RE: First Qpid 0.6 Beta Release available for download and testing
> Another thing I have noticed is that there is no WCF component > artifact in the beta build, is that something we plan to release with > 0.6? Yes. I contacted Andrew about this yesterday. JIRA coming soon. Cliff -Original Message- From: Robbie Gemmell [mailto:robbie.gemm...@gmail.com] Sent: Thursday, December 10, 2009 6:15 AM To: dev@qpid.apache.org Subject: RE: First Qpid 0.6 Beta Release available for download and testing I have updated the release script to set the SVN version as a property on the Ant command line and modified the ant build scripts as mentioned below. I Also updated some straggling version numbers in the Java tree for the 0.6 release. Another thing I have noticed is that there is no WCF component artifact in the beta build, is that something we plan to release with 0.6? Robbie > -Original Message- > From: Robbie Gemmell [mailto:robbie.gemm...@gmail.com] > Sent: 09 December 2009 22:42 > To: dev@qpid.apache.org > Subject: RE: First Qpid 0.6 Beta Release available for download and > testing > > You are indeed correct. I didn't actually look at the script before > mentioning it as I just presumed it was the Git addition that did it > hehe. Oh well, still needs fixing :) > > I'll look at it tomorrow, shouldn't take long to add a property check. > I'll probably standardise the different areas that doo it too as I > think they do it independently and slightly differently, and I recall > there being an issue whereby if you dont actually have the svnversion > command on your path Ant encounters an infinite loop property > definition and refuses to build a second time ;) > > Robbie > > > -Original Message- > > From: Andrew Stitcher [mailto:astitc...@redhat.com] > > Sent: 09 December 2009 22:24 > > To: dev@qpid.apache.org > > Subject: RE: First Qpid 0.6 Beta Release available for download and > > testing > > > > On Wed, 2009-12-09 at 22:01 +, Robbie Gemmell wrote: > > > No that explains it, it's just the same effect as using git would > > have caused, I just assumed since you added that... :) > > > > > > By using svn export you don't get the .svn/ metadata directories > > within each directory, and so svnversion cant detect the version when > > run and assumes it was exported (which in this case, it actually > was). > > > > > > I will have a look at updating the Ant scripting to allow > specifying > > the revision to handle this. > > > > FWIW in that case this isn't a regression as I believe previous > > releases > > were produced using svn export also. > > > > A > > > > > > > > - > > Apache Qpid - AMQP Messaging Implementation > > Project: http://qpid.apache.org > > Use/Interact: mailto:dev-subscr...@qpid.apache.org > - Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:dev-subscr...@qpid.apache.org
Re: First Qpid 0.6 Beta Release available for download and testing
On 12/02/2009 07:53 PM, Andrew Stitcher wrote: Apologies for the long message, but there actually is a fair bit to say: I've uploaded the artifacts for a 0.6 beta release to qpid.apache.org: http://qpid.apache.org/dist/qpid-0.6beta1/ Please download and test this release. No, really, please download and at least look at it. Kicked the tires on RHEL5, no problem with the following: - automake built & tested the c++ subdir of the big source package tarball. - automake built, tested & installed cpp distro - cmake built & tested cpp distro (after copying tests/cluster.cmake to work-around the known & fixed cmake build problem) - ran a cluster from the install - built the installed examples, ran direct example against cluster. - Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:dev-subscr...@qpid.apache.org
RE: First Qpid 0.6 Beta Release available for download and testing
I have updated the release script to set the SVN version as a property on the Ant command line and modified the ant build scripts as mentioned below. I Also updated some straggling version numbers in the Java tree for the 0.6 release. Another thing I have noticed is that there is no WCF component artifact in the beta build, is that something we plan to release with 0.6? Robbie > -Original Message- > From: Robbie Gemmell [mailto:robbie.gemm...@gmail.com] > Sent: 09 December 2009 22:42 > To: dev@qpid.apache.org > Subject: RE: First Qpid 0.6 Beta Release available for download and > testing > > You are indeed correct. I didn't actually look at the script before > mentioning it as I just presumed it was the Git addition that did it > hehe. Oh well, still needs fixing :) > > I'll look at it tomorrow, shouldn't take long to add a property check. > I'll probably standardise the different areas that doo it too as I > think they do it independently and slightly differently, and I recall > there being an issue whereby if you dont actually have the svnversion > command on your path Ant encounters an infinite loop property > definition and refuses to build a second time ;) > > Robbie > > > -Original Message- > > From: Andrew Stitcher [mailto:astitc...