Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
Hi Niels, Bob, others, On Fri, Sep 25, 2015 at 8:34 AM, Niels Thykierwrote: > Thanks for getting it uploaded to NEW. > > I have prodded the FTP masters about fast tracking and will try to > finish the resulting transition as fast as possible. :) Now it's accepted, built fine on almost all architectures and the tracker seems to be correct. It fails on mipsel[1] with: -- cut -- Magick: abort due to signal 11 (SIGSEGV) "Segmentation Fault"... /«PKGBUILDDIR»/scripts/tap-functions.shi: line 58: 2290 Aborted ( "$@" ) EXEC: ./rwfile -stdio -filespec out_truecolor_stdio_%d /«PKGBUILDDIR»/tests/input_truecolor.miff PCDS not ok 244 - PCDS truecolor (stdio) FAIL: tests/rwfile.tap 244 - PCDS truecolor (stdio) -- cut -- It was failed previously a few times as well, but a rescheduled build solved it. To be honest the reasons were: signal 01 (SIGBUS) "Bus Error" But I suspect the cause may be the same, please reschedule the build of graphicsmagick on mipsel. The build is strange on mips just for the record. Sometimes (just like in the past) it builds in a hour. Nowadays it's over twelve hours sometimes, but it seems to be at lease six hours[2]. Bob, how may I help you to find the root cause of the slow building on mips at least? Thanks, Laszlo/GCS [1] https://buildd.debian.org/status/fetch.php?pkg=graphicsmagick=mipsel=1.3.21-4=1443407301 [2] https://buildd.debian.org/status/logs.php?pkg=graphicsmagick=mips
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
On 2015-09-28 08:46, László Böszörményi (GCS) wrote: > Hi Niels, Bob, others, Hi, > On Fri, Sep 25, 2015 at 8:34 AM, Niels Thykierwrote: > > Thanks for getting it uploaded to NEW. > > > > I have prodded the FTP masters about fast tracking and will try to > > finish the resulting transition as fast as possible. :) > Now it's accepted, built fine on almost all architectures and the > tracker seems to be correct. > It fails on mipsel[1] with: > -- cut -- > Magick: abort due to signal 11 (SIGSEGV) "Segmentation Fault"... > /«PKGBUILDDIR»/scripts/tap-functions.shi: line 58: 2290 Aborted > ( "$@" ) > EXEC: ./rwfile -stdio -filespec out_truecolor_stdio_%d > /«PKGBUILDDIR»/tests/input_truecolor.miff PCDS > not ok 244 - PCDS truecolor (stdio) > FAIL: tests/rwfile.tap 244 - PCDS truecolor (stdio) > -- cut -- > > It was failed previously a few times as well, but a rescheduled build > solved it. To be honest the reasons were: signal 01 (SIGBUS) "Bus > Error" > But I suspect the cause may be the same, please reschedule the build > of graphicsmagick on mipsel. The failure is strange, but indeed the new build was successful. Looks like something to be investigated. > The build is strange on mips just for the record. Sometimes (just like > in the past) it builds in a hour. Nowadays it's over twelve hours > sometimes, but it seems to be at lease six hours[2]. > Bob, how may I help you to find the root cause of the slow building on > mips at least? graphicsmagick uses FP instructions, and it happens that the machines we currently use for the mips port do not have an FPU. The instructions are emulated, that's why it takes longer to build (or actually to run the testsuite). We have recently changed the ISA to mips32r2, which improves things a bit (up to 40% on some packages), but the gain for graphicsmagick seems to be around 10% only. The real solution is to get new machines with an FPU, we are currently working on that. Aurelien -- Aurelien Jarno GPG: 4096R/1DDD8C9B aurel...@aurel32.net http://www.aurel32.net
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
reassign 796310 src:graphicsmagick 1.3.20-4 fixed 796310 src:graphicsmagick 1.3.21-4 thanks Hi, Since libgraphicsmagick3 no longer exists, the BTS gets confused and thinks this bug isn't fixed, and so britney won't attempt to migrate graphicsmagick: Updating libgraphicsmagick3 introduces new bugs: #796310 Let's see if this fixes that. Emilio
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
On Fri, 25 Sep 2015 06:56:30 +0200 =?UTF-8?B?TMOhc3psw7MgQsO2c3rDtnJtw6lueWkgKEdDUyk=?=wrote: > On Fri, Sep 25, 2015 at 12:41 AM, Emilio Pozuelo Monfort > wrote: > > Yeah, sorry if that came out too harsh. I just meant to say that fixing the > > RC > > bug should take priority as this is blocking other transitions. It's good > > that > > you're looking at it, so no worries. > It was a bit harsh, but no problem. I guess we all have enough things > to do. The updated package is now in NEW, hopefully will be accepted > to the archives soon. > > Kind regards, > Laszlo/GCS > > Hi László, Thanks for getting it uploaded to NEW. I have prodded the FTP masters about fast tracking and will try to finish the resulting transition as fast as possible. :) Thanks, ~Niels signature.asc Description: OpenPGP digital signature
