_archiver(a)lists.apple.com

1 0

RE: static initializers do not get run in dylibs with gcc 3.3
by Jan Hardenbergh 07 Dec '04

07 Dec '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Thread-index: AcTb2bcdza9TmOSMTk+pDn0waPSX6wAgbVlQ Thread-topic: static initializers do not get run in dylibs with gcc 3.3 That's a good idea, but, that is not what is happening. This is a dylib that is registering procedure pointers to be called externally, so, understanding who is calliing them is hard. But, no matter what optimization level I use (-g, -O0, -O2) the code is there. If I call hackMyClass(), then magically, the constructor for myClass is run. So, the code is there. The optimization levels do make a difference. The dylib has about 25 classes in it. With -O2, no static initializers run. With -O0 and -g, 9 of the static initializers run. So, currently, I compile -O0 and call 16 hackMyClass() routines and it works. I'd like to get rid of those 16 calls. It seems like many people have run into this problem and there seem to be many piecewise solutions. YON - Jan C. Hardenbergh, hardenbergh(a)terarecon.com, 978-369-6500x272 Director of VolumePro Software Engineering, TeraRecon, Inc. Concord Office -----Original Message----- From: Jerry Pendergraft [mailto:jerry@endocardial.com] Sent: Monday, December 06, 2004 4:21 PM To: Jan Hardenbergh Cc: darwin-dev(a)lists.apple.com Subject: Re: static initializers do not get run in dylibs with gcc 3.3 Perhaps this is way too simple, but in your example there is no reference to myClass other than its declaration. Is it possibly the compiler recognized that and "optimized" the code away? -- Jerry Pendergraft jerry.pendergraft(a)endocardial.com Endocardial Solutions voice: 651-523-6935 1350 Energy Lane, Suite 110 fax: 651-644-7897 St Paul, MN 55108-5254 mobile: 651-491-0163 On Mon, 6 Dec 2004, Jan Hardenbergh wrote: > I am having a problem with static initializers. This problem is well documented if you Google: > "mac osx gcc static initializers" or variations. It works on about 1/4 of the classes. I cannot > discern which ones it will or will not work with. The static is instantiated if I call hackMyClass() > > My flavor of the Problem is > > class MyClass { > MyClass() > { > do some init > } > ... > } > > static MyClass myClass; > > hackMyClass() { return; } > > This works on Linux with gcc 3.3, MSVC++, and Solaris. > > This was rumored to have been fixed with GCC 3.3, but, it is not for me. I have tried all fo the link flags that seemed to make any sense. > > Any ideas? > > YON - Jan C. Hardenbergh > > > > _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: gettimeofday() seems to return local time
by Ragnar Sundblad 07 Dec '04

07 Dec '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com I'm using gettimeofday(), on OS X 10.3.6, to get the time resolution I need (milliseconds). If I'm reading the documentation right, I should be getting time in UTC. However, I'm getting local time. This is not what you asked, put just for the correctness of the discussion; Actually, my MacOS X man page says "Greenwich time" and not UTC, and this is in some senses more correct than the latter. I think the man pages for gettimeofday() on the OSes that mention UTC are wrong, as I doubt they will ever return UTC. UTC is a time scale with all seconds equally long. That means it has to be corrected for the variations in the movement of the Earth to keep in sync with the Earth's day and year by inserting or removing leap seconds every now and then. These are inserted just before midnight on June 30 or Dec 31. Since the introduction in 1972, UTC has gotten an extra 22 seconds inserted. In other words, the time 23h 59m 60s has existed 22 times in this time scale. I doubt any implementation of gettimeofday() ever returns time stamps that relates to this time scale (except maybe during the first half of 1972, depending on how you choose to look at it), since it would make all programs' conversion routines have to be updated with when the leap seconds happened, and those have happened every 1.5 year or so. On the other hand, the clock may be synchronized using NTP that actually uses the same second ticks as UTC (except when there are leap seconds which NTP doesn't support so it takes a little while before everything synchs together again), so in the sense gettimeofday kind of returns UTC *ticks*. But gettimeofday still does not return UTC seconds since 1970, rather GMT seconds (or UT, or UT1, or whatever scale you prefer). So - get better timing by don't confusing the time scales! :-) /ragge _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… --On Wed, 1 Dec 2004 17:57:32 -0500 "Michael Rubinstein" <mrubinstein(a)rai-dev.com> wrote: Should gettimeofday() return local time? This email sent to site_archiver(a)lists.apple.com

1 0

Re: static initializers do not get run in dylibs with gcc 3.3
by Paul Forgey 06 Dec '04

06 Dec '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Dec 6, 2004, at 8:47 AM, Jan Hardenbergh wrote: My flavor of the Problem is class MyClass { MyClass() { do some init } ... } static MyClass myClass; hackMyClass() { return; } This works on Linux with gcc 3.3, MSVC++, and Solaris. Any ideas? YON - Jan C. Hardenbergh _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/paulf%40aphrodite.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… The reason the static initializer works only when you call the no-op function is initializers are only run in a module when any symbols from that module are referenced. I think the theory (and I'm quoting my theory of the design rational, not my opinion) is there is no reason to run any static initializers in a module until anything from that module is referenced. Even without the hackMyClass() call, if you were to reference anything to do with MyClass, your static initializer will be run the first time you do. I am having a problem with static initializers. This problem is well documented if you Google: "mac osx gcc static initializers" or variations. It works on about 1/4 of the classes. I cannot discern which ones it will or will not work with. The static is instantiated if I call hackMyClass() This was rumored to have been fixed with GCC 3.3, but, it is not for me. I have tried all fo the link flags that seemed to make any sense. This email sent to paulf(a)aphrodite.com This email sent to site_archiver(a)lists.apple.com

1 0