Re: [Fink-devel] r-base215. r-base-30 and r-base31
On Mon, 28 Apr 2014 22:38:02 -0400, Daniel Macks dma...@netspace.org wrote: On Mon, 28 Apr 2014 20:55:18 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: The r-base214 packaging seems to have test suite issues when built against Xcode 5.1 on darwin12… Testing examples for package ‘utils’ /sw/src/fink.build/r-base214-2.14.2-9/R-2.14.2/bin/BATCH: line 60: 34097 Trace/BPT trap: 5 ${R_HOME}/bin/R -f ${in} ${opts} ${R_BATCH_OPTIONS} ${out} 21 Error: testing 'utils' failed Execution halted The failure appears to be due to an unresolved __pcre_valid_utf (or similarly named) symbol in libR.dylib. The R library is trying to use a private symbol in libpcre by guessing what it's called (and what its parameters are) in various different libpcre versions, but our most recent libpcre dropped that symbol altogether. Obviously a fragile situation to rely on undocumented non-public content. One useful change we can make is to patch out the -undefined dynamic_lookup from the configure script. That flag causes undefined symbols to be ignored by the linker, leaving them to cause problems at runtime. By removing the flag, the linking itself fails right away rather than leaving a possibly mis-built library. ..which reveals that util.dylib is missing -llzma in even in r-base30 and r-base31 (ones that do not have the pcre problem). dan -- Daniel Macks dma...@netspace.org -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free. http://p.sf.net/sfu/SauceLabs ___ 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] r-base215. r-base-30 and r-base31
Daniel, Isn't the real problem that r-base in fink is being built with --with-system-pcre but the fink lib pcre.1.dylib ends up linked into libR. I assume we need a BuildConflicts on libpcre1 in order to have the system pcre used. Jack On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: On Mon, 28 Apr 2014 22:38:02 -0400, Daniel Macks dma...@netspace.org wrote: On Mon, 28 Apr 2014 20:55:18 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: The r-base214 packaging seems to have test suite issues when built against Xcode 5.1 on darwin12… Testing examples for package ‘utils’ /sw/src/fink.build/r-base214-2.14.2-9/R-2.14.2/bin/BATCH: line 60: 34097 Trace/BPT trap: 5 ${R_HOME}/bin/R -f ${in} ${opts} ${R_BATCH_OPTIONS} ${out} 21 Error: testing 'utils' failed Execution halted The failure appears to be due to an unresolved __pcre_valid_utf (or similarly named) symbol in libR.dylib. The R library is trying to use a private symbol in libpcre by guessing what it's called (and what its parameters are) in various different libpcre versions, but our most recent libpcre dropped that symbol altogether. Obviously a fragile situation to rely on undocumented non-public content. One useful change we can make is to patch out the -undefined dynamic_lookup from the configure script. That flag causes undefined symbols to be ignored by the linker, leaving them to cause problems at runtime. By removing the flag, the linking itself fails right away rather than leaving a possibly mis-built library. ..which reveals that util.dylib is missing -llzma in even in r-base30 and r-base31 (ones that do not have the pcre problem). dan -- Daniel Macks dma...@netspace.org -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free. http://p.sf.net/sfu/SauceLabs___ 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] r-base215. r-base-30 and r-base31
Using (apparently-)undocumented, internal implementation details that are known to change in different versions is *always* a problem. Any solution that doesn't involve actually not-doing-that is merely deferring the same failure from occurring again in the future when those details change again. dan On Tue, 29 Apr 2014 13:12:14 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: Daniel, Isn't the real problem that r-base in fink is being built with --with-system-pcre but the fink lib pcre.1.dylib ends up linked into libR. I assume we need a BuildConflicts on libpcre1 in order to have the system pcre used. Jack On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: On Mon, 28 Apr 2014 22:38:02 -0400, Daniel Macks dma...