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
* László Böszörményi (GCS), 2015-09-23, 22:13: [1] http://www.barcikacomp.hu/gcs/graphicsmagick_1.3.21-4.dsc Here's my review: -Breaks: pdf2djvu (<= 0.7.21-2) Dropping the Breaks is correct, but I would expect such changes to be documented in the changelog. -Conflicts: libgraphicsmagick -Replaces: libgraphicsmagick +Conflicts: libgraphicsmagick, libgraphicsmagick3 +Replaces: libgraphicsmagick, libgraphicsmagick3 Conflicts/replaces on "libgraphicsmagick" is long obsolete and should be removed. Conflicts/replaces on "libgraphicsmagick3" in necessary because of /usr/{lib,share}/GraphicsMagick-1.3.21/ directories. :-/ Fortunately, can make the new package co-installable with the jessie version by making conflict/replaces versioned: Conflicts: libgraphicsmagick3 (>= 1.3.21) Replaces: libgraphicsmagick3 (>= 1.3.21) +++ graphicsmagick-1.3.21/debian/graphicsmagick.install 2015-09-22 21:55:12.0 +0200 @@ -1,2 +1,3 @@ usr/bin/gm usr/share/doc/graphicsmagick/www +usr/share/man/man1/gm.1 I don't understand what why this change is needed. It's not documented in the changelog. +++ graphicsmagick-1.3.21/debian/libgraphicsmagick++1-dev.links 2015-09-23 00:40:01.0 +0200 @@ -1 +1,2 @@ usr/share/doc/graphicsmagick/www/images usr/share/doc/libgraphicsmagick++1-dev/images +usr/lib/libGraphicsMagick++-Q16.so.11.0.0 usr/lib/libGraphicsMagick++-Q16.so I don't think these symlinks are useful. If upstream build system doesn't create them, then we shouldn't either. +++ graphicsmagick-1.3.21/debian/libgraphicsmagick1-dev.links 2015-09-23 00:23:12.0 +0200 @@ -1 +1,3 @@ usr/share/doc/graphicsmagick/www/images usr/share/doc/libgraphicsmagick1-dev/images +usr/lib/libGraphicsMagick-Q16.so.3.13.0 usr/lib/libGraphicsMagick-Q16.so +usr/lib/libGraphicsMagickWand-Q16.so.2.7.1 usr/lib/libGraphicsMagickWand-Q16.so Ditto. - dh_shlibdeps -a -L libgraphicsmagick3 \ - -l debian/libgraphicsmagick3/usr/lib + dh_shlibdeps -a -L libgraphicsmagick-q16-3 \ + -l debian/libgraphicsmagick-q16-3/usr/lib These days -l and -L shouldn't be needed. -- Jakub Wilk
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
On Thu, Sep 24, 2015 at 1:38 PM, Jakub Wilkwrote: > Conflicts/replaces on "libgraphicsmagick3" in necessary because of > /usr/{lib,share}/GraphicsMagick-1.3.21/ directories. :-/ Fortunately, can > make the new package co-installable with the jessie version by making > conflict/replaces versioned: > > Conflicts: libgraphicsmagick3 (>= 1.3.21) > Replaces: libgraphicsmagick3 (>= 1.3.21) +1 >> +++ graphicsmagick-1.3.21/debian/graphicsmagick.install 2015-09-22 >> 21:55:12.0 +0200 >> @@ -1,2 +1,3 @@ >> usr/bin/gm >> usr/share/doc/graphicsmagick/www >> +usr/share/man/man1/gm.1 > > I don't understand what why this change is needed. It's not documented in > the changelog. A quick check showed the manpage is not installed otherwise, need check. >> +++ graphicsmagick-1.3.21/debian/libgraphicsmagick++1-dev.links 2015-09-23 >> 00:40:01.0 +0200 >> @@ -1 +1,2 @@ >> usr/share/doc/graphicsmagick/www/images >> usr/share/doc/libgraphicsmagick++1-dev/images >> +usr/lib/libGraphicsMagick++-Q16.so.11.0.0 >> usr/lib/libGraphicsMagick++-Q16.so > > I don't think these symlinks are useful. > If upstream build system doesn't create them, then we shouldn't either. I was afraid that other packages may use it to detect the QD support of the library if not now, but in the close future. But sure, if upstream doesn't create this then why should I be such pedantic. >> - dh_shlibdeps -a -L libgraphicsmagick3 \ >> - -l debian/libgraphicsmagick3/usr/lib >> + dh_shlibdeps -a -L libgraphicsmagick-q16-3 \ >> + -l debian/libgraphicsmagick-q16-3/usr/lib > > These days -l and -L shouldn't be needed. OK. Thanks! Will act when I get back home. Laszlo/GCS
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
On 25/09/15 00:10, László Böszörményi (GCS) wrote: > On Wed, Sep 23, 2015 at 8:30 PM, Emilio Pozuelo Monfort >wrote: >> On Tue, 22 Sep 2015 21:58:46 +0200 wrote: >>> You make me wonder. Would it worth to have packages with different >>> quantum depth support in parallel? I mean I would compile / install >>> the package twice; first for QD=16 and then QD=32 to different paths >>> so I can ship the libraries parallel for any given release. >> >> Please fix the aforementioned bug as it's blocking the GCC5 transition, and >> worry about compiling things twice and providing multiple libraries later. > It was just a question for future upstream releases. Life is easier - > my slowness is due to IRL things, just got back home recently and need > to get up early in the morning. But working on the fix / update and > going to upload it soon. Yeah, sorry if that came out too harsh. I just meant to say that fixing the RC bug should take priority as this is blocking other transitions. It's good that you're looking at it, so no worries. Thanks, Emilio
