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

Re: SDKs and linking
by Paul Forgey 23 Nov '04

23 Nov '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Nov 23, 2004, at 2:30 PM, Sam Vaughan wrote: Sam On 24/11/2004, at 9:24 AM, Shawn Erickson wrote: On Nov 23, 2004, at 1:51 PM, Sam Vaughan wrote: Forwarding from Nick: libauto.dylib doesn't exist on 10.3 but does exist in 10.4's SDK _______________________________________________ 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… DYLD_LIBRARY_PATH may help you get there, but it won't completely solve the problem. Let's say you have app which dynamically links to lib1 which links to lib2. lib1 and lib2 are in /that/directory although they have dylib names in /usr/lib. Setting DYLD_LIBRARY_PATH to include /that/directory will let the app link to lib1 without -L or -F, but it still won't find lib2. You can include both libraries in -l. While that's a pain to do because you have to anticipate the entire dependency graph, it might be easier than what you are doing now. The same problem exists in 10.3 with the Developer SDKs. Substitute all mention of 10.4 with 10.3 and 10.3 with 10.2 if you like. Hi, I am trying to build some things on 10.3 using the 10.4 SDK, and the behaviour of ld is making life difficult. It seems that implicitly linked libraries do not honour -L paths (even if -Z is passed), hence, a program which links in libSystem.dylib will fail to link. eg: $ gcc -Z -L/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib -F/Developer/SDKs/MacOSX10.4.0.sdk//System/Library/Frameworks -o bin-10.4.0/XVMprobe obj-10.4.0/mac_probe.o -framework CoreFoundation -framework IOKit ld: warning can't open dynamic library: /usr/lib/libauto.dylib (checking for undefined symbols may be affected) (No such file or directory, errno = 2) ld: Undefined symbols: _auto_register_weak_reference referenced from CoreFoundation expected to be defined in /usr/lib/libauto.dylib _auto_unregister_weak_reference referenced from CoreFoundation expected to be defined in /usr/lib/libauto.dylib <...snip...> _utrans_openU referenced from CoreFoundation expected to be defined in libicucore What is the proper solution to this? Explicitly linking the libraries will work, but that's a large number of potentially changing libraries. Alternatively (and the solution I've taken so far), I can do something like this: $ gcc -Z -L/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib -dylib_file /usr/lib/libauto.dylib:/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib/ libauto.dylib -dylib_file /usr/lib/libobjc.A.dylib:/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib/ libobjc.A.dylib -dylib_file /usr/lib/libicucore.A.dylib:/Developer/SDKs/MacOSX10.4.0.sdk//usr/ lib/libicucore.A.dylib -F/Developer/SDKs/MacOSX10.4.0.sdk//System/Library/Frameworks -o bin-10.4.0/XVMprobe obj-10.4.0/mac_probe.o -framework CoreFoundation -framework IOKit which does work, but is very ugly... and I'm concerned over the amount of work needed to maintain the dylib_file list. The base of this problem stems from full paths being stored in the libraries. It would be better to have an independent search path, ala LD_LIBRARY_PATH for runtime and -L for link time. Mac OS X 10.4 (Tiger) is under NDA and is not to be discussed on any public lists. You may want to contact Apple's DTS folks about how best to go about this at this time. -Shawn This email sent to paulf(a)aphrodite.com This email sent to site_archiver(a)lists.apple.com

