Bug#897945: [Openjdk] Bug#920037: Bug#897945: #897945 still present/breaks with Java 8
Le 05/02/2019 à 15:05, Emmanuel Bourg a écrit : > The real issue is lombok, it needs both Java 8 and 11 to build (and even > 6 and 7! But we managed do to without that). Erratum: I've just figured out how to build lombok with Java 11 only. Once ivyplusplus is taught about the new javac 'release' option it's easy. Uploads will follow soon. Emmanuel Bourg
Bug#897945: [Openjdk] Bug#920037: Bug#897945: #897945 still present/breaks with Java 8
Hi all, Le 05/02/2019 à 14:24, Per Lundberg a écrit : > is this the correct list of packages which can only be built w/ openjdk-8 That's almost correct: - jzmq and openjfx are built with openjdk-11 already - openjdk-8-jre-dcevm has just been removed, replaced by openjdk-11-jre-dcevm (not in testing yet) - leiningen-clojure is about to be updated with Java 11 compatibility - uwsgi builds with openjdk-11, but supports openjdk-8 on kfreebsd. The Java 8 plugin can probably be dropped. - virtualbox switched to openjdk-11 a few days ago - icedtea-web should support Java 11 in the next upstream release The real issue is lombok, it needs both Java 8 and 11 to build (and even 6 and 7! But we managed do to without that). This is a complicated package that is now a key part of the Java ecosystem in Debian, and we can't really do without it. It looks like the latest releases have improved the Java 11 support but I doubt it can build without Java 8. Note that we'll still need OpenJDK 8 as part of the SBT packaging effort (which is required to build Scala 2.12). I wouldn't be surprised to see it required as well to bootstrap Kotlin. So even if we managed to ship Buster without openjdk-8, the package should remain in unstable until this is sorted out. Emmanuel Bourg
Bug#897945: [Openjdk] Bug#920037: Bug#897945: #897945 still present/breaks with Java 8
On 2/4/19 10:07 PM, Moritz Mühlenhoff wrote: > What is the specific use case for this, is there some package which > needs 8 and can't be fixed in time for 11? Yes, it was stated in this thread earlier that such packages do exist. Emmanuel/others, is this the correct list of packages which can only be built w/ openjdk-8 or am I missing something out? If so, it doesn't seem like a huge list and making all of these work on openjdk-11 "could" be doable/the release could possibly live without them. (well, ideally not of course) $ grep-dctrl -FBuild-Depends openjdk-8 -sPackage /var/lib/apt/lists/*Sources Package: virtualbox Package: icedtea-web Package: jzmq Package: leiningen-clojure Package: libbluray Package: lombok Package: openjdk-8 Package: openjdk-8-jre-dcevm Package: openjdk-8 Package: openjfx Package: uwsgi Personally, I still rely on openjdk-8 for my work (because of customer environments still using it for at least 3-5 more years), but I can live with getting it from an unofficial repository. It's more important to provide a smooth user experience for the majority of people, who are perhaps not _developers_ of Java-based software but more _users_ of the same. Best regards, -- Per
Bug#897945: [Openjdk] Bug#920037: Bug#897945: #897945 still present/breaks with Java 8
On Mon, Feb 04, 2019 at 03:04:41PM +0100, Markus Koschany wrote: > I would add a NEWS file to OpenJDK 8 and explain the situation and in > addition create a wrapper around the OpenJDK 8 java command that prints > out a message and quits. javac shall continue to work because it is > needed to build some packages. The debconf approach could also work but > it is a bit more work to implement. Then we also announce it via release > news. Just some thoughts. What is the specific use case for this, is there some package which needs 8 and can't be fixed in time for 11? If we do keep openjdk-8 after all, could we simply omit the jre binary packages when building for distrel=buster? That should ensure that noone uses an unsupported Java (along with a NEWS file maybe) to run Java applications. Cheers, Moritz
Bug#897945: [Openjdk] Bug#920037: Bug#897945: #897945 still present/breaks with Java 8
