February 2008 - Darwin-dev - lists.apple.com

Re: Trouble updating a complex software solution
by David Elliott 12 Feb '08

12 Feb '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Hi Sorin, On Feb 11, 2008, at 11:41 AM, Sorin Negoita wrote: [...] -Dave _______________________________________________ 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… Another solution is to replace all files at startup using a script or another launchd daemon, but I must be sure that this one will be executed before the execution of my daemons and agents. I couldn't find in the launchd documentation any information to make this possible, something like Startup items dependencies or MoveFileEx function from Windows. Maybe I've missed something :( Does it really matter? Inform the user that the first reboot will boot and update some files then reboot again. This way you ensure the daemons are reloaded. Just add a reboot to the end of your update script. Or alternatively don't reboot but instead force a relaunch of all of your daemons after you update the files. The launchctl command should be able to do this. Be careful though with your argument quoting when you do updates using commands in a shell script. Another option is to replace the files the usual way (as part of the update) then restart all of the daemons and agents from your post install script. This way you don't need to have the user reboot at all which is much much nicer, particularly if this is a server-related product. You may also even shut down the daemons and agents pre- install and start them up again post-install. I'm not sure that a MoveFileEx equivalent is really desirable. You know better than Apple ever could what specific needs your daemons have. This email sent to site_archiver(a)lists.apple.com

1 0

Re: getcontext() -> Bus error?
by Edd Dawson 12 Feb '08

12 Feb '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Thunderbird 2.0.0.9 (Windows/20071031) Sorry? You lost me! Which stack variable? Kind regards, Edd On Feb 11, 2008, at 11:10 AM, Edd Dawson <lists(a)mr-edd.co.uk> wrote: _______________________________________________ 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… Terry Lambert wrote: Given your use of a stack variable, it makes all sorts if sense. 8-). This isn't an entirely pointless follow-up; if you can shed some light on why the code failed on Darwin but not on Linux, I'm interested to know more. All I thought I needed to do before calling makecontext() was ensure that I had a stack allocated and uc_link set appropriately? Edd Dawson wrote: Kevin Van Vechten wrote: On Feb 3, 2008, at 4:43 PM, Edd Dawson wrote: I hope this is my final question on the matter. The following code is failing on Leopard. The assert() in main() is triggered. The vector's size appears to be 3?! #define _XOPEN_SOURCE #include <ucontext.h> #include <signal.h> #include <cassert> #include <cstdio> #include <cstdlib> #include <vector> struct context { context(ucontext_t ctx, void (*func)()) : ctx_(ctx), func_(func), stack_(MINSIGSTKSZ) { ctx_.uc_stack.ss_sp = &stack_.front(); ctx_.uc_stack.ss_size = stack_.size(); ctx_.uc_link = 0; makecontext(&ctx_, func_, 0); } context(ucontext_t ctx) : ctx_(ctx), func_(0), stack_() { } static void change(context &from, context &to) { swapcontext(&from.ctx_, &to.ctx_); } ucontext_t ctx_; void (*func_)(); std::vector<char> stack_; }; std::vector<context *> all; void die(const char *msg) { perror(msg); exit(666); } void hello() { puts("hello"); assert(all.size() == 2); context::change(*all[1], *all[0]); } int main() { ucontext_t ctx; if (getcontext(&ctx) == -1) die("getcontext failed"); all.push_back(new context(ctx)); if (getcontext(&ctx) == -1) die("getcontext failed"); all.push_back(new context(ctx, &hello)); assert(all.size() == 2); context::change(*all[0], *all[1]); return 0; } This all works fine when I move getcontext() in to the same function as the associated makecontext(), which makes sense from a certain point of view. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Trouble updating a complex software solution
by Dave Zarzycki 12 Feb '08