@netspace.org wrote: On Mon, 28 Apr 2014 20:55:18 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: The r-base214 packaging seems to have test suite issues when built against Xcode 5.1 on darwin12… Testing examples for package ‘utils’ /sw/src/fink.build/r-base214-2.14.2-9/R-2.14.2/bin/BATCH: line 60: 34097 Trace/BPT trap: 5 ${R_HOME}/bin/R -f ${in} ${opts} ${R_BATCH_OPTIONS} ${out} 21 Error: testing 'utils' failed Execution halted The failure appears to be due to an unresolved __pcre_valid_utf (or similarly named) symbol in libR.dylib. The R library is trying to use a private symbol in libpcre by guessing what it's called (and what its parameters are) in various different libpcre versions, but our most recent libpcre dropped that symbol altogether. Obviously a fragile situation to rely on undocumented non-public content. One useful change we can make is to patch out the -undefined dynamic_lookup from the configure script. That flag causes undefined symbols to be ignored by the linker, leaving them to cause problems at runtime. By removing the flag, the linking itself fails right away rather than leaving a possibly mis-built library. ..which reveals that util.dylib is missing -llzma in even in r-base30 and r-base31 (ones that do not have the pcre problem). dan -- Daniel Macks dma...@netspace.org - -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free. http://p.sf.net/sfu/SauceLabs - ___ 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 -- Daniel Macks dma...@netspace.org -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free. http://p.sf.net/sfu/SauceLabs ___ 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] r-base215. r-base-30 and r-base31
Isn't this just a header mismatch? We have both a build depends on libpcre1 and --with-system-pcre which is illogical. I am testing with --with-system-pcre changed to --with-pcre=%p. FYI, macports doesn't pass either flag and incorrectly ignores pcre. On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: Using (apparently-)undocumented, internal implementation details that are known to change in different versions is *always* a problem. Any solution that doesn't involve actually not-doing-that is merely deferring the same failure from occurring again in the future when those details change again. dan On Tue, 29 Apr 2014 13:12:14 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: Daniel, Isn't the real problem that r-base in fink is being built with --with-system-pcre but the fink lib pcre.1.dylib ends up linked into libR. I assume we need a BuildConflicts on libpcre1 in order to have the system pcre used. Jack On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: On Mon, 28 Apr 2014 22:38:02 -0400, Daniel Macks dma...@netspace.org wrote: On Mon, 28 Apr 2014 20:55:18 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: The r-base214 packaging seems to have test suite issues when built against Xcode 5.1 on darwin12… Testing examples for package ‘utils’ /sw/src/fink.build/r-base214-2.14.2-9/R-2.14.2/bin/BATCH: line 60: 34097 Trace/BPT trap: 5 ${R_HOME}/bin/R -f ${in} ${opts} ${R_BATCH_OPTIONS} ${out} 21 Error: testing 'utils' failed Execution halted The failure appears to be due to an unresolved __pcre_valid_utf (or similarly named) symbol in libR.dylib. The R library is trying to use a private symbol in libpcre by guessing what it's called (and what its parameters are) in various different libpcre versions, but our most recent libpcre dropped that symbol altogether. Obviously a fragile situation to rely on undocumented non-public content. One useful change we can make is to patch out the -undefined dynamic_lookup from the configure script. That flag causes undefined symbols to be ignored by the linker, leaving them to cause problems at runtime. By removing the flag, the linking itself fails right away rather than leaving a possibly mis-built library. ..which reveals that util.dylib is missing -llzma in even in r-base30 and r-base31 (ones that do not have the pcre problem). dan -- Daniel Macks dma...@netspace.org - -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free. http://p.sf.net/sfu/SauceLabs - ___ 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 -- Daniel Macks dma...@netspace.org -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free. http://p.sf.net/sfu/SauceLabs___ 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] r-base215. r-base-30 and r-base31
(sorry if this email goes out twice!) I don't think that's the meaning of system. R appears to have optional onboard/internal sources of various dependencies as an alternative to using ones existing on the system. *Where* on the system is a different issue. You can probably check the .d files to see exactly which headers are being loaded to see if it's matched. dan On Tue, 29 Apr 2014 13:21:08 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: Isn't this just a header mismatch? We have both a build depends on libpcre1 and --with-system-pcre which is illogical. I am testing with --with-system-pcre changed to --with-pcre=%p. FYI, macports doesn't pass either flag and incorrectly ignores pcre. On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: Using (apparently-)undocumented, internal implementation details that are known to change in different versions is *always* a problem. Any solution that doesn't involve actually not-doing-that is merely deferring the same failure from occurring again in the future when those details change again. dan On Tue, 29 Apr 2014 13:12:14 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: Daniel, Isn't the real problem that r-base in fink is being built with --with-system-pcre but the fink lib pcre.1.dylib ends up linked into libR. I assume we need a BuildConflicts on libpcre1 in order to have the system pcre used. Jack On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: On Mon, 28 Apr 2014 22:38:02 -0400, Daniel Macks dma...