1 0

Re: SDKs and linking
by Sam Vaughan 23 Nov '04

23 Nov '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Sam On 24/11/2004, at 9:24 AM, Shawn Erickson wrote: On Nov 23, 2004, at 1:51 PM, Sam Vaughan wrote: Forwarding from Nick: libauto.dylib doesn't exist on 10.3 but does exist in 10.4's SDK _______________________________________________ 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 same problem exists in 10.3 with the Developer SDKs. Substitute all mention of 10.4 with 10.3 and 10.3 with 10.2 if you like. Hi, I am trying to build some things on 10.3 using the 10.4 SDK, and the behaviour of ld is making life difficult. It seems that implicitly linked libraries do not honour -L paths (even if -Z is passed), hence, a program which links in libSystem.dylib will fail to link. eg: $ gcc -Z -L/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib -F/Developer/SDKs/MacOSX10.4.0.sdk//System/Library/Frameworks -o bin-10.4.0/XVMprobe obj-10.4.0/mac_probe.o -framework CoreFoundation -framework IOKit ld: warning can't open dynamic library: /usr/lib/libauto.dylib (checking for undefined symbols may be affected) (No such file or directory, errno = 2) ld: Undefined symbols: _auto_register_weak_reference referenced from CoreFoundation expected to be defined in /usr/lib/libauto.dylib _auto_unregister_weak_reference referenced from CoreFoundation expected to be defined in /usr/lib/libauto.dylib <...snip...> _utrans_openU referenced from CoreFoundation expected to be defined in libicucore What is the proper solution to this? Explicitly linking the libraries will work, but that's a large number of potentially changing libraries. Alternatively (and the solution I've taken so far), I can do something like this: $ gcc -Z -L/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib -dylib_file /usr/lib/libauto.dylib:/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib/ libauto.dylib -dylib_file /usr/lib/libobjc.A.dylib:/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib/ libobjc.A.dylib -dylib_file /usr/lib/libicucore.A.dylib:/Developer/SDKs/MacOSX10.4.0.sdk//usr/ lib/libicucore.A.dylib -F/Developer/SDKs/MacOSX10.4.0.sdk//System/Library/Frameworks -o bin-10.4.0/XVMprobe obj-10.4.0/mac_probe.o -framework CoreFoundation -framework IOKit which does work, but is very ugly... and I'm concerned over the amount of work needed to maintain the dylib_file list. The base of this problem stems from full paths being stored in the libraries. It would be better to have an independent search path, ala LD_LIBRARY_PATH for runtime and -L for link time. Mac OS X 10.4 (Tiger) is under NDA and is not to be discussed on any public lists. You may want to contact Apple's DTS folks about how best to go about this at this time. -Shawn This email sent to site_archiver(a)lists.apple.com

1 0

Re: SDKs and linking
by Shawn Erickson 23 Nov '04

23 Nov '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Nov 23, 2004, at 1:51 PM, Sam Vaughan wrote: Forwarding from Nick: libauto.dylib doesn't exist on 10.3 but does exist in 10.4's SDK -Shawn _______________________________________________ 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… Hi, I am trying to build some things on 10.3 using the 10.4 SDK, and the behaviour of ld is making life difficult. It seems that implicitly linked libraries do not honour -L paths (even if -Z is passed), hence, a program which links in libSystem.dylib will fail to link. eg: $ gcc -Z -L/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib -F/Developer/SDKs/MacOSX10.4.0.sdk//System/Library/Frameworks -o bin-10.4.0/XVMprobe obj-10.4.0/mac_probe.o -framework CoreFoundation -framework IOKit ld: warning can't open dynamic library: /usr/lib/libauto.dylib (checking for undefined symbols may be affected) (No such file or directory, errno = 2) ld: Undefined symbols: _auto_register_weak_reference referenced from CoreFoundation expected to be defined in /usr/lib/libauto.dylib _auto_unregister_weak_reference referenced from CoreFoundation expected to be defined in /usr/lib/libauto.dylib <...snip...> _utrans_openU referenced from CoreFoundation expected to be defined in libicucore What is the proper solution to this? Explicitly linking the libraries will work, but that's a large number of potentially changing libraries. Alternatively (and the solution I've taken so far), I can do something like this: $ gcc -Z -L/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib -dylib_file /usr/lib/libauto.dylib:/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib/ libauto.dylib -dylib_file /usr/lib/libobjc.A.dylib:/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib/ libobjc.A.dylib -dylib_file /usr/lib/libicucore.A.dylib:/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib/ libicucore.A.dylib -F/Developer/SDKs/MacOSX10.4.0.sdk//System/Library/Frameworks -o bin-10.4.0/XVMprobe obj-10.4.0/mac_probe.o -framework CoreFoundation -framework IOKit which does work, but is very ugly... and I'm concerned over the amount of work needed to maintain the dylib_file list. The base of this problem stems from full paths being stored in the libraries. It would be better to have an independent search path, ala LD_LIBRARY_PATH for runtime and -L for link time. Mac OS X 10.4 (Tiger) is under NDA and is not to be discussed on any public lists. You may want to contact Apple's DTS folks about how best to go about this at this time. This email sent to site_archiver(a)lists.apple.com