@redhat.com] > > Sent: 09 December 2009 22:24 > > To: dev@qpid.apache.org > > Subject: RE: First Qpid 0.6 Beta Release available for download and > > testing > > > > On Wed, 2009-12-09 at 22:01 +, Robbie Gemmell wrote: > > > No that explains it, it's just the same effect as using git would > > have caused, I just assumed since you added that... :) > > > > > > By using svn export you don't get the .svn/ metadata directories > > within each directory, and so svnversion cant detect the version when > > run and assumes it was exported (which in this case, it actually > was). > > > > > > I will have a look at updating the Ant scripting to allow > specifying > > the revision to handle this. > > > > FWIW in that case this isn't a regression as I believe previous > > releases > > were produced using svn export also. > > > > A > > > > > > > > - > > Apache Qpid - AMQP Messaging Implementation > > Project: http://qpid.apache.org > > Use/Interact: mailto:dev-subscr...@qpid.apache.org > - Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:dev-subscr...@qpid.apache.org
RE: First Qpid 0.6 Beta Release available for download and testing
You are indeed correct. I didn't actually look at the script before mentioning it as I just presumed it was the Git addition that did it hehe. Oh well, still needs fixing :) I'll look at it tomorrow, shouldn't take long to add a property check. I'll probably standardise the different areas that doo it too as I think they do it independently and slightly differently, and I recall there being an issue whereby if you dont actually have the svnversion command on your path Ant encounters an infinite loop property definition and refuses to build a second time ;) Robbie > -Original Message- > From: Andrew Stitcher [mailto:astitc...@redhat.com] > Sent: 09 December 2009 22:24 > To: dev@qpid.apache.org > Subject: RE: First Qpid 0.6 Beta Release available for download and > testing > > On Wed, 2009-12-09 at 22:01 +, Robbie Gemmell wrote: > > No that explains it, it's just the same effect as using git would > have caused, I just assumed since you added that... :) > > > > By using svn export you don't get the .svn/ metadata directories > within each directory, and so svnversion cant detect the version when > run and assumes it was exported (which in this case, it actually was). > > > > I will have a look at updating the Ant scripting to allow specifying > the revision to handle this. > > FWIW in that case this isn't a regression as I believe previous > releases > were produced using svn export also. > > A > > > > - > Apache Qpid - AMQP Messaging Implementation > Project: http://qpid.apache.org > Use/Interact: mailto:dev-subscr...@qpid.apache.org - Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:dev-subscr...@qpid.apache.org
RE: First Qpid 0.6 Beta Release available for download and testing
On Wed, 2009-12-09 at 22:01 +, Robbie Gemmell wrote: > No that explains it, it's just the same effect as using git would have > caused, I just assumed since you added that... :) > > By using svn export you don't get the .svn/ metadata directories within each > directory, and so svnversion cant detect the version when run and assumes it > was exported (which in this case, it actually was). > > I will have a look at updating the Ant scripting to allow specifying the > revision to handle this. FWIW in that case this isn't a regression as I believe previous releases were produced using svn export also. A - Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:dev-subscr...@qpid.apache.org
RE: First Qpid 0.6 Beta Release available for download and testing
No that explains it, it's just the same effect as using git would have caused, I just assumed since you added that... :) By using svn export you don't get the .svn/ metadata directories within each directory, and so svnversion cant detect the version when run and assumes it was exported (which in this case, it actually was). I will have a look at updating the Ant scripting to allow specifying the revision to handle this. Robbie > -Original Message- > From: Andrew Stitcher [mailto:astitc...@redhat.com] > Sent: 09 December 2009 21:55 > To: dev@qpid.apache.org > Subject: RE: First Qpid 0.6 Beta Release available for download and > testing > > On Wed, 2009-12-09 at 20:36 +, Robbie Gemmell wrote: > > I have updated (r888939) the build process to and release script so > the QMan package is generated and copied over. > > > > Additionally, I have noticed that because you are using Git > (presumably, since the ability was added recently) to do the checkout > the Java modules are failing to identify the SVN revision used in the > build and so are generating property files that contain "exported" > where the revision should be, since svnversion didnt find the .svn > folders. As a result, "exported" will show up wherever the revision > would normally be mentioned, eg broker startup. > > > > The actual released artifacts were produced from an svn export, not git > - that was for convenience leading up to a final release as "svn > export" > is slow because it needs to go over the internet. > > So that doesn't explain this (unless I've made a mistake). > > Andrew > > > One way to handle this would be to add a property in the ant build > process that can be set via command line in the release script with the > value parsed from Git, and modify the build process to only use > svnversion to figure it out the revision if that hasn't been set. > > That's certainly not in my skill set re java, I'm just about capable of > running release.sh and seeing whether java built! > > Andrew > > > > - > Apache Qpid - AMQP Messaging Implementation > Project: http://qpid.apache.org > Use/Interact: mailto:dev-subscr...@qpid.apache.org - Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:dev-subscr...@qpid.apache.org