@netspace.org wrote: On Mon, 28 Apr 2014 20:55:18 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: The r-base214 packaging seems to have test suite issues when built against Xcode 5.1 on darwin12… Testing examples for package ‘utils’ /sw/src/fink.build/r-base214-2.14.2-9/R-2.14.2/bin/BATCH: line 60: 34097 Trace/BPT trap: 5 ${R_HOME}/bin/R -f ${in} ${opts} ${R_BATCH_OPTIONS} ${out} 21 Error: testing 'utils' failed Execution halted The failure appears to be due to an unresolved __pcre_valid_utf (or similarly named) symbol in libR.dylib. The R library is trying to use a private symbol in libpcre by guessing what it's called (and what its parameters are) in various different libpcre versions, but our most recent libpcre dropped that symbol altogether. Obviously a fragile situation to rely on undocumented non-public content. One useful change we can make is to patch out the -undefined dynamic_lookup from the configure script. That flag causes undefined symbols to be ignored by the linker, leaving them to cause problems at runtime. By removing the flag, the linking itself fails right away rather than leaving a possibly mis-built library. ..which reveals that util.dylib is missing -llzma in even in r-base30 and r-base31 (ones that do not have the pcre problem). dan -- Daniel Macks dma...@netspace.org - -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free. http://p.sf.net/sfu/SauceLabs - ___ 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 -- Daniel Macks dma...@netspace.org - -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free. http://p.sf.net/sfu/SauceLabs - ___ 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 -- Daniel Macks dma...@netspace.org -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free. http://p.sf.net/sfu/SauceLabs ___ Fink-devel mailing list Fink-devel@lists.sourceforge.net
Re: [Fink-devel] r-base215. r-base-30 and r-base31
Well, empirically it fixed the r-base214 build on 10.8 for me by just changing --with-system-pcre to --with-pcre=%p. It would be highly irregular for the --with-system-pcre option not to be pushing the headers in /usr/include to be used. On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: (sorry if this email goes out twice!) I don't think that's the meaning of system. R appears to have optional onboard/internal sources of various dependencies as an alternative to using ones existing on the system. *Where* on the system is a different issue. You can probably check the .d files to see exactly which headers are being loaded to see if it's matched. dan On Tue, 29 Apr 2014 13:21:08 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: Isn't this just a header mismatch? We have both a build depends on libpcre1 and --with-system-pcre which is illogical. I am testing with --with-system-pcre changed to --with-pcre=%p. FYI, macports doesn't pass either flag and incorrectly ignores pcre. On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: Using (apparently-)undocumented, internal implementation details that are known to change in different versions is *always* a problem. Any solution that doesn't involve actually not-doing-that is merely deferring the same failure from occurring again in the future when those details change again. dan On Tue, 29 Apr 2014 13:12:14 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: Daniel, Isn't the real problem that r-base in fink is being built with --with-system-pcre but the fink lib pcre.1.dylib ends up linked into libR. I assume we need a BuildConflicts on libpcre1 in order to have the system pcre used.Jack On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: On Mon, 28 Apr 2014 22:38:02 -0400, Daniel Macks dma...