Re: [Fink-devel] Compiler mess on Lion (Was Re: libgeos problem)
On 2/12/11 22:04, Jack Howarth wrote: On Fri, Dec 02, 2011 at 09:59:31PM +0100, Martin Costabel wrote: [] I installed llvm30 from your new package on the submission tracker. No problem installing. Unfortunately, the fatal bug is still there: Building scribus with the new clang fails exactly as with /usr/bin/clang: [] I would suggest opening a PR on http://llvm.org/bugs/ against clang and attach the preprocessed source obtained with --save-temps for the failing compilation. This may be a corner-case not covered by r141515. After looking at this some more, I am currently convinced that this is actually a bug in the Scribus sources that is undetected by the other compilers. I think clang is right, and the resemblance with the error fixed by r141515 is a coincidence. But I am not a C++ inheritance lawyer. In addition, I found a tiny patch that fools clang into overlooking the error as well. With this patch, I am now able to build scribus on Lion, with your new clang-3. With the old /usr/bin/clang (and the patch), the compilation stops at the same place where llvm-g++-4.2 stops. Or rather, it doesn't stop, it goes into an infinite loop. Thus, the new clang-3 from your llvm30 package fixes an important bug, and I am eager to see it added to Fink (maintainer CCed). -- Martin -- All the data continuously generated in your IT infrastructure contains a definitive record of customers, application performance, security threats, fraudulent activity, and more. Splunk takes this data and makes sense of it. IT sense. And common sense. http://p.sf.net/sfu/splunk-novd2d ___ Fink-devel mailing list Fink-devel@lists.sourceforge.net List archive: http://news.gmane.org/gmane.os.apple.fink.devel Subscription management: https://lists.sourceforge.net/lists/listinfo/fink-devel
Re: [Fink-devel] Compiler mess on Lion (Was Re: libgeos problem)
On 28/11/11 15:54, Jack Howarth wrote: On Mon, Nov 28, 2011 at 09:42:44AM +0100, Martin Costabel wrote: [] I have been trying to build the scribus package on Lion, without success: - With clang, it runs into a clang bug that was detected a year ago http://lists.cs.uiuc.edu/pipermail/llvmbugs/2010-November/015821.html and has been fixed in the clang sources some time ago, but is still present in current xcode-4.2.1. Martin, Considering that llvm/clang 3.0 will be released shortly a better option might be to BuildDepends on llvm30 once it is added to fink and use those clang compilers. I installed llvm30 from your new package on the submission tracker. No problem installing. Unfortunately, the fatal bug is still there: Building scribus with the new clang fails exactly as with /usr/bin/clang: In file included from /sw/src/fink.build/scribus-x11-1.4.0-72.rc6/scribus-1.4.0.rc6/scribus/desaxe/digester.cpp:20: /sw/src/fink.build/scribus-x11-1.4.0-72.rc6/scribus-1.4.0.rc6/scribus/desaxe/actions.h:160:45: error: 'body' is a private member of 'desaxe::Action' return static_castGenerator_bodyType*(body)-eval(dig, tag, attr); ^ /sw/src/fink.build/scribus-x11-1.4.0-72.rc6/scribus-1.4.0.rc6/scribus/desaxe/actions.h:101:15: note: declared private here Action_body* body; ^ From the llvm bug system, I got the impression that this was fixed by now. In fact, svn r141515 was supposed to fix it. For scribus, it didn't. I find still no way to build scribus on Lion. -- Martin -- All the data continuously generated in your IT infrastructure contains a definitive record of customers, application performance, security threats, fraudulent activity, and more. Splunk takes this data and makes sense of it. IT sense. And common sense. http://p.sf.net/sfu/splunk-novd2d ___ Fink-devel mailing list Fink-devel@lists.sourceforge.net List archive: http://news.gmane.org/gmane.os.apple.fink.devel Subscription management: https://lists.sourceforge.net/lists/listinfo/fink-devel
Re: [Fink-devel] Compiler mess on Lion (Was Re: libgeos problem)
On Fri, Dec 02, 2011 at 09:59:31PM +0100, Martin Costabel wrote: On 28/11/11 15:54, Jack Howarth wrote: On Mon, Nov 28, 2011 at 09:42:44AM +0100, Martin Costabel wrote: [] I have been trying to build the scribus package on Lion, without success: - With clang, it runs into a clang bug that was detected a year ago http://lists.cs.uiuc.edu/pipermail/llvmbugs/2010-November/015821.html and has been fixed in the clang sources some time ago, but is still present in current xcode-4.2.1. Martin, Considering that llvm/clang 3.0 will be released shortly a better option might be to BuildDepends on llvm30 once it is added to fink and use those clang compilers. I installed llvm30 from your new package on the submission tracker. No problem installing. Unfortunately, the fatal bug is still there: Building scribus with the new clang fails exactly as with /usr/bin/clang: In file included from /sw/src/fink.build/scribus-x11-1.4.0-72.rc6/scribus-1.4.0.rc6/scribus/desaxe/digester.cpp:20: /sw/src/fink.build/scribus-x11-1.4.0-72.rc6/scribus-1.4.0.rc6/scribus/desaxe/actions.h:160:45: error: 'body' is a private member of 'desaxe::Action' return static_castGenerator_bodyType*(body)-eval(dig, tag, attr); ^ /sw/src/fink.build/scribus-x11-1.4.0-72.rc6/scribus-1.4.0.rc6/scribus/desaxe/actions.h:101:15: note: declared private here Action_body* body; ^ From the llvm bug system, I got the impression that this was fixed by now. In fact, svn r141515 was supposed to fix it. For scribus, it didn't. I find still no way to build scribus on Lion. I would suggest opening a PR on http://llvm.org/bugs/ against clang and attach the preprocessed source obtained with --save-temps for the failing compilation. This may be a corner-case not covered by r141515. Jack -- Martin -- All the data continuously generated in your IT infrastructure contains a definitive record of customers, application performance, security threats, fraudulent activity, and more. Splunk takes this data and makes sense of it. IT sense. And common sense. http://p.sf.net/sfu/splunk-novd2d ___ Fink-devel mailing list Fink-devel@lists.sourceforge.net List archive: http://news.gmane.org/gmane.os.apple.fink.devel Subscription management: https://lists.sourceforge.net/lists/listinfo/fink-devel