RE: First Qpid 0.6 Beta Release available for download and testing
On Wed, 2009-12-09 at 20:36 +, Robbie Gemmell wrote: > I have updated (r888939) the build process to and release script so the QMan > package is generated and copied over. > > Additionally, I have noticed that because you are using Git (presumably, > since the ability was added recently) to do the checkout the Java modules are > failing to identify the SVN revision used in the build and so are generating > property files that contain "exported" where the revision should be, since > svnversion didnt find the .svn folders. As a result, "exported" will show up > wherever the revision would normally be mentioned, eg broker startup. > The actual released artifacts were produced from an svn export, not git - that was for convenience leading up to a final release as "svn export" is slow because it needs to go over the internet. So that doesn't explain this (unless I've made a mistake). Andrew > One way to handle this would be to add a property in the ant build process > that can be set via command line in the release script with the value parsed > from Git, and modify the build process to only use svnversion to figure it > out the revision if that hasn't been set. That's certainly not in my skill set re java, I'm just about capable of running release.sh and seeing whether java built! Andrew - Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:dev-subscr...@qpid.apache.org
RE: First Qpid 0.6 Beta Release available for download and testing
I have updated (r888939) the build process to and release script so the QMan package is generated and copied over. Additionally, I have noticed that because you are using Git (presumably, since the ability was added recently) to do the checkout the Java modules are failing to identify the SVN revision used in the build and so are generating property files that contain "exported" where the revision should be, since svnversion didnt find the .svn folders. As a result, "exported" will show up wherever the revision would normally be mentioned, eg broker startup. One way to handle this would be to add a property in the ant build process that can be set via command line in the release script with the value parsed from Git, and modify the build process to only use svnversion to figure it out the revision if that hasn't been set. Robbie < 2nd attempt at send as I didn't notice the first time that it only replied directly to me ;-) > > -Original Message- > From: Robbie Gemmell [mailto:robbie.gemm...@gmail.com] > Sent: 09 December 2009 17:10 > To: dev@qpid.apache.org > Subject: RE: First Qpid 0.6 Beta Release available for download and > testing > > The single component package for QMan is not currently being created > (although I seem to recall this happening with 0.5 as well). I think it > uses a slightly non standard target to build the package just now, so > ill see if I can modify it to be the same as the others and then the > release script just needs updated to perform the copy. > > Robbie > > > -Original Message- > > From: Andrew Stitcher [mailto:astitc...@redhat.com] > > Sent: 03 December 2009 00:54 > > To: dev@qpid.apache.org > > Subject: First Qpid 0.6 Beta Release available for download and > testing > > > > Apologies for the long message, but there actually is a fair bit to > > say: > > > > I've uploaded the artifacts for a 0.6 beta release to > qpid.apache.org: > > > > http://qpid.apache.org/dist/qpid-0.6beta1/ > > > > Please download and test this release. No, really, please download > and > > at least look at it. > > > > I've called it a beta because, it seems to me in the area that I know > > (C > > ++) and in the currently outstanding bugs it seems very usable. > > > > I plan that every one should give this a good beating for 1-2 weeks > and > > then I'll spin a release candidate, or if there are many problems > found > > and fixed a beta2. > > > > So all being well I'd like to vote on releasing 0.6 in a couple of > > weeks. > > > > --- > > > > The status of this beta: > > > > My testing has been building the C++ tar source under various linux > > platforms (using this release or very slightly earlier ones). > > Essentially "configure; make check". > > > > Fedora 11/Fedora 12 (these both tested with the same results): > > > > Autotools build is fine including all broker modules. It doesn't seem > > that the build as it stands in the C++ source release builds the qmf > > bindings. > > > > Cmake build is mostly fine - however the clustering module doesn't > > build > > due to a complication of recent versions of