@netspace.org wrote: On Mon, 28 Apr 2014 20:55:18 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: The r-base214 packaging seems to have test suite issues when built against Xcode 5.1 on darwin12… Testing examples for package ‘utils’ /sw/src/fink.build/r-base214-2.14.2-9/R-2.14.2/bin/BATCH: line 60: 34097 Trace/BPT trap: 5 ${R_HOME}/bin/R -f ${in} ${opts} ${R_BATCH_OPTIONS} ${out} 21 Error: testing 'utils' failed Execution halted The failure appears to be due to an unresolved __pcre_valid_utf (or similarly named) symbol in libR.dylib. The R library is trying to use a private symbol in libpcre by guessing what it's called (and what its parameters are) in various different libpcre versions, but our most recent libpcre dropped that symbol altogether. Obviously a fragile situation to rely on undocumented non-public content. One useful change we can make is to patch out the -undefined dynamic_lookup from the configure script. That flag causes undefined symbols to be ignored by the linker, leaving them to cause problems at runtime. By removing the flag, the linking itself fails right away rather than leaving a possibly mis-built library. ..which reveals that util.dylib is missing -llzma in even in r-base30 and r-base31 (ones that do not have the pcre problem). dan -- Daniel Macks dma...@netspace.org - -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free. http://p.sf.net/sfu/SauceLabs - ___ 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 -- Daniel Macks dma...@netspace.org - -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free. http://p.sf.net/sfu/SauceLabs - ___ 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/https://lists.sourceforge.net/lists/listinfo/fink-devel -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get
Re: [Fink-devel] r-base215. r-base-30 and r-base31
Okay, I now see that disables the use of pcre. Guess someone should open a PR upstream, no? Do we really gain anything of use by having R-base build against pcre? On Tuesday, April 29, 2014, Jack Howarth howarth.at.f...@gmail.com wrote: Well, empirically it fixed the r-base214 build on 10.8 for me by just changing --with-system-pcre to --with-pcre=%p. It would be highly irregular for the --with-system-pcre option not to be pushing the headers in /usr/include to be used. On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: (sorry if this email goes out twice!) I don't think that's the meaning of system. R appears to have optional onboard/internal sources of various dependencies as an alternative to using ones existing on the system. *Where* on the system is a different issue. You can probably check the .d files to see exactly which headers are being loaded to see if it's matched. dan On Tue, 29 Apr 2014 13:21:08 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: Isn't this just a header mismatch? We have both a build depends on libpcre1 and --with-system-pcre which is illogical. I am testing with --with-system-pcre changed to --with-pcre=%p. FYI, macports doesn't pass either flag and incorrectly ignores pcre. On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: Using (apparently-)undocumented, internal implementation details that are known to change in different versions is *always* a problem. Any solution that doesn't involve actually not-doing-that is merely deferring the same failure from occurring again in the future when those details change again. dan On Tue, 29 Apr 2014 13:12:14 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: Daniel, Isn't the real problem that r-base in fink is being built with --with-system-pcre but the fink lib pcre.1.dylib ends up linked into libR. I assume we need a BuildConflicts on libpcre1 in order to have the system pcre used.Jack On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: On Mon, 28 Apr 2014 22:38:02 -0400, Daniel Macks dma...@netspace.org wrote: On Mon, 28 Apr 2014 20:55:18 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: The r-base214 packaging seems to have test suite issues when built against Xcode 5.1 on darwin12… Testing examples for package ‘utils’ /sw/src/fink.build/r-base214-2.14.2-9/R-2.14.2/bin/BATCH: line 60: 34097 Trace/BPT trap: 5 ${R_HOME}/bin/R -f ${in} ${opts} ${R_BATCH_OPTIONS} ${out} 21 Error: testing 'utils' failed Execution halted The failure appears to be due to an unresolved __pcre_valid_utf (or similarly named) symbol in libR.dylib. The R library is trying to use a private symbol in libpcre by guessing what it's called (and what its parameters are) in various different libpcre versions, but our most recent libpcre dropped that symbol altogether. Obviously a fragile situation to rely on undocumented non-public content. One useful change we can make is to patch out the -undefined dynamic_lookup from the configure script. That flag causes undefined symbols to be ignored by the linker, leaving them to cause problems at runtime. By removing the flag, the linking itself fails right away rather than leaving a possibly mis-built library. ..which reveals that util.dylib is missing -llzma in even in r-base30 and r-base31 (ones that do not have the pcre problem). dan -- Daniel Macks dma...@netspace.org - -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free. http://p.sf.net/sfu/SauceLabs - ___ 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 -- Daniel Macks dma...@netspace.org - -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free.