12 Feb '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com davez On Feb 11, 2008, at 8:41 AM, Sorin Negoita wrote: Hello, Any help would be appreciated. Thank you in advance ! _______________________________________________ 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/zarzycki%40apple.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… We feel your pain. We have similar problems every time we do a software update for the operating system. Unfortunately, our solutions are not ready for third-parties to adopt. Can you please file a bug report? Thanks! I have a software solution formed of 3 kexts, 3 launchd daemons, 3 agents and 1 IPC mechanism used for communication between daemons and agents. I can't find a reliable way of updating this solution. My problem appears if the new executables (or IPC files) are not compatibles anymore with the old ones. So, if I replace all files right after the download of the updates and the user restart the system, everything is OK. But if the user does not restart the system and one of my daemons or agents restarts or crashes, then the new one restarted will not be able to "talk" with the old ones (because of the incompatibilities). Another solution is to replace all files at startup using a script or another launchd daemon, but I must be sure that this one will be executed before the execution of my daemons and agents. I couldn't find in the launchd documentation any information to make this possible, something like Startup items dependencies or MoveFileEx function from Windows. Maybe I've missed something :( This email sent to zarzycki(a)apple.com This email sent to site_archiver(a)lists.apple.com

1 0

Mac OS X 10.5.2 Kernel Panic
by Rand Childs 11 Feb '08

11 Feb '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.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

1 0

Re: getcontext() -> Bus error?
by Terry Lambert 11 Feb '08

11 Feb '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Given your use of a stack variable, it makes all sorts if sense. 8-). -- Terry On Feb 11, 2008, at 11:10 AM, Edd Dawson <lists(a)mr-edd.co.uk> wrote: _______________________________________________ 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… Edd Dawson wrote: Kevin Van Vechten wrote: On Feb 3, 2008, at 4:43 PM, Edd Dawson wrote: I hope this is my final question on the matter. The following code is failing on Leopard. The assert() in main() is triggered. The vector's size appears to be 3?! #define _XOPEN_SOURCE #include <ucontext.h> #include <signal.h> #include <cassert> #include <cstdio> #include <cstdlib> #include <vector> struct context { context(ucontext_t ctx, void (*func)()) : ctx_(ctx), func_(func), stack_(MINSIGSTKSZ) { ctx_.uc_stack.ss_sp = &stack_.front(); ctx_.uc_stack.ss_size = stack_.size(); ctx_.uc_link = 0; makecontext(&ctx_, func_, 0); } context(ucontext_t ctx) : ctx_(ctx), func_(0), stack_() { } static void change(context &from, context &to) { swapcontext(&from.ctx_, &to.ctx_); } ucontext_t ctx_; void (*func_)(); std::vector<char> stack_; }; std::vector<context *> all; void die(const char *msg) { perror(msg); exit(666); } void hello() { puts("hello"); assert(all.size() == 2); context::change(*all[1], *all[0]); } int main() { ucontext_t ctx; if (getcontext(&ctx) == -1) die("getcontext failed"); all.push_back(new context(ctx)); if (getcontext(&ctx) == -1) die("getcontext failed"); all.push_back(new context(ctx, &hello)); assert(all.size() == 2); context::change(*all[0], *all[1]); return 0; } This all works fine when I move getcontext() in to the same function as the associated makecontext(), which makes sense from a certain point of view. This email sent to site_archiver(a)lists.apple.com

1 0

Re: getcontext() -> Bus error?
by Edd Dawson 11 Feb '08

11 Feb '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Thunderbird 2.0.0.9 (Windows/20071031) #define _XOPEN_SOURCE #include <ucontext.h> #include <signal.h> #include <cassert> #include <cstdio> #include <cstdlib> #include <vector> struct context { context(ucontext_t ctx, void (*func)()) : ctx_(ctx), func_(func), stack_(MINSIGSTKSZ) { ctx_.uc_stack.ss_sp = &stack_.front(); ctx_.uc_stack.ss_size = stack_.size(); ctx_.uc_link = 0; makecontext(&ctx_, func_, 0); } context(ucontext_t ctx) : ctx_(ctx), func_(0), stack_() { } static void change(context &from, context &to) { swapcontext(&from.ctx_, &to.ctx_); } ucontext_t ctx_; void (*func_)(); std::vector<char> stack_; }; std::vector<context *> all; void die(const char *msg) { perror(msg); exit(666); } void hello() { puts("hello"); assert(all.size() == 2); context::change(*all[1], *all[0]); } int main() { ucontext_t ctx; if (getcontext(&ctx) == -1) die("getcontext failed"); all.push_back(new context(ctx)); if (getcontext(&ctx) == -1) die("getcontext failed"); all.push_back(new context(ctx, &hello)); assert(all.size() == 2); context::change(*all[0], *all[1]); return 0; } I can work with this, so problem solved as far as I'm concerned. Kind regards, Edd _______________________________________________ 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… Edd Dawson wrote: Kevin Van Vechten wrote: On Feb 3, 2008, at 4:43 PM, Edd Dawson wrote: I hope this is my final question on the matter. The following code is failing on Leopard. The assert() in main() is triggered. The vector's size appears to be 3?! This all works fine when I move getcontext() in to the same function as the associated makecontext(), which makes sense from a certain point of view. This email sent to site_archiver(a)lists.apple.com