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
On Wed, Sep 23, 2015 at 8:30 PM, Emilio Pozuelo Monfortwrote: > On Tue, 22 Sep 2015 21:58:46 +0200 wrote: >> You make me wonder. Would it worth to have packages with different >> quantum depth support in parallel? I mean I would compile / install >> the package twice; first for QD=16 and then QD=32 to different paths >> so I can ship the libraries parallel for any given release. > > Please fix the aforementioned bug as it's blocking the GCC5 transition, and > worry about compiling things twice and providing multiple libraries later. It was just a question for future upstream releases. Life is easier - my slowness is due to IRL things, just got back home recently and need to get up early in the morning. But working on the fix / update and going to upload it soon. Laszlo/GCS
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
On Fri, Sep 25, 2015 at 12:41 AM, Emilio Pozuelo Monfortwrote: > Yeah, sorry if that came out too harsh. I just meant to say that fixing the RC > bug should take priority as this is blocking other transitions. It's good that > you're looking at it, so no worries. It was a bit harsh, but no problem. I guess we all have enough things to do. The updated package is now in NEW, hopefully will be accepted to the archives soon. Kind regards, Laszlo/GCS
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
* László Böszörményi (GCS), 2015-09-22, 19:40: Also, now might be a good moment to move libGraphicsMagickWand to a seperate binary package. Do you know any package which needs only that library (ie does it worth to be a separate one)? I don't know. I mentioned because the current arrangement is not policy-compliant. Policy §8.1 allows lumping together multiple shared libraries in a single package only if all the SONAMEs always change together. But libGraphicsMagick and libGraphicsMagickWand are versioned independently, and in fact their soversions are already different. -- Jakub Wilk
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
On Tue, 22 Sep 2015 21:58:46 +0200wrote: > On Tue, Sep 22, 2015 at 8:27 PM, Bob Friesenhahn > wrote: > > If there are two packages with different quantum depth, then they should be > > able to co-exist without conflict. Likewise, the existing package should be > > able to co-exist with new packages which are distinguished via quantum depth > > so that existing software continues to work. > You make me wonder. Would it worth to have packages with different > quantum depth support in parallel? I mean I would compile / install > the package twice; first for QD=16 and then QD=32 to different paths > so I can ship the libraries parallel for any given release. Please fix the aforementioned bug as it's blocking the GCC5 transition, and worry about compiling things twice and providing multiple libraries later. Thanks, Emilio
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
On Tue, Sep 22, 2015 at 8:10 PM, Julien Cristauwrote: > On Tue, Sep 22, 2015 at 19:40:42 +0200, László Böszörményi (GCS) wrote: >> Yes, I know that. Still the packages can be tracked during a >> transition via the package name if those were compiled against the >> quantum-depth change or not. I have to change the package name anyway, >> I can't just change the SONAME. Then I don't see the mandatory to >> change the SONAME, the package name change will make it distinct of >> the quantum-depth difference. > Please do change the SONAMEs when you change the package names for this, > especially if upstream provide an option to do so. OK, updated the package[1] if any of you would like to review it - no other change for now. It's getting late, will upload the package tomorrow. Cheers, Laszlo/GCS [1] http://www.barcikacomp.hu/gcs/graphicsmagick_1.3.21-4.dsc
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