Re: [Fink-devel] r-base215. r-base-30 and r-base31
Why would they want a bug report against such an old version? Are newer versions not-broken? I'm pretty sure I already mentioned exactly what change they made in newer versions to avoid trying to access the internals of the external libpcre. dan On Tue, 29 Apr 2014 13:44:34 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: Okay, I now see that disables the use of pcre. Guess someone should open a PR upstream, no? Do we really gain anything of use by having R-base build against pcre? On Tuesday, April 29, 2014, Jack Howarth howarth.at.f...@gmail.com wrote: Well, empirically it fixed the r-base214 build on 10.8 for me by just changing --with-system-pcre to --with-pcre=%p. It would be highly irregular for the --with-system-pcre option not to be pushing the headers in /usr/include to be used. On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: (sorry if this email goes out twice!) I don't think that's the meaning of system. R appears to have optional onboard/internal sources of various dependencies as an alternative to using ones existing on the system. *Where* on the system is a different issue. You can probably check the .d files to see exactly which headers are being loaded to see if it's matched. dan On Tue, 29 Apr 2014 13:21:08 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: Isn't this just a header mismatch? We have both a build depends on libpcre1 and --with-system-pcre which is illogical. I am testing with --with-system-pcre changed to --with-pcre=%p. FYI, macports doesn't pass either flag and incorrectly ignores pcre. On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: Using (apparently-)undocumented, internal implementation details that are known to change in different versions is *always* a problem. Any solution that doesn't involve actually not-doing-that is merely deferring the same failure from occurring again in the future when those details change again. dan On Tue, 29 Apr 2014 13:12:14 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: Daniel, Isn't the real problem that r-base in fink is being built with --with-system-pcre but the fink lib pcre.1.dylib ends up linked into libR. I assume we need a BuildConflicts on libpcre1 in order to have the system pcre used. Jack On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: On Mon, 28 Apr 2014 22:38:02 -0400, Daniel Macks dma...@netspace.org wrote: On Mon, 28 Apr 2014 20:55:18 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: The r-base214 packaging seems to have test suite issues when built against Xcode 5.1 on darwin12… Testing examples for package ‘utils’ /sw/src/fink.build/r-base214-2.14.2-9/R-2.14.2/bin/BATCH: line 60: 34097 Trace/BPT trap: 5 ${R_HOME}/bin/R -f ${in} ${opts} ${R_BATCH_OPTIONS} ${out} 21 Error: testing 'utils' failed Execution halted The failure appears to be due to an unresolved __pcre_valid_utf (or similarly named) symbol in libR.dylib. The R library is trying to use a private symbol in libpcre by guessing what it's called (and what its parameters are) in various different libpcre versions, but our most recent libpcre dropped that symbol altogether. Obviously a fragile situation to rely on undocumented non-public content. One useful change we can make is to patch out the -undefined dynamic_lookup from the configure script. That flag causes undefined symbols to be ignored by the linker, leaving them to cause problems at runtime. By removing the flag, the linking itself fails right away rather than leaving a possibly mis-built library. ..which reveals that util.dylib is missing -llzma in even in r-base30 and r-base31 (ones that do not have the pcre problem). dan -- Daniel Macks dma...@netspace.org - -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free. http://p.sf.net/sfu/SauceLabs - ___ 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 -- Daniel Macks dma...@netspace.org - -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the -
Re: [Fink-devel] r-base215. r-base-30 and r-base31
