March 2005 - Darwin-dev - lists.apple.com

Re: Apple gcc bug
by Paul Forgey 17 Mar '05

17 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com #define DIVIDE(a,b) (assert((b)!=0),(a)/(b)) And use DIVIDE(n,m) instead of n/m. On Mar 17, 2005, at 2:25 PM, Aaron Jackson wrote: On Mar 17, 2005, at 4:55 PM, Matt Watson wrote: fpsetmask(FP_X_INV | FP_X_DZ); signal(SIGFPE, sig_handler); _______________________________________________ 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… If the PPC processor does not throw an exception in this situation, then there's no exception to unmask. Can't do it that way. It is a bug in your code, much like dereferencing a null pointer. All the macros in the header files won't change CPU behavior. For integer division, you may be able to at least break on the attempt to help isolate your problem by doing something like: Once you have found the problem, don't leave it like that if other people are supposed to maintain your code.. This is not a bug. Per the ISO/IEC 9899:1999 C Standard (§6.5.5): 5 The result of the / operator is the quotient from the division of the first operand by the second; the result of the % operator is the remainder. In both operations, if the value of the second operand is zero, the behavior is undefined. Division by zero results in undefined behavior and the compiler is free to do whatever it likes, including crashing the program, or setting the result to zero and continuing. Sorry for jumping to the wrong conclusion. The problem is that I am doing some numerical calculations (solving nonlinear equations with a nonlinear constraint) and the algorithm I am using sometimes wanders off into nowhere (I am still debugging things). I haven't done any c programming in quite a while, and got used to the crashing behavior when I first learned and thought that was how things operated. Since this is undefined, I would like for the program to throw a SIGFPE, as it does on other platforms, for both integer and floating point division by zero. On FreeBSD and OpenBSD, there is a function fpsetmask that I could use to trap invalid operations and division by zero: Unfortunately, the fpsetmask function doesn't seem to exist on my mac, or at least ieeefp.h doesn't. Is there an equivalent function I can use on my mac? Thanks. Aaron This email sent to site_archiver(a)lists.apple.com

1 0

Re: Apple gcc bug
by Graham J Lee 17 Mar '05

17 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On 17 Mar 2005, at 22:25, Aaron Jackson wrote: On Mar 17, 2005, at 4:55 PM, Matt Watson 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… This is not a bug. Per the ISO/IEC 9899:1999 C Standard (§6.5.5): 5 The result of the / operator is the quotient from the division of the first operand by the second; the result of the % operator is the remainder. In both operations, if the value of the second operand is zero, the behavior is undefined. Division by zero results in undefined behavior and the compiler is free to do whatever it likes, including crashing the program, or setting the result to zero and continuing. Unfortunately, the fpsetmask function doesn't seem to exist on my mac, or at least ieeefp.h doesn't. Is there an equivalent function I can use on my mac? Thanks. Does /usr/include/architecture/`arch`/fenv.h do the same? I haven't programmed for FreeBSD. -- Graham Lee GPG Key ID: 01D5B9D8 UNIX Systems Manager, Oxford Physics Practical Course Coordinator, UKUUG Apple SIG (apple-sig-request(a)lists.ukuug.org) Technical bod, Oxford Mac Users' Group (http://www.oxmug.org) This email sent to site_archiver(a)lists.apple.com

1 0

Re: Apple gcc bug
by Andrew Pinski 17 Mar '05

17 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Mar 17, 2005, at 5:25 PM, Aaron Jackson wrote: -- Pinski _______________________________________________ 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… Unfortunately, the fpsetmask function doesn't seem to exist on my mac, or at least ieeefp.h doesn't. Is there an equivalent function I can use on my mac? Thanks. Again as I have said before PPC, does not throw any exceptions for integer divide by zero. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Apple gcc bug
by Aaron Jackson 17 Mar '05

17 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Mar 17, 2005, at 4:55 PM, Matt Watson wrote: fpsetmask(FP_X_INV | FP_X_DZ); signal(SIGFPE, sig_handler); Aaron _______________________________________________ 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 is not a bug. Per the ISO/IEC 9899:1999 C Standard (§6.5.5): 5 The result of the / operator is the quotient from the division of the first operand by the second; the result of the % operator is the remainder. In both operations, if the value of the second operand is zero, the behavior is undefined. Division by zero results in undefined behavior and the compiler is free to do whatever it likes, including crashing the program, or setting the result to zero and continuing. Sorry for jumping to the wrong conclusion. The problem is that I am doing some numerical calculations (solving nonlinear equations with a nonlinear constraint) and the algorithm I am using sometimes wanders off into nowhere (I am still debugging things). I haven't done any c programming in quite a while, and got used to the crashing behavior when I first learned and thought that was how things operated. Since this is undefined, I would like for the program to throw a SIGFPE, as it does on other platforms, for both integer and floating point division by zero. On FreeBSD and OpenBSD, there is a function fpsetmask that I could use to trap invalid operations and division by zero: Unfortunately, the fpsetmask function doesn't seem to exist on my mac, or at least ieeefp.h doesn't. Is there an equivalent function I can use on my mac? Thanks. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Apple gcc bug
by Matt Watson 17 Mar '05

17 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com matt. _______________________________________________ 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 is not a bug. Per the ISO/IEC 9899:1999 C Standard (§6.5.5): 5 The result of the / operator is the quotient from the division of the first operand by the second; the result of the % operator is the remainder. In both operations, if the value of the second operand is zero, the behavior is undefined. Division by zero results in undefined behavior and the compiler is free to do whatever it likes, including crashing the program, or setting the result to zero and continuing. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Apple gcc bug
by Andrew Pinski 17 Mar '05

