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

Re: Getting entire Darwin source tree via CVS
by seebs＠plethora.net 02 Mar '05

02 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com In message <a06210205be4acef47420(a)[10.250.3.49]>, OL&L Lists writes: >Is there a way to download the entire Darwin CVS source tree in one >shot? I connected to Apple's Darwin CVS repository, but it lists only >a few certain modules as being available. Is there a way to download >the entire source tree via CVS? There doesn't appear to be, and if you get all the packages from the page, they're not really a build tree. I would love a way to check out a complete tree for a given version automatically; unfortunately, the OpenDarwin people are quite out of date, and Apple doesn't offer a complete CVS tree. -s _______________________________________________ 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

Getting entire Darwin source tree via CVS
by OL&L Lists 02 Mar '05

02 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Thanks, Michael _______________________________________________ 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… Is there a way to download the entire Darwin CVS source tree in one shot? I connected to Apple's Darwin CVS repository, but it lists only a few certain modules as being available. Is there a way to download the entire source tree via CVS? This email sent to site_archiver(a)lists.apple.com

1 0

Re: __attribute__((constructor)) in dylib does not run with DYLD_INSERT_LIBRARIES
by Brian Dantes 02 Mar '05

02 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Microsoft-Entourage/10.1.4.030702.0 Peter, I was hoping you might chime in - but of course I don't like the answer. ;-> There may not be any committee official standard requiring the preload static constructor behavior I want, but I would argue there certainly is a de facto one. I'm trying to think of some hack where I override some weak symbol (does dyld even support those?) that's referenced during the static initialization phase of any program. Is there any way I could hijack _main itself? Of course, my library would need to find its way back to the original entry point (ala something like Jonathan Rentzsch's mach_override) somehow. I haven't tried this - I suspect dyld will either complain at runtime or simply ignore my definition of _main entirely. Is there some undocumented hook in Panther? -Brian > From: "Peter O'Gorman" <peter(a)pogma.com> > Date: Wed, 02 Mar 2005 08:24:36 +0900 > To: Andreas Tobler <toa(a)pop.agri.ch> > Cc: darwin-dev(a)lists.apple.com > Subject: Re: __attribute__((constructor)) in dylib does not run with > DYLD_INSERT_LIBRARIES > > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > Andreas Tobler wrote: > > | > | So if you build libhello with -bundle instead of -dynamiclib it works > | somehow. But this is a no go for me. > > Why? > > | > | I'm not aware of any workaround for panther. > > Since you are dlopening the library, and panther uses dlcompat, and dlcompat > looks for a symbol named "_init", you can use a hack. Inserting a "void > _init() { ; }" into the .c file which has your static constructors should > cause them to be called. They must appear in the same translation unit as > the static constructor, or you must link your library using the > - -single_module flag if you want dyld to call them. > > If I were you I'd check for different dlopen and dyld behavior on tiger... > > There is no workaround for getting libraries inserted with > DYLD_INSERT_LIBRARIES to have their static constructors called on panther > that I know of. Nor is there any standard that I am aware of which requires > this functionality. > > Peter > - -- > Peter O'Gorman - http://www.pogma.com > -----BEGIN PGP SIGNATURE----- > Version: GnuPG v1.2.6 (Darwin) > > iQCVAwUBQiT5s7iDAg3OZTLPAQIPogP/b0oAFLodhHCTH3aQHq7o++Za/b61kTm/ > TpL8y9mSmtlqdxLKFdLklncZ8foGd+QE0sGDCGs6oJCwRQiYRga6lSFwnAqoe2aE > crLnwwKCKfUHlqfF4XZMdut7qu3BgrgBAVklKMCH5HzVx/NWwQmIMEp5x+MWalZK > hI4gzVszV3s= > =b4cx > -----END PGP SIGNATURE----- > _______________________________________________ > 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/brian%40things.org > > This email sent to brian(a)things.org > _______________________________________________ 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: __attribute__((constructor)) in dylib does not run with DYLD_INSERT_LIBRARIES
by Peter O'Gorman 01 Mar '05