boost. And r886259 breaks > > building the clustering tests (due to missing the new > > tests/cluster.cmake from EXTRA_DIST) > > > > Red Hat Enterprise Linux 5: > > > > Autotools builds fine. > > > > Cmake builds fine (but this was a slightly earlier release and this > > would have been broken by r886259 as above). > > > > Debian 5 (lenny). > > > > Neither the clustering module nor the XML module have their necessary > > prerequisites so can't be built here. > > > > Given this both autotools and cmake builds were fine. > > > > --- > > > > Overall Release Status: > > > > I've put up a page for the release on the wiki: > > > > http://cwiki.apache.org/confluence/display/qpid/0.6+Release > > > > (don't use the mirror on qpid.apache.org as it won't have the > > dynamically updated bug list) > > > > Please edit the release page to add your understanding to it - it's a > > little biased to the areas that I know about presently. > > > > One area that needs attention is the Release notes, I appeal to you > to > > add any release notes to that wiki page and I will turn them into > > something to put in the release itself. Otherwise the only release > > notes > > will be for the C++ code and whatever I can get from the 0.6 fixed >
RE: First Qpid 0.6 Beta Release available for download and testing
The single component package for QMan is not currently being created (although I seem to recall this happening with 0.5 as well). I think it uses a slightly non standard target to build the package just now, so ill see if I can modify it to be the same as the others and then the release script just needs updated to perform the copy. Robbie > -Original Message- > From: Andrew Stitcher [mailto:astitc...@redhat.com] > Sent: 03 December 2009 00:54 > To: dev@qpid.apache.org > Subject: First Qpid 0.6 Beta Release available for download and testing > > Apologies for the long message, but there actually is a fair bit to > say: > > I've uploaded the artifacts for a 0.6 beta release to qpid.apache.org: > > http://qpid.apache.org/dist/qpid-0.6beta1/ > > Please download and test this release. No, really, please download and > at least look at it. > > I've called it a beta because, it seems to me in the area that I know > (C > ++) and in the currently outstanding bugs it seems very usable. > > I plan that every one should give this a good beating for 1-2 weeks and > then I'll spin a release candidate, or if there are many problems found > and fixed a beta2. > > So all being well I'd like to vote on releasing 0.6 in a couple of > weeks. > > --- > > The status of this beta: > > My testing has been building the C++ tar source under various linux > platforms (using this release or very slightly earlier ones). > Essentially "configure; make check". > > Fedora 11/Fedora 12 (these both tested with the same results): > > Autotools build is fine including all broker modules. It doesn't seem > that the build as it stands in the C++ source release builds the qmf > bindings. > > Cmake build is mostly fine - however the clustering module doesn't > build > due to a complication of recent versions of boost. And r886259 breaks > building the clustering tests (due to missing the new > tests/cluster.cmake from EXTRA_DIST) > > Red Hat Enterprise Linux 5: > > Autotools builds fine. > > Cmake builds fine (but this was a slightly earlier release and this > would have been broken by r886259 as above). > > Debian 5 (lenny). > > Neither the clustering module nor the XML module have their necessary > prerequisites so can't be built here. > > Given this both autotools and cmake builds were fine. > > --- > > Overall Release Status: > > I've put up a page for the release on the wiki: > > http://cwiki.apache.org/confluence/display/qpid/0.6+Release > > (don't use the mirror on qpid.apache.org as it won't have the > dynamically updated bug list) > > Please edit the release page to add your understanding to it - it's a > little biased to the areas that I know about presently. > > One area that needs attention is the Release notes, I appeal to you to > add any release notes to that wiki page and I will turn them into > something to put in the release itself. Otherwise the only release > notes > will be for the C++ code and whatever I can get from the 0.6 fixed > bugs. > > --- > > Deficiencies: > > I've removed the separate dotnet package from the release as it doesn't > build for me using the release script. More about this separately. > > I've not packaged the wcf code separately as I've no idea how it should > look and I doubt I could build it under Linux anyway. > > I'm concerned there appear to be 5 13Mb eclipse plugins labelled for > various platforms - is this really necessary - it certainly took a long > time to upload and constitutes a large amount of space. > > --- > > At this point I suggest that any bugs important enough to be fixed > before the 0.6 release should be tagged with "Fix version" 0.6 and be a > blocker or critical. > > So if you find an issue that must be fixed before release then put it > in > Jira like this. > > Conversely if you know of a bug that is in this state that can be > resolved then resolve it poste haste. > > I will assume that as long as there *any* bugs like this 0.6 release is > blocked. > > Regards > > Andrew > > > - > Apache Qpid - AMQP Messaging Implementation > Project: http://qpid.apache.org > Use/Interact: mailto:dev-subscr...@qpid.apache.org - Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:dev-subscr...@qpid.apache.org