17 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Mar 17, 2005, at 4:36 PM, Aaron Jackson wrote: [Aaron-Jacksons-Computer:~] jackson% gcc -v Reading specs from /usr/libexec/gcc/darwin/ppc/3.3/specs Thread model: posix gcc version 3.3 20030304 (Apple Computer, Inc. build 1671) -- Pinski _______________________________________________ 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 am trying to do some numerical calculations and ran into an interesting bug: This is not a bug. division by zero is only undefined. on PPC, division by zero does not raise any exceptions but just the results are undefined. This email sent to site_archiver(a)lists.apple.com

1 0

Apple gcc bug
by Aaron Jackson 17 Mar '05

17 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com [Aaron-Jacksons-Computer:~] jackson% gcc -v Reading specs from /usr/libexec/gcc/darwin/ppc/3.3/specs Thread model: posix gcc version 3.3 20030304 (Apple Computer, Inc. build 1671) [Aaron-Jacksons-Computer:~] jackson% cat test4.c #include <stdio.h> int main(int argc, char **argv) { int k; k=1/0; printf("k = %d\n",k); k=0/0; printf("k = %d\n",k); exit(0); } [Aaron-Jacksons-Computer:~] jackson% gcc -o test4 test4.c test4.c: In function `main': test4.c:7: warning: division by zero test4.c:9: warning: division by zero [Aaron-Jacksons-Computer:~] jackson% ./test4 k = 0 k = 0 Aaron _______________________________________________ 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 am trying to do some numerical calculations and ran into an interesting bug: I have tried this on two other macs with the same result. Of course, this program properly bombs out on the openbsd and solaris systems that I tried to run it on. I did some google searching and looked through the list archive, but didn't see this mentioned. Is this a known bug? If so, are there any workarounds for this, or which versions of gcc are not effected? Thanks. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Can't mount_afp japanese volumes
by Alexey Proskuryakov 17 Mar '05

17 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Thu, 17 Mar 2005 18:30:55 +0100 Stephane Thiell <sthiell(a)intego.com> wrote: - WBR, Alexey Proskuryakov _______________________________________________ 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… But these tools work well with, for example, volumes with french accents in UTF-8. Have you tried running the tools with primary Japanese language (as set in the International control panel)? Looks like AppleShare client uses the system encoding for server and volume names. Actually, the AFP 3.1 documentation states that a volume name is a string of up to 27 8-bit ASCII characters (Concepts->File System Structure). I've never heard of 8-bit ASCII, but that should probably mean the system encoding, such as MacRoman or MacJapanese. I'd also try running tcpdump to check how the French accented characters are actually sent, in UTF-8 or MacRoman. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Sending notification from a time-constraint thread
by Mike Smith 17 Mar '05

17 Mar '05

site_archiver(a)lists.apple.com Delivered-To: Darwin-dev(a)lists.apple.com Jim, Ok. To be sure I've understood correctly: when using RPC: (leaving this text for context) = Mike _______________________________________________ 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 in over my head at this point when it comes to giving accurate advice; I appreciate you're terribly busy... On Mar 17, 2005, at 1:04 AM, Stéphane Letz wrote: Le 16 mars 05, à 22:10, Mike Smith a écrit : On Mar 16, 2005, at 12:57 PM, Stéphane LETZ wrote: In a server/client system we are developing, we need a time-constraint thread in the server to send a notification to another thread in the server. We use MIG generated RPC calls to communicate between threads. Why do you do this? Because the server "non time-constraint" thread already "listen" to RPC calls coming from clients. So the idea was to use the same RPC system for server inter-thread communication also. Ok. But this then puts your time-constraint thread at the mercy of the other thread's workload, as you have discovered. - an asynchronous call (simpleroutine) *without* timeout will return immediately if the receiver queue is not full - an asynchronous call (simpleroutine) *without* timeout will block if the received queue is full - an asynchronous call (simpleroutine) *with* timeout will return immediately if the receiver queue is not full - an asynchronous call (simpleroutine) *with* timeout will return an "error (ipc/send) timed out " message if the receiver queue is full Is is correct? I'm going to leave this one to Jim. Your conclusions look correct to me, but I'm not the authority on this. Unless you mean "between client and server", you would generally be much better off implementing a trivial messaging system using posix condition variables. Is is possible to have the server thread answer to RPC calls and handle condition variables at the same time? No; if it's blocked waiting for an incoming RPC, it can't also block waiting for a cvar. Your best bet is probably to have a helper thread blocked on the cvar, which will turn around and message the server thread as needed. This decouples your TC thread from the RPC; depending on your architecture you can have this helper thread handle fan-in for multiple TC threads. It's a bit miserable needing yet another thread, but the realities of queueing (RPC) vs. asynchronous messaging (cvars) mean that you need a transition interface of some sort. This email sent to site_archiver(a)lists.apple.com

1 0

Can't mount_afp japanese volumes
by Stephane Thiell 17 Mar '05

17 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Hello, Thanks in advance! St~{(&~}phane Thiell _______________________________________________ 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 seeking a way to mount, in command line, a AFP volume with a _japanese_ name, like "~{92SP~}X". The Finder, at least since Jaguar, is able to list and mount a remote japanese volume without any problem. I tried without success the mount_afp command and osascript -e with "tell application Finder to mount volume afp://...". Often I get afpParmErr (A specified parameter was out of allowable range). But these tools work well with, for example, volumes with french accents in UTF-8. I tried to escape the UTF-8 characters with CFURLCreateStringByAddingPercentEscapes() too, but it didn't work neither. Is it a way to mount such volumes without having to use the Carbon layer? Maybe their names should be escaped in a special manner? This email sent to site_archiver(a)lists.apple.com

1 0