01 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Mozilla Thunderbird 1.0 (Macintosh/20041206) -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Andreas Tobler wrote: | | So if you build libhello with -bundle instead of -dynamiclib it works | somehow. But this is a no go for me. Why? | | I'm not aware of any workaround for panther. Since you are dlopening the library, and panther uses dlcompat, and dlcompat looks for a symbol named "_init", you can use a hack. Inserting a "void _init() { ; }" into the .c file which has your static constructors should cause them to be called. They must appear in the same translation unit as the static constructor, or you must link your library using the - -single_module flag if you want dyld to call them. If I were you I'd check for different dlopen and dyld behavior on tiger... There is no workaround for getting libraries inserted with DYLD_INSERT_LIBRARIES to have their static constructors called on panther that I know of. Nor is there any standard that I am aware of which requires this functionality. Peter - -- Peter O'Gorman - http://www.pogma.com -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.6 (Darwin) iQCVAwUBQiT5s7iDAg3OZTLPAQIPogP/b0oAFLodhHCTH3aQHq7o++Za/b61kTm/ TpL8y9mSmtlqdxLKFdLklncZ8foGd+QE0sGDCGs6oJCwRQiYRga6lSFwnAqoe2aE crLnwwKCKfUHlqfF4XZMdut7qu3BgrgBAVklKMCH5HzVx/NWwQmIMEp5x+MWalZK hI4gzVszV3s= =b4cx -----END PGP SIGNATURE----- _______________________________________________ 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: __attribute__((constructor)) in dylib does not run with DYLD_INSERT_LIBRARIES
by Andreas Tobler 01 Mar '05

01 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Mozilla Thunderbird 1.0 (Macintosh/20041206) Andreas, Thanks mucho for the confirmation, but aaargh. Any workaround in Panther you're aware of? Do you have a Radar number? I'd like to reference yours in the bug report I made, if so. 3824565 My sample I pasted there looks like this: ------ libhello.cc #include <stdio.h> int i = printf("hello, I'm a static constructor\n"); void foo_print(void) { } ------ demo_hello.cc // Compiled with: g++ \ // -g -o demo_hello demo_hello.cc -L. -ldl #include <dlfcn.h> #include <stdlib.h> #include <stdio.h> int main(void) { const char *error; void *module; module = dlopen("libhello.so", RTLD_LAZY); if (!module) { fprintf(stderr, "Couldn't open libhello.so: %s\n", dlerror()); exit(1); } /* All done, close things cleanly */ dlclose(module); return 0; } I'm not aware of any workaround for panther. 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… Brian Dantes wrote: So if you build libhello with -bundle instead of -dynamiclib it works somehow. But this is a no go for me. This email sent to site_archiver(a)lists.apple.com

1 0

Re: __attribute__((constructor)) in dylib does not run with DYLD_INSERT_LIBRARIES
by Brian Dantes 01 Mar '05

01 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Microsoft-Entourage/10.1.6.040913.0 Andreas, Thanks mucho for the confirmation, but aaargh. Any workaround in Panther you're aware of? Do you have a Radar number? I'd like to reference yours in the bug report I made, if so. -Brian > From: "Andreas Tobler" <toa(a)pop.agri.ch> > Date: Tue, 1 Mar 2005 11:38:33 +0100 > To: "Brian Dantes" <brian(a)things.org>, darwin-dev(a)lists.apple.com > Subject: Re: __attribute__((constructor)) in dylib does not run with > DYLD_INSERT_LIBRARIES > > This is a bug in dyld/dylib. I filed a report back in Oct 04. > > Seems to be fixed in tiger. > > [wolfram:~] andreast% ./bar > Constructor called! > > 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… This email sent to site_archiver(a)lists.apple.com