1 0

Re: FSEventStream with more than one Path
by Joachim Deelen 11 Feb '08

11 Feb '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.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

1 0

FSEventStream with more than one Path
by Joachim Deelen 11 Feb '08

11 Feb '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.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

1 0

Trouble updating a complex software solution
by Sorin Negoita 11 Feb '08

11 Feb '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:received:received:message-id:date:from:to:subject:mime-version:content-type; bh=SK7q7Fk0t7r3ZGBy14tht+SyWPYTIc5B+DNM6wQSqy8=; b=xZtT0kn2XsE0LaN+Wx4kN+2q9QYaOBjBkyN4XWGDATwV4CZ/bxEXUAHceTe6Upl3CishQucqKt1eizmNuy211CO/Rc6MU975ayGXJxQ4vXo9IDdP7l2Bhw20swEqAuBctXqdIIegjhOzEL9KtIQKBc6Ap/wIZwtR8GzI2r2B7Wo= Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=message-id:date:from:to:subject:mime-version:content-type; b=gD/Re7G/33hB4uxglnu+d6uryr0C18YwK1+Eu2GQEKwAEzmOI4RLuz03CGujFGg/KsjUkECpRrLWgS/bcQVQX3vS3zEHl7Yf9Hd/mppQ5pBmxD/DgRU+GrWrupansfoGkNU7DWtkf2RFDwIXEW/I3CdsVV5uV582MFzgIY8MgBA= _______________________________________________ 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: getcontext() -> Bus error?
by Edd Dawson 11 Feb '08

11 Feb '08

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Thunderbird 2.0.0.9 (Windows/20071031) On Feb 3, 2008, at 4:43 PM, Edd Dawson wrote: Thanks again, Kevin. That worked great! #define _XOPEN_SOURCE #include <ucontext.h> #include <signal.h> #include <cassert> #include <cstdio> #include <cstdlib> #include <vector> struct context { context(ucontext_t ctx, void (*func)()) : ctx_(ctx), func_(func), stack_(MINSIGSTKSZ) { ctx_.uc_stack.ss_sp = &stack_.front(); ctx_.uc_stack.ss_size = stack_.size(); ctx_.uc_link = 0; makecontext(&ctx_, func_, 0); } context(ucontext_t ctx) : ctx_(ctx), func_(0), stack_() { } static void change(context &from, context &to) { swapcontext(&from.ctx_, &to.ctx_); } ucontext_t ctx_; void (*func_)(); std::vector<char> stack_; }; std::vector<context *> all; void die(const char *msg) { perror(msg); exit(666); } void hello() { puts("hello"); assert(all.size() == 2); context::change(*all[1], *all[0]); } int main() { ucontext_t ctx; if (getcontext(&ctx) == -1) die("getcontext failed"); all.push_back(new context(ctx)); if (getcontext(&ctx) == -1) die("getcontext failed"); all.push_back(new context(ctx, &hello)); assert(all.size() == 2); context::change(*all[0], *all[1]); return 0; } I'm compiling with: g++ getcontext3.cpp -o getcontext3 -W -Wall -ansi -pedantic Thanks again for all your help so far! Kind regards, Edd _______________________________________________ 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… Kevin Van Vechten wrote: That seems to have stopped the bus error -- thanks! -- but now I'm getting what I believe is fishy behaviour from swapcontext(). The following program prints "about to swapcontext()..." indefinitely. The 4096-byte stack storage provided is too small for Darwin. The stack size should be at least MINSIGSTKSZ (include signal.h to get this constant). I hope this is my final question on the matter. The following code is failing on Leopard. The assert() in main() is triggered. The vector's size appears to be 3?! This email sent to site_archiver(a)lists.apple.com

1 0