Re: First Qpid 0.6 Beta Release available for download and testing
On 12/07/2009 11:04 AM, Andrew Stitcher wrote: On Mon, 2009-12-07 at 10:28 -0500, Alan Conway wrote: On 12/04/2009 12:20 PM, Andrew Stitcher wrote: On Fri, 2009-12-04 at 11:07 -0500, Alan Conway wrote: ...On 12/02/2009 07:53 PM, Andrew Stitcher wrote: Cmake build is mostly fine - however the clustering module doesn't build due to a complication of recent versions of boost. The cluster& cluster test build& pass for me on f11 with boost-1.37.0-7.fc11.x86_64, which is the yum latest. Is there a machine I can log into where you see the problem? In response - I don't understand how you can be building this! The StoreStatus.o has references to symbols that are only in libboost_system but it's not directly linked. However having said this I also can't understand why the autotools build does work by the same token - it's a bit of a mystery! ldd shows that in my build cluster.so is indeed linked with boost_system-mt. [acon...@rolf cmake]$ ldd src/cluster.so | grep boost libboost_filesystem-mt.so.4 => /usr/lib64/libboost_filesystem-mt.so.4 (0x7f1d97f67000) libboost_program_options-mt.so.4 => /usr/lib64/libboost_program_options-mt.so.4 (0x7f1d978ac000) libboost_system-mt.so.4 => /usr/lib64/libboost_system-mt.so.4 (0x7f1d95f44000) Also note that libcommon uses boost::filesystem and links without problems, although perhaps StoreStatus is using different symbols. What missing symbols are you seeing? Ok, I haven't explained in enough detail: libqpidcommon isn't a problem because by default shared objects are allowed to have unresolved symbols at link time on the Unix systems I know - if they weren't resolved at the run time load you'd get an error - generally sigkill to the process. ldd doesn't tell you what you need to know here as it gives the transitive linkage - ie in this case libboost_system is loaded because libboost_filesystem requires it. To get the actual (non transitive) requires of an object you need to use objdump -p. When we build plugin modules for qpid we tell the linker to give an error if it finds unresolved symbols when linking against the non transitive requirements (--no-undefined to the linker -Wl,--no-undefined to the compiler). It's possible that libtool adds another option to make it check the symbols transitively, which would explain why the autotools build works, but I can't see why your cmake build could be different from mine. My cmake build does not have those flags in the link line for cluster.so: /usr/bin/ccache /usr/bin/distcc g++ -fPIC -g -O -shared -Wl,-soname,cluster.so -o cluster.so CMakeFiles/cluster.dir/qpid/cluster/Quorum_cman.o CMakeFiles/cluster.dir/qpid/cluster/Cluster.o CMakeFiles/cluster.dir/qpid/cluster/Decoder.o CMakeFiles/cluster.dir/qpid/cluster/ClusterMap.o CMakeFiles/cluster.dir/qpid/cluster/ClusterPlugin.o CMakeFiles/cluster.dir/qpid/cluster/Connection.o CMakeFiles/cluster.dir/qpid/cluster/ConnectionCodec.o CMakeFiles/cluster.dir/qpid/cluster/Cpg.o CMakeFiles/cluster.dir/qpid/cluster/UpdateClient.o CMakeFiles/cluster.dir/qpid/cluster/RetractClient.o CMakeFiles/cluster.dir/qpid/cluster/ErrorCheck.o CMakeFiles/cluster.dir/qpid/cluster/Event.o CMakeFiles/cluster.dir/qpid/cluster/EventFrame.o CMakeFiles/cluster.dir/qpid/cluster/ExpiryPolicy.o CMakeFiles/cluster.dir/qpid/cluster/FailoverExchange.o CMakeFiles/cluster.dir/qpid/cluster/Multicaster.o CMakeFiles/cluster.dir/qpid/cluster/OutputInterceptor.o CMakeFiles/cluster.dir/qpid/cluster/PollerDispatch.o CMakeFiles/cluster.dir/qpid/cluster/InitialStatusMap.o CMakeFiles/cluster.dir/qpid/cluster/MemberSet.o CMakeFiles/cluster.dir/qpid/cluster/StoreStatus.o -lcpg -lcman libqpidbroker.so.0.6 libqpidclient.so.0.6 -lboost_filesystem-mt libqpidcommon.so.0.6 -lboost_program_options-mt -lboost_filesystem-mt -luuid -ldl -lrt -lsasl2 -Wl,-rpath,/home/aconway/qpid/qpid/cmake/src I don't know why those flags aren't present, maybe something wrong with the CMAKE_COMPILER_IS_GNUCXX that's causing it to fail on my box? How can I debug what cmake is doing here? The symbols that are unresolved in all of the shared objects that use boost_filesystem due to inline expansions of things from it are: U boost::system::get_system_category() U boost::system::get_generic_category() Yep, they're undefined for me as well but not causing a compile error due to the missing flags. I'm still not clear on what the solution is here: - drop the --no-undefined-flags - explicitly link with boost system if present We could drop the --no-undefined flag for cluster.so only as a workaround if linking with boost system is tricky. - Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:dev-subscr...@qpid.apache.org
Re: First Qpid 0.6 Beta Release available for download and testing