On Tue, Sep 22, 2015 at 7:17 PM, Jakub Wilkwrote: > * László Böszörményi (GCS) , 2015-09-22, 08:25: >> >> [2] dget -x http://www.barcikacomp.hu/gcs/graphicsmagick_1.3.21-4.dsc > > You changed the package name, but not the SONAME. That doesn't sound right. Yes, I know that. Still the packages can be tracked during a transition via the package name if those were compiled against the quantum-depth change or not. I have to change the package name anyway, I can't just change the SONAME. Then I don't see the mandatory to change the SONAME, the package name change will make it distinct of the quantum-depth difference. > Changing the SONAME should be a matter of passing > --enable-quantum-library-names to configure, as suggested by Bob Friesenhahn > in #795102. Will re-read that mail from Bob. > Also, now might be a good moment to move libGraphicsMagickWand to a seperate > binary package. Do you know any package which needs only that library (ie does it worth to be a separate one)? Regards, Laszlo/GCS
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
On Tue, 22 Sep 2015, László Böszörményi wrote: On Tue, Sep 22, 2015 at 7:17 PM, Jakub Wilkwrote: * László Böszörményi (GCS) , 2015-09-22, 08:25: [2] dget -x http://www.barcikacomp.hu/gcs/graphicsmagick_1.3.21-4.dsc You changed the package name, but not the SONAME. That doesn't sound right. Yes, I know that. Still the packages can be tracked during a transition via the package name if those were compiled against the quantum-depth change or not. I have to change the package name anyway, I can't just change the SONAME. Then I don't see the mandatory to change the SONAME, the package name change will make it distinct of the quantum-depth difference. If there are two packages with different quantum depth, then they should be able to co-exist without conflict. Likewise, the existing package should be able to co-exist with new packages which are distinguished via quantum depth so that existing software continues to work. Dependent applications should be re-compiled and re-linked against the new packages so they stop using the old library. A factor to consider is that GraphicsMagick packages of the same GraphicsMagick release version and quantum depth will share coder and filter modules which are under a path like /usr/lib/GraphicsMagick-1.3.21/modules-Q16. Changing to the new scheme without also transitioning to a new version will result in over-writing existing coder/filter modules and these will now reference the new GraphicsMagick library rather than the old one. Bob -- Bob Friesenhahn bfrie...@simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/ GraphicsMagick Maintainer,http://www.GraphicsMagick.org/
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
On Tue, Sep 22, 2015 at 8:27 PM, Bob Friesenhahnwrote: > If there are two packages with different quantum depth, then they should be > able to co-exist without conflict. Likewise, the existing package should be > able to co-exist with new packages which are distinguished via quantum depth > so that existing software continues to work. You make me wonder. Would it worth to have packages with different quantum depth support in parallel? I mean I would compile / install the package twice; first for QD=16 and then QD=32 to different paths so I can ship the libraries parallel for any given release. > Dependent applications should be re-compiled and re-linked against the new > packages so they stop using the old library. Does it mean that run-time loading is not possible? A dependent package would ask which one (16/32) you would like and then load the library that provides the specified QD to process the images. Regards, Laszlo/GCS
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
* László Böszörményi (GCS), 2015-09-22, 08:25: [2] dget -x http://www.barcikacomp.hu/gcs/graphicsmagick_1.3.21-4.dsc You changed the package name, but not the SONAME. That doesn't sound right. Changing the SONAME should be a matter of passing --enable-quantum-library-names to configure, as suggested by Bob Friesenhahn in #795102. Also, now might be a good moment to move libGraphicsMagickWand to a seperate binary package. -- Jakub Wilk
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
On Tue, Sep 22, 2015 at 19:40:42 +0200, László Böszörményi (GCS) wrote: > On Tue, Sep 22, 2015 at 7:17 PM, Jakub Wilkwrote: > > * László Böszörményi (GCS) , 2015-09-22, 08:25: > >> > >> [2] dget -x http://www.barcikacomp.hu/gcs/graphicsmagick_1.3.21-4.dsc > > > > You changed the package name, but not the SONAME. That doesn't sound right. > Yes, I know that. Still the packages can be tracked during a > transition via the package name if those were compiled against the > quantum-depth change or not. I have to change the package name anyway, > I can't just change the SONAME. Then I don't see the mandatory to > change the SONAME, the package name change will make it distinct of > the quantum-depth difference. > Please do change the SONAMEs when you change the package names for this, especially if upstream provide an option to do so. Cheers, Julien signature.asc Description: Digital signature