1 0

SDKs and linking
by Sam Vaughan 23 Nov '04

23 Nov '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Forwarding from Nick: libauto.dylib doesn't exist on 10.3 but does exist in 10.4's SDK Nick _______________________________________________ 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… Hi, I am trying to build some things on 10.3 using the 10.4 SDK, and the behaviour of ld is making life difficult. It seems that implicitly linked libraries do not honour -L paths (even if -Z is passed), hence, a program which links in libSystem.dylib will fail to link. eg: $ gcc -Z -L/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib -F/Developer/SDKs/MacOSX10.4.0.sdk//System/Library/Frameworks -o bin-10.4.0/XVMprobe obj-10.4.0/mac_probe.o -framework CoreFoundation -framework IOKit ld: warning can't open dynamic library: /usr/lib/libauto.dylib (checking for undefined symbols may be affected) (No such file or directory, errno = 2) ld: Undefined symbols: _auto_register_weak_reference referenced from CoreFoundation expected to be defined in /usr/lib/libauto.dylib _auto_unregister_weak_reference referenced from CoreFoundation expected to be defined in /usr/lib/libauto.dylib <...snip...> _utrans_openU referenced from CoreFoundation expected to be defined in libicucore What is the proper solution to this? Explicitly linking the libraries will work, but that's a large number of potentially changing libraries. Alternatively (and the solution I've taken so far), I can do something like this: $ gcc -Z -L/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib -dylib_file /usr/lib/libauto.dylib:/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib/ libauto.dylib -dylib_file /usr/lib/libobjc.A.dylib:/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib/ libobjc.A.dylib -dylib_file /usr/lib/libicucore.A.dylib:/Developer/SDKs/MacOSX10.4.0.sdk//usr/lib/ libicucore.A.dylib -F/Developer/SDKs/MacOSX10.4.0.sdk//System/Library/Frameworks -o bin-10.4.0/XVMprobe obj-10.4.0/mac_probe.o -framework CoreFoundation -framework IOKit which does work, but is very ugly... and I'm concerned over the amount of work needed to maintain the dylib_file list. The base of this problem stems from full paths being stored in the libraries. It would be better to have an independent search path, ala LD_LIBRARY_PATH for runtime and -L for link time. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Maximum number of threads
by Andreas Yankopolus 23 Nov '04

23 Nov '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Kenny, Thanks, Andreas _______________________________________________ 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… Surely, uless you have a 15,000 Processor system, that at some point your system will become processor bound? It will spend more time context switching than it actually does running the processes. I suspect that this would be the case if each thread were active a large portion of the time, but the way simulations pan out, the threads are largely dormant and primarily used to maintain state between the transmission and reception messages. Can't one thread handle multiple queues? I'm looking at ways to aggregate functionality, but it would be easier to just make more threads :-) This email sent to site_archiver(a)lists.apple.com