On Mon, 2009-12-07 at 10:28 -0500, Alan Conway wrote: > On 12/04/2009 12:20 PM, Andrew Stitcher wrote: > > On Fri, 2009-12-04 at 11:07 -0500, Alan Conway wrote: > >> ...On 12/02/2009 07:53 PM, Andrew Stitcher wrote: > >> > >>> Cmake build is mostly fine - however the clustering module doesn't build > >>> due to a complication of recent versions of boost. > >> > >> The cluster& cluster test build& pass for me on f11 with > >> boost-1.37.0-7.fc11.x86_64, which is the yum latest. Is there a machine I > >> can > >> log into where you see the problem? > > > > In response - I don't understand how you can be building this! The > > StoreStatus.o has references to symbols that are only in libboost_system > > but it's not directly linked. However having said this I also can't > > understand why the autotools build does work by the same token - it's a > > bit of a mystery! > > > > ldd shows that in my build cluster.so is indeed linked with boost_system-mt. > > [acon...@rolf cmake]$ ldd src/cluster.so | grep boost > libboost_filesystem-mt.so.4 => /usr/lib64/libboost_filesystem-mt.so.4 > (0x7f1d97f67000) > libboost_program_options-mt.so.4 => > /usr/lib64/libboost_program_options-mt.so.4 (0x7f1d978ac000) > libboost_system-mt.so.4 => /usr/lib64/libboost_system-mt.so.4 > (0x7f1d95f44000) > > Also note that libcommon uses boost::filesystem and links without problems, > although perhaps StoreStatus is using different symbols. What missing symbols > are you seeing? Ok, I haven't explained in enough detail: libqpidcommon isn't a problem because by default shared objects are allowed to have unresolved symbols at link time on the Unix systems I know - if they weren't resolved at the run time load you'd get an error - generally sigkill to the process. ldd doesn't tell you what you need to know here as it gives the transitive linkage - ie in this case libboost_system is loaded because libboost_filesystem requires it. To get the actual (non transitive) requires of an object you need to use objdump -p. When we build plugin modules for qpid we tell the linker to give an error if it finds unresolved symbols when linking against the non transitive requirements (--no-undefined to the linker -Wl,--no-undefined to the compiler). It's possible that libtool adds another option to make it check the symbols transitively, which would explain why the autotools build works, but I can't see why your cmake build could be different from mine. The symbols that are unresolved in all of the shared objects that use boost_filesystem due to inline expansions of things from it are: U boost::system::get_system_category() U boost::system::get_generic_category() Regards Andrew > > - > Apache Qpid - AMQP Messaging Implementation > Project: http://qpid.apache.org > Use/Interact: mailto:dev-subscr...@qpid.apache.org > - Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:dev-subscr...@qpid.apache.org
Re: First Qpid 0.6 Beta Release available for download and testing
On 12/04/2009 12:20 PM, Andrew Stitcher wrote: On Fri, 2009-12-04 at 11:07 -0500, Alan Conway wrote: ...On 12/02/2009 07:53 PM, Andrew Stitcher wrote: Cmake build is mostly fine - however the clustering module doesn't build due to a complication of recent versions of boost. The cluster& cluster test build& pass for me on f11 with boost-1.37.0-7.fc11.x86_64, which is the yum latest. Is there a machine I can log into where you see the problem? In response - I don't understand how you can be building this! The StoreStatus.o has references to symbols that are only in libboost_system but it's not directly linked. However having said this I also can't understand why the autotools build does work by the same token - it's a bit of a mystery! ldd shows that in my build cluster.so is indeed linked with boost_system-mt. [acon...@rolf cmake]$ ldd src/cluster.so | grep boost libboost_filesystem-mt.so.4 => /usr/lib64/libboost_filesystem-mt.so.4 (0x7f1d97f67000) libboost_program_options-mt.so.4 => /usr/lib64/libboost_program_options-mt.so.4 (0x7f1d978ac000) libboost_system-mt.so.4 => /usr/lib64/libboost_system-mt.so.4 (0x7f1d95f44000) Also note that libcommon uses boost::filesystem and links without problems, although perhaps StoreStatus is using different symbols. What missing symbols are you seeing? - Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:dev-subscr...@qpid.apache.org
Re: First Qpid 0.6 Beta Release available for download and testing