[Fink-devel] Compiler mess on Lion (Was Re: libgeos problem)
On 27/11/11 21:50, Alexander Hansen wrote: [] We should probably do an apple-gcc-4.2 package, then. It might even be handy on 10.7. If it builds on 10.7, I am all for it. I have been trying to build the scribus package on Lion, without success: - With clang, it runs into a clang bug that was detected a year ago http://lists.cs.uiuc.edu/pipermail/llvmbugs/2010-November/015821.html and has been fixed in the clang sources some time ago, but is still present in current xcode-4.2.1. - With llvm-gcc42, the compiler goes into an infinite loop (100% cpu, runs for hours without doing anything). (If anyone has the ear of an Apple employee, would you please box or twist it violently, for giving us an operating system version whose compilers have serious - and known - bugs) - With Fink's gcc-fsf-4.6, building fails already at the (cmake-) configure stage, because cmake insists on placing a -arch x86_64 flag on the compiler line that makes gcc46 error out. And as usual with cmake, it is impossible to find out where this flag comes from and how to remove it. Why doesn't our gcc46 simply ignore the -arch flag instead of giving an error? These problems have also been detected by macports https://trac.macports.org/ticket/30835 In that bug ticket, one finds the remark gcc42 does not build for Lion. No viable workaround. I don't know if this means that their apple-gcc42 does not build on Lion, or that scribus doesn't even build with that compiler either. Note: This version of scribus builds OK on 10.5 and 10.6/pre-xcode-4.2. -- Martin -- All the data continuously generated in your IT infrastructure contains a definitive record of customers, application performance, security threats, fraudulent activity, and more. Splunk takes this data and makes sense of it. IT sense. And common sense. http://p.sf.net/sfu/splunk-novd2d ___ Fink-devel mailing list Fink-devel@lists.sourceforge.net List archive: http://news.gmane.org/gmane.os.apple.fink.devel Subscription management: https://lists.sourceforge.net/lists/listinfo/fink-devel
Re: [Fink-devel] Compiler mess on Lion (Was Re: libgeos problem)
On Mon, Nov 28, 2011 at 09:42:44AM +0100, Martin Costabel wrote: On 27/11/11 21:50, Alexander Hansen wrote: [] We should probably do an apple-gcc-4.2 package, then. It might even be handy on 10.7. If it builds on 10.7, I am all for it. I have been trying to build the scribus package on Lion, without success: - With clang, it runs into a clang bug that was detected a year ago http://lists.cs.uiuc.edu/pipermail/llvmbugs/2010-November/015821.html and has been fixed in the clang sources some time ago, but is still present in current xcode-4.2.1. Martin, Considering that llvm/clang 3.0 will be released shortly a better option might be to BuildDepends on llvm30 once it is added to fink and use those clang compilers. - With llvm-gcc42, the compiler goes into an infinite loop (100% cpu, runs for hours without doing anything). (If anyone has the ear of an Apple employee, would you please box or twist it violently, for giving us an operating system version whose compilers have serious - and known - bugs) Unfortunately, I really doubt we will see many fixes applied to Apple's llvm-gcc. They have based it on an older llvm release which means any fixes will have to be backported (and their general concept is to avoid non-essential fixes when clang works). Using llvm.org's clang in these problem cases and pinging Apple to backport those fixes to their current clang compiler is far more likely to be successful. - With Fink's gcc-fsf-4.6, building fails already at the (cmake-) configure stage, because cmake insists on placing a -arch x86_64 flag on the compiler line that makes gcc46 error out. And as usual with cmake, it is impossible to find out where this flag comes from and how to remove it. Why doesn't our gcc46 simply ignore the -arch flag instead of giving an error? File an upstream PR against FSF gcc if you want that enhancement. IMHO, the compiler might as well error to provide a clear notice to the user that their use of -arch won't be honored. These problems have also been detected by macports https://trac.macports.org/ticket/30835 In that bug ticket, one finds the remark gcc42 does not build for Lion. No viable workaround. I don't know if this means that their apple-gcc42 does not build on Lion, or that scribus doesn't even build with that compiler either. I'd rather avoid the apple-gcc42 approach if possible since Apple seems unlikely to continue to provide fixes to their gcc-4.2 sources for later Xcode releases. I can well imagine that that this will cause issues later on. It would be far better to use clang from llvm30 if possible. Jack ps I have been doing regular builds of llvm 3.0 and its works well under Lion. Also major progress has been made in allowing dragonegg to provide full support for vectorization using -msse4 -fplugin-arg-dragonegg-enable-gcc-optzns. http://lists.cs.uiuc.edu/pipermail/llvmdev/2011-October/044091.html ` Note: This version of scribus builds OK on 10.5 and 10.6/pre-xcode-4.2. -- Martin -- All the data continuously generated in your IT infrastructure contains a definitive record of customers, application performance, security threats, fraudulent activity, and more. Splunk takes this data and makes sense of it. IT sense. And common sense. http://p.sf.net/sfu/splunk-novd2d ___ Fink-devel mailing list Fink-devel@lists.sourceforge.net List archive: http://news.gmane.org/gmane.os.apple.fink.devel Subscription management: https://lists.sourceforge.net/lists/listinfo/fink-devel