1 0

RE: Maximum number of threads
by Kenny Millar 23 Nov '04

23 Nov '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Thread-index: AcTRcYGothPuH/JxR1yo8/H6Bnfc3QABKBaw Surely, uless you have a 15,000 Processor system, that at some point your system will become processor bound? It will spend more time context switching than it actually does running the processes. Can't one thread handle multiple queues? -Kenny > -----Original Message----- > From: darwin-dev-bounces+kennymillar=mac.com(a)lists.apple.com > [mailto:darwin-dev-bounces+kennymillar=mac.com@lists.apple.com > ] On Behalf Of Andreas Yankopolus > Sent: 23 November 2004 15:31 > To: Shawn Erickson > Cc: darwin-dev(a)lists.apple.com > Subject: Re: Maximum number of threads > > Shawn, > > > Wow... why so many threads? > > > > Why not a thread pool of worker threads that you can take > tasks off of > > a queue or something? > > Somehow I knew this question would come up sooner or later :-) > > I'm working on a communications simulation where the > straightforward approach involves creating one thread per > node. This thread both generates new messages based on > traffic models and listens for messages from other nodes. In > some instances, more than one thread per node would prove > handy. The network has 15k+ nodes, and therein lies the problem. > > Cheers, > > Andreas > > _______________________________________________ > 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/kennymillar% > 40mac.com > > This email sent to kennymillar(a)mac.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: Maximum number of threads
by Andreas Yankopolus 23 Nov '04

23 Nov '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Shawn, Wow... why so many threads? Somehow I knew this question would come up sooner or later :-) Cheers, Andreas _______________________________________________ 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… Why not a thread pool of worker threads that you can take tasks off of a queue or something? I'm working on a communications simulation where the straightforward approach involves creating one thread per node. This thread both generates new messages based on traffic models and listens for messages from other nodes. In some instances, more than one thread per node would prove handy. The network has 15k+ nodes, and therein lies the problem. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Maximum number of threads
by Shawn Erickson 23 Nov '04

23 Nov '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Nov 23, 2004, at 4:10 AM, Andreas Yankopolus wrote: (don't know the answer to your question but...) Wow... why so many threads? -Shawn _______________________________________________ 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… I'm bumping into what appears to be Darwin's pthread limit: my process stops creating threads after spinning off about 2558 of them (top gives a system total of 2711 threads). I've reduced the stack size of each thread, so I'm not running out of memory. On Linux, I'd recompile the threading library with a larger number for PTHREAD_THREADS_MAX. How do I go about raising the limit on OS X? Why not a thread pool of worker threads that you can take tasks off of a queue or something? This email sent to site_archiver(a)lists.apple.com

1 0

Maximum number of threads
by Andreas Yankopolus 23 Nov '04

23 Nov '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Thanks, Andreas _______________________________________________ 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… I'm bumping into what appears to be Darwin's pthread limit: my process stops creating threads after spinning off about 2558 of them (top gives a system total of 2711 threads). I've reduced the stack size of each thread, so I'm not running out of memory. On Linux, I'd recompile the threading library with a larger number for PTHREAD_THREADS_MAX. How do I go about raising the limit on OS X? This email sent to site_archiver(a)lists.apple.com

1 0

Re: BOM format and Open Source
by Peter Bierman 22 Nov '04

22 Nov '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Le 21 nov. 04, à 22:17, Stéphane Sudre a écrit : I also think you'll be impressed with some upcoming improvements. :-) -pmb _______________________________________________ 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… At 10:34 PM +0100 11/21/04, Damien Bobillot wrote: Any sign of progress on the open sourcing of the BOM format since Kevin Van Vechten's request a long time ago? I wonder why Apple don't provide at least a public API for the Bom framework. It may be very useful to develop custom installer makers. Apple's one (and the mkbom command) is hard to use in automatic tasks and not very flexible. Are there specific problems you experience when using the available tools? Have you filed bug reports about those issues? I'm one of the engineers working on the Installer, and I can tell you that bug reports carry lots of weight when setting priorities and direction for these tools. This email sent to site_archiver(a)lists.apple.com

1 0