Hi Andrew, Thanks for doing this ! I'll take a look/test shortly. On the release notes front, afaik for the java that should be generated from JIRA as this is pretty much accurate (for all completed items). Hth, Marnie On Thu, Dec 3, 2009 at 12:53 AM, Andrew Stitcher wrote: > Apologies for the long message, but there actually is a fair bit to say: > > I've uploaded the artifacts for a 0.6 beta release to qpid.apache.org: > > http://qpid.apache.org/dist/qpid-0.6beta1/ > > Please download and test this release. No, really, please download and > at least look at it. > > I've called it a beta because, it seems to me in the area that I know (C > ++) and in the currently outstanding bugs it seems very usable. > > I plan that every one should give this a good beating for 1-2 weeks and > then I'll spin a release candidate, or if there are many problems found > and fixed a beta2. > > So all being well I'd like to vote on releasing 0.6 in a couple of > weeks. > > --- > > The status of this beta: > > My testing has been building the C++ tar source under various linux > platforms (using this release or very slightly earlier ones). > Essentially "configure; make check". > > Fedora 11/Fedora 12 (these both tested with the same results): > > Autotools build is fine including all broker modules. It doesn't seem > that the build as it stands in the C++ source release builds the qmf > bindings. > > Cmake build is mostly fine - however the clustering module doesn't build > due to a complication of recent versions of boost. And r886259 breaks > building the clustering tests (due to missing the new > tests/cluster.cmake from EXTRA_DIST) > > Red Hat Enterprise Linux 5: > > Autotools builds fine. > > Cmake builds fine (but this was a slightly earlier release and this > would have been broken by r886259 as above). > > Debian 5 (lenny). > > Neither the clustering module nor the XML module have their necessary > prerequisites so can't be built here. > > Given this both autotools and cmake builds were fine. > > --- > > Overall Release Status: > > I've put up a page for the release on the wiki: > > http://cwiki.apache.org/confluence/display/qpid/0.6+Release > > (don't use the mirror on qpid.apache.org as it won't have the > dynamically updated bug list) > > Please edit the release page to add your understanding to it - it's a > little biased to the areas that I know about presently. > > One area that needs attention is the Release notes, I appeal to you to > add any release notes to that wiki page and I will turn them into > something to put in the release itself. Otherwise the only release notes > will be for the C++ code and whatever I can get from the 0.6 fixed bugs. > > --- > > Deficiencies: > > I've removed the separate dotnet package from the release as it doesn't > build for me using the release script. More about this separately. > > I've not packaged the wcf code separately as I've no idea how it should > look and I doubt I could build it under Linux anyway. > > I'm concerned there appear to be 5 13Mb eclipse plugins labelled for > various platforms - is this really necessary - it certainly took a long > time to upload and constitutes a large amount of space. > > --- > > At this point I suggest that any bugs important enough to be fixed > before the 0.6 release should be tagged with "Fix version" 0.6 and be a > blocker or critical. > > So if you find an issue that must be fixed before release then put it in > Jira like this. > > Conversely if you know of a bug that is in this state that can be > resolved then resolve it poste haste. > > I will assume that as long as there *any* bugs like this 0.6 release is > blocked. > > Regards > > Andrew > > > - > Apache Qpid - AMQP Messaging Implementation > Project: http://qpid.apache.org > Use/Interact: mailto:dev-subscr...@qpid.apache.org > >
Re: First Qpid 0.6 Beta Release available for download and testing
On Fri, 2009-12-04 at 11:07 -0500, Alan Conway wrote: > ...On 12/02/2009 07:53 PM, Andrew Stitcher wrote: > > > Cmake build is mostly fine - however the clustering module doesn't build > > due to a complication of recent versions of boost. > > The cluster & cluster test build & pass for me on f11 with > boost-1.37.0-7.fc11.x86_64, which is the yum latest. Is there a machine I can > log into where you see the problem? In response - I don't understand how you can be building this! The StoreStatus.o has references to symbols that are only in libboost_system but it's not directly linked. However having said this I also can't understand why the autotools build does work by the same token - it's a bit of a mystery! > > > And r886259 breaks > > building the clustering tests (due to missing the new > > tests/cluster.cmake from EXTRA_DIST) > > I think you already fixed this in 886779 - is there still an issue to resolve? Yes fixed. A - Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:dev-subscr...@qpid.apache.org
RE: First Qpid 0.6 Beta Release available for download and testing
On Fri, 2009-12-04 at 11:53 -0500, Steve Huston wrote: > > -Original Message- > > From: Alan Conway [mailto:acon...@redhat.com] > > Sent: Friday, December 04, 2009 11:07 AM > > To: dev@qpid.apache.org; Andrew Stitcher > > Subject: Re: First Qpid 0.6 Beta Release available for > > download and testing > > > > > > On 12/02/2009 07:53 PM, Andrew Stitcher wrote: > > > > > Cmake build is mostly fine - however the clustering module > > doesn't build > > > due to a complication of recent versions of boost. > > > > The cluster & cluster test build & pass for me on f11 with > > boost-1.37.0-7.fc11.x86_64, which is the yum latest. Is there > > a machine I can > > log into where you see the problem? > > It's probably this: > https://issues.apache.org/jira/browse/QPID-2202 It's related to this in that it's symbols from boost::filesystem which aren't found. However the details of Unix and Windows linking are different enough that the bugs can't be related (essentially because the dll linking under windows is driven from the header files, but that isn't possible under Unix). Andrew - Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:dev-subscr...@qpid.apache.org