My mistake. I thought you had said the problem existed in the newer versions but was latent rather than fixed. On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: Why would they want a bug report against such an old version? Are newer versions not-broken? I'm pretty sure I already mentioned exactly what change they made in newer versions to avoid trying to access the internals of the external libpcre. dan On Tue, 29 Apr 2014 13:44:34 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: Okay, I now see that disables the use of pcre. Guess someone should open a PR upstream, no? Do we really gain anything of use by having R-base build against pcre? On Tuesday, April 29, 2014, Jack Howarth howarth.at.f...@gmail.com wrote: Well, empirically it fixed the r-base214 build on 10.8 for me by just changing --with-system-pcre to --with-pcre=%p. It would be highly irregular for the --with-system-pcre option not to be pushing the headers in /usr/include to be used. On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: (sorry if this email goes out twice!) I don't think that's the meaning of system. R appears to have optional onboard/internal sources of various dependencies as an alternative to using ones existing on the system. *Where* on the system is a different issue. You can probably check the .d files to see exactly which headers are being loaded to see if it's matched. dan On Tue, 29 Apr 2014 13:21:08 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: Isn't this just a header mismatch? We have both a build depends on libpcre1 and --with-system-pcre which is illogical. I am testing with --with-system-pcre changed to --with-pcre=%p. FYI, macports doesn't pass either flag and incorrectly ignores pcre. On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: Using (apparently-)undocumented, internal implementation details that are known to change in different versions is *always* a problem. Any solution that doesn't involve actually not-doing-that is merely deferring the same failure from occurring again in the future when those details change again. dan On Tue, 29 Apr 2014 13:12:14 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: Daniel, Isn't the real problem that r-base in fink is being built with --with-system-pcre but the fink lib pcre.1.dylib ends up linked into libR. I assume we need a BuildConflicts on libpcre1 in order to have the system pcre used.Jack On Tuesday, April 29, 2014, Daniel Macks dma...@netspace.org wrote: On Mon, 28 Apr 2014 22:38:02 -0400, Daniel Macks dma...@netspace.org wrote: On Mon, 28 Apr 2014 20:55:18 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: The r-base214 packaging seems to have test suite issues when built against Xcode 5.1 on darwin12… Testing examples for package ‘utils’ /sw/src/fink.build/r-base214-2.14.2-9/R-2.14.2/bin/BATCH: line 60: 34097 Trace/BPT trap: 5 ${R_HOME}/bin/R -f ${in} ${opts} ${R_BATCH_OPTIONS} ${out} 21 Error: testing 'utils' failed Execution halted The failure appears to be due to an unresolved __pcre_valid_utf (or similarly named) symbol in libR.dylib. The R library is trying to use a private symbol in libpcre by guessing what it's called (and what its parameters are) in various different libpcre versions, but our most recent libpcre dropped that symbol altogether. Obviously a fragile situation to rely on undocumented non-public content. One useful change we can make is to patch out the -undefined dynamic_lookup from the configure script. That flag causes undefined symbols to be ignored by the linker, leaving them to cause problems at runtime. By removing the flag, the linking itself fails right away rather than leaving a possibly mis-built library. ..which reveals that util.dylib is missing -llzma in even in r-base30 and r-base31 (ones that do not have the pcre problem). dan -- Daniel Macks dma...@netspace.org - -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free. http://p.sf.net/sfu/SauceLabs - ___ 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 -- Daniel Macks dma...@netspace.org - -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE
[Fink-devel] r-base215. r-base-30 and r-base31