1 0

Re: __attribute__((constructor)) in dylib does not run with DYLD_INSERT_LIBRARIES
by Andreas Tobler 01 Mar '05

01 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com This is a bug in dyld/dylib. I filed a report back in Oct 04. Seems to be fixed in tiger. [wolfram:~] andreast% ./bar Constructor called! 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… This email sent to site_archiver(a)lists.apple.com

1 0

Re: __attribute__((constructor)) in dylib does not run with DYLD_INSERT_LIBRARIES
by Brian Dantes 01 Mar '05

01 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Microsoft-Entourage/10.1.6.040913.0 Static objects and constructors of any kind must be by their very nature always be considered "used" - unless the compiler can determine for a fact that the code initializing them has no side-effects. It's not the same as a simple integral global variable which is never referenced. The same would be true of a basic global variable in the main image initialized by a function call of some sort (C++, not C, of course). But even if what you said were the case, even under MacOS X, the static constructors run for *all* static objects in a library if even one symbol in that library is referenced. So the behavior is not consistent with your reasoning. So the Mach-O lazy symbol evaluation is coming into play here. I tried setting DYLD_BIND_AT_LAUNCH - but it didn't help. But the bottom line is that there needs to be some way of injecting a custom preamble into third party code. If not this technique that works on other Unixes, then what is the right sauce for Mach? -Brian > From: Markus Hitter <mah(a)jump-ing.de> > Date: Tue, 1 Mar 2005 10:20:39 +0100 > To: Brian Dantes <brian(a)things.org> > Cc: <darwin-dev(a)lists.apple.com> > Subject: Re: __attribute__((constructor)) in dylib does not run with > DYLD_INSERT_LIBRARIES > > > Am 01.03.2005 um 00:58 schrieb Brian Dantes: > >> If this were a static library, you certainly would expect any static >> constructors to run that were defined in that library. > > (No). Constructors should be run when an object is created, if it is > "used". > > With your understanding, behaviour of a program would be different > depending on wether unused code is stripped or not. > > > Markus > > - - - - - - - - - - - - - - - - - - - > Dipl. Ing. Markus Hitter > http://www.jump-ing.de/ _______________________________________________ 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: __attribute__((constructor)) in dylib does not run with DYLD_INSERT_LIBRARIES
by Markus Hitter 01 Mar '05

01 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Am 01.03.2005 um 00:58 schrieb Brian Dantes: Markus - - - - - - - - - - - - - - - - - - - Dipl. Ing. Markus Hitter http://www.jump-ing.de/ _______________________________________________ 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 this were a static library, you certainly would expect any static constructors to run that were defined in that library. (No). Constructors should be run when an object is created, if it is "used". With your understanding, behaviour of a program would be different depending on wether unused code is stripped or not. This email sent to site_archiver(a)lists.apple.com

1 0

Re: __attribute__((constructor)) in dylib does not run with DYLD_INSERT_LIBRARIES
by Chris Ridd 01 Mar '05

01 Mar '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Microsoft-Entourage/11.1.0.040913 On 28/2/05 11:58 pm, Brian Dantes <brian(a)things.org> wrote: > If this were a static library, you certainly would expect any static > constructors to run that were defined in that library. The only difference > in that scenario and this one is when the library is loaded into the image. > I would hazard to say that Linux (and FreeBSD, HPUX, Solaris and other It would be surprising if Linux behaved differently from FreeBSD and Solaris, because they all use ELF. Some versions of HP-UX use ELF too. So your "weight of numbers" is only really 1 :-( > variants I've directly tried this on) are correct and MacOS X is not. > > I have submitted this as Radar bug report to Apple as well, so we'll see if > anything comes of that. Also check the archives of this list, as this subject has kind of come up before. (Mostly wrt C++ constructors, but it sounds like the same sort of thing.) Cheers, Chris _______________________________________________ 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