RE: First Qpid 0.6 Beta Release available for download and testing
> -Original Message- > From: Alan Conway [mailto:acon...@redhat.com] > Sent: Friday, December 04, 2009 11:07 AM > To: dev@qpid.apache.org; Andrew Stitcher > Subject: Re: First Qpid 0.6 Beta Release available for > download and testing > > > On 12/02/2009 07:53 PM, Andrew Stitcher wrote: > > > Cmake build is mostly fine - however the clustering module > doesn't build > > due to a complication of recent versions of boost. > > The cluster & cluster test build & pass for me on f11 with > boost-1.37.0-7.fc11.x86_64, which is the yum latest. Is there > a machine I can > log into where you see the problem? It's probably this: https://issues.apache.org/jira/browse/QPID-2202 -Steve - Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:dev-subscr...@qpid.apache.org
Re: First Qpid 0.6 Beta Release available for download and testing
On 12/02/2009 07:53 PM, Andrew Stitcher wrote: Cmake build is mostly fine - however the clustering module doesn't build due to a complication of recent versions of boost. The cluster & cluster test build & pass for me on f11 with boost-1.37.0-7.fc11.x86_64, which is the yum latest. Is there a machine I can log into where you see the problem? > And r886259 breaks > building the clustering tests (due to missing the new > tests/cluster.cmake from EXTRA_DIST) I think you already fixed this in 886779 - is there still an issue to resolve? - Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:dev-subscr...@qpid.apache.org
RE: First Qpid 0.6 Beta Release available for download and testing
Andrew Stitcher wrote: > I've not packaged the wcf code separately as I've no idea how it > should look and I doubt I could build it under Linux anyway. I will take a look at it ASAP and get you what you need. Cliff
Re: First Qpid 0.6 Beta Release available for download and testing
Hi Andrew, I've updated the 0.6 Relese page with another JIRA filter showing all other unresolved JIRAs tagged Fix For : 0.6. These should be resolved or migrated to Affects 0.6 and the fix for cleared. http://cwiki.apache.org/confluence/display/qpid/0.6+Release Martin 2009/12/3 Robbie Gemmell : > The reason there are multiple 'eclipse-plugin' artifacts is that it > isnt actually just an Eclipse plugin, but actually a full Eclipse RCP > application and the launcher and SWT code for these is platform > specific. The artifacts are generated from the > management/eclipse-plugin module, which is just an Eclipse RCP plugin, > hence its name. > > I know this has also confused for a couple of users in the past, so I > would be in favour of updating the release artifact names, Eg: > qpid-jmx-management-console-0.6-linux-gtk-x86.tar.gz instead of > qpid-management-eclipse-plugin-0.6-linux-gtk-x86.tar.gz > > What do people think? > > Robbie > > 2009/12/3 Andrew Stitcher : > > > >> Deficiencies: >> >> I've removed the separate dotnet package from the release as it doesn't >> build for me using the release script. More about this separately. >> >> I've not packaged the wcf code separately as I've no idea how it should >> look and I doubt I could build it under Linux anyway. >> >> I'm concerned there appear to be 5 13Mb eclipse plugins labelled for >> various platforms - is this really necessary - it certainly took a long >> time to upload and constitutes a large amount of space. >> > > > > - > Apache Qpid - AMQP Messaging Implementation > Project: http://qpid.apache.org > Use/Interact: mailto:dev-subscr...@qpid.apache.org > > -- Martin Ritchie - Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:dev-subscr...@qpid.apache.org
Re: First Qpid 0.6 Beta Release available for download and testing
The reason there are multiple 'eclipse-plugin' artifacts is that it isnt actually just an Eclipse plugin, but actually a full Eclipse RCP application and the launcher and SWT code for these is platform specific. The artifacts are generated from the management/eclipse-plugin module, which is just an Eclipse RCP plugin, hence its name. I know this has also confused for a couple of users in the past, so I would be in favour of updating the release artifact names, Eg: qpid-jmx-management-console-0.6-linux-gtk-x86.tar.gz instead of qpid-management-eclipse-plugin-0.6-linux-gtk-x86.tar.gz What do people think? Robbie 2009/12/3 Andrew Stitcher : > Deficiencies: > > I've removed the separate dotnet package from the release as it doesn't > build for me using the release script. More about this separately. > > I've not packaged the wcf code separately as I've no idea how it should > look and I doubt I could build it under Linux anyway. > > I'm concerned there appear to be 5 13Mb eclipse plugins labelled for > various platforms - is this really necessary - it certainly took a long > time to upload and constitutes a large amount of space. > - Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:dev-subscr...@qpid.apache.org