Baba, I have added r-base31 packaging to the 10.7 tree and updated the r-base215 and r-base30 packaging to build with gcc49-compiler like r-base31 and to be aware of the new package. The r-base214 packaging seems to have test suite issues when built against Xcode 5.1 on darwin12… Testing examples for package ‘utils’ /sw/src/fink.build/r-base214-2.14.2-9/R-2.14.2/bin/BATCH: line 60: 34097 Trace/BPT trap: 5 ${R_HOME}/bin/R -f ${in} ${opts} ${R_BATCH_OPTIONS} ${out} 21 Error: testing 'utils' failed Execution halted There are quite a few differences between the r-base214 and r-base215 info files so I am not sure what is the exact cause of the build issue. However, since upstream no longer lists the 2.14.x series as supported on their home page, we should just drop that package like we did with r-base213. Jack ps I also modified the rnmr-r.info file to be aware of the new r-base31 release and it builds fine against it. -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free. http://p.sf.net/sfu/SauceLabs___ 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] r-base215. r-base-30 and r-base31
Jack, 2014/04/29 9:55、Jack Howarth howarth.at.f...@gmail.com のメール: I have added r-base31 packaging to the 10.7 tree and updated the r-base215 and r-base30 packaging to build with gcc49-compiler like r-base31 and to be aware of the new package. The r-base214 packaging seems to have test suite issues when built against Xcode 5.1 on darwin12… Testing examples for package ‘utils’ /sw/src/fink.build/r-base214-2.14.2-9/R-2.14.2/bin/BATCH: line 60: 34097 Trace/BPT trap: 5 ${R_HOME}/bin/R -f ${in} ${opts} ${R_BATCH_OPTIONS} ${out} 21 Error: testing 'utils' failed Execution halted There are quite a few differences between the r-base214 and r-base215 info files so I am not sure what is the exact cause of the build issue. However, since upstream no longer lists the 2.14.x series as supported on their home page, we should just drop that package like we did with r-base213. Jack ps I also modified the rnmr-r.info file to be aware of the new r-base31 release and it builds fine against it. Thanks. I’ve have been too busy to commit r-base31. We can drop 2.14, though some people may still be interested in 2.15. FYI, there are several reports to R-SIG-Mac mailing list that some CRAN packages do not work on R 3.1+Marvericks combination. Maybe because Marvericks uses libc++ instead of libstdc++, but I haven’t looked yet. https://stat.ethz.ch/pipermail/r-sig-mac/2014-April/thread.html — BABA Yoshihiko-- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free. http://p.sf.net/sfu/SauceLabs___ 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] r-base215. r-base-30 and r-base31
On Mon, 28 Apr 2014 20:55:18 -0400, Jack Howarth howarth.at.f...@gmail.com wrote: The r-base214 packaging seems to have test suite issues when built against Xcode 5.1 on darwin12… Testing examples for package ‘utils’ /sw/src/fink.build/r-base214-2.14.2-9/R-2.14.2/bin/BATCH: line 60: 34097 Trace/BPT trap: 5 ${R_HOME}/bin/R -f ${in} ${opts} ${R_BATCH_OPTIONS} ${out} 21 Error: testing 'utils' failed Execution halted The failure appears to be due to an unresolved __pcre_valid_utf (or similarly named) symbol in libR.dylib. The R library is trying to use a private symbol in libpcre by guessing what it's called (and what its parameters are) in various different libpcre versions, but our most recent libpcre dropped that symbol altogether. Obviously a fragile situation to rely on undocumented non-public content. One useful change we can make is to patch out the -undefined dynamic_lookup from the configure script. That flag causes undefined symbols to be ignored by the linker, leaving them to cause problems at runtime. By removing the flag, the linking itself fails right away rather than leaving a possibly mis-built library. There are quite a few differences between the r-base214 and r-base215 info files so I am not sure what is the exact cause of the build issue. However, since upstream no longer lists the 2.14.x series as supported on their home page, we should just drop that package like we did with r-base213. Looks like R now (as of 215) contains the actual internal pcre code it wants (copied from pcre's sources), so it becomes insensitive to changes in libpcre's internals. compare src/main/util.c around line 2159, and see the new valid_utf.h file added to support it. dan -- Daniel Macks dma...@netspace.org -- Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free. http://p.sf.net/sfu/SauceLabs ___ 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