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
On Tue, 22 Sep 2015, László Böszörményi wrote: On Tue, Sep 22, 2015 at 8:27 PM, Bob Friesenhahnwrote: If there are two packages with different quantum depth, then they should be able to co-exist without conflict. Likewise, the existing package should be able to co-exist with new packages which are distinguished via quantum depth so that existing software continues to work. You make me wonder. Would it worth to have packages with different quantum depth support in parallel? I mean I would compile / install the package twice; first for QD=16 and then QD=32 to different paths so I can ship the libraries parallel for any given release. This would be useful since some applications need absolute minimum use of resources (e.g. web applications or processing typical JPEG files) while others care more about the degree of color precision. Dependent applications should be re-compiled and re-linked against the new packages so they stop using the old library. Does it mean that run-time loading is not possible? A dependent package would ask which one (16/32) you would like and then load the library that provides the specified QD to process the images. That would require explicit code in the application and creating a loadable module compiled against each one. Bob -- Bob Friesenhahn bfrie...@simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/ GraphicsMagick Maintainer,http://www.GraphicsMagick.org/
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
On Mon, Sep 21, 2015 at 9:22 AM, László Böszörményi (GCS)wrote: > Correct. It seems all dependencies could be built with the > QuantumDepth change[1], but the C library package name change is > strongly advised. Will do it today with appending q16 to the library > name (as seen with imagemagick which also done the QuantumDepth=16 > change) if there's no objections. [...] > [1] https://release.debian.org/transitions/html/auto-graphicsmagick.html I've updated it and it's ready to be uploaded[2]. But finished late in the evening, would like to get a round of local testing today. Sorry for the delay, Laszlo/GCS [2] dget -x http://www.barcikacomp.hu/gcs/graphicsmagick_1.3.21-4.dsc
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
On Sat, Aug 22, 2015 at 00:03:22 +0200, László Böszörményi wrote: > On Fri, Aug 21, 2015 at 11:28 AM, Jakub Wilkwrote: > > Package: libgraphicsmagick3 > > Version: 1.3.20-4 > > Severity: serious > > > > Please either revert the QuantumDepth change, or change the SONAME. > Of course, I'm going to fix it. Will change the SONAME as the > QuantumDepth change is inevitable. > Ping. It's been a month, and this bug is blocking graphicsmagick's gcc-5 transition. Cheers, Julien signature.asc Description: Digital signature
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
On Mon, Sep 21, 2015 at 8:12 AM, Julien Cristauwrote: > On Sat, Aug 22, 2015 at 00:03:22 +0200, László Böszörményi wrote: >> On Fri, Aug 21, 2015 at 11:28 AM, Jakub Wilk wrote: >> > Please either revert the QuantumDepth change, or change the SONAME. >> Of course, I'm going to fix it. Will change the SONAME as the >> QuantumDepth change is inevitable. >> > Ping. It's been a month, and this bug is blocking graphicsmagick's > gcc-5 transition. Correct. It seems all dependencies could be built with the QuantumDepth change[1], but the C library package name change is strongly advised. Will do it today with appending q16 to the library name (as seen with imagemagick which also done the QuantumDepth=16 change) if there's no objections. Cheers, Laszlo/GCS [1] https://release.debian.org/transitions/html/auto-graphicsmagick.html
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
On Fri, Aug 21, 2015 at 11:28 AM, Jakub Wilk jw...@debian.org wrote: Package: libgraphicsmagick3 Version: 1.3.20-4 Severity: serious Please either revert the QuantumDepth change, or change the SONAME. Of course, I'm going to fix it. Will change the SONAME as the QuantumDepth change is inevitable. Regards, Laszlo/GCS
Bug#796310: libgraphicsmagick3: --with-quantum-depth=16 breaks ABI
Package: libgraphicsmagick3 Version: 1.3.20-4 Severity: serious As I said in #795102, building with QuantumDepth=16 breaks the ABI. Please either revert the QuantumDepth change, or change the SONAME. -- Jakub Wilk