Am 04.02.19 um 14:56 schrieb Per Lundberg: > On 2/1/19 11:20 AM, Matthias Klose wrote: >> On 01.02.19 10:03, Emmanuel Bourg wrote: > >>> This is an excellent suggestion. We should file a bug for openjdk-8 to >>> implement that. >> please attach the patch. > > Sure, I should be able to write something up. Forbear my ignorance: > should I use debconf to show the message or what do you think would be > right approach? (I haven't been involved at this level in Debian since I > stepped down as package maintainer 15 years ago. :) > > Best regards, > Per > Hi, I would add a NEWS file to OpenJDK 8 and explain the situation and in addition create a wrapper around the OpenJDK 8 java command that prints out a message and quits. javac shall continue to work because it is needed to build some packages. The debconf approach could also work but it is a bit more work to implement. Then we also announce it via release news. Just some thoughts. Best, Markus signature.asc Description: OpenPGP digital signature
Bug#897945: [Openjdk] Bug#920037: Bug#897945: #897945 still present/breaks with Java 8
On 2/1/19 11:20 AM, Matthias Klose wrote: > On 01.02.19 10:03, Emmanuel Bourg wrote: >> This is an excellent suggestion. We should file a bug for openjdk-8 to >> implement that. > please attach the patch. Sure, I should be able to write something up. Forbear my ignorance: should I use debconf to show the message or what do you think would be right approach? (I haven't been involved at this level in Debian since I stepped down as package maintainer 15 years ago. :) Best regards, Per
Bug#897945: [Openjdk] Bug#920037: Bug#897945: #897945 still present/breaks with Java 8
On 01.02.19 10:03, Emmanuel Bourg wrote: > Le 01/02/2019 à 09:17, Per Lundberg a écrit : > >> I think that risk is significant. If we go that route, I would suggest a >> postinst/debhelper message saying that "OpenJDK 8 is included but >> unsupported. Many packages will not work with it. Use at your own risk." >> or something similar. > > This is an excellent suggestion. We should file a bug for openjdk-8 to > implement that. please attach the patch.
Bug#897945: #897945 still present/breaks with Java 8
Le 01/02/2019 à 09:17, Per Lundberg a écrit : > I think that risk is significant. If we go that route, I would suggest a > postinst/debhelper message saying that "OpenJDK 8 is included but > unsupported. Many packages will not work with it. Use at your own risk." > or something similar. This is an excellent suggestion. We should file a bug for openjdk-8 to implement that. Emmanuel Bourg
Bug#897945: #897945 still present/breaks with Java 8
On 1/29/19 3:36 PM, Markus Koschany wrote: > We usually ship only with one JDK per release because of security > support. OpenJDK 8 will be EOL before the end of the Debian 10 release > cycle. We still have a couple of packages that will not compile with > OpenJDK 11 hence we also thought about shipping OpenJDK 8 for building > those packages but declaring it as unsupported at runtime. However, as > this bug report shows, people tend to assume that OpenJDK 8 is supported > as well. Yes, it's easy to get this impression and I think it'll get even worse when more people start using Buster. > So we either end up without OpenJDK 8 in Buster at all and with > less packages, that are only there to build other packages, or we > declare OpenJDK 8 as unsupported at runtime in our release notes which > carries the risk of nobody reading them. I think that risk is significant. If we go that route, I would suggest a postinst/debhelper message saying that "OpenJDK 8 is included but unsupported. Many packages will not work with it. Use at your own risk." or something similar. But removing it altogether is probably less cumulative work for the whole Debian team in the long run, so I think that option should be strongly considered. It will also increase the motivation for people to fix the broken packages to build correctly on OpenJDK 11... :-) (Do we have a list of these packages btw?) Best regards, -- Per
Bug#897945: #897945 still present/breaks with Java 8
Am 29.01.19 um 14:19 schrieb Per Lundberg: > Hi Markus, > > On 1/29/19 1:32 PM, Markus Koschany wrote: >> >>> I am sorry but OpenJDK 8 will not be supported at runtime in >>> Buster. > Another question: if this is the case, should the openjdk-8-jdk package > (and friends) be removed altogether in sid? I looked briefly but > couldn't find any bugs mentioning this. This is Debian bug: https://bugs.debian.org/920037 We usually ship only with one JDK per release because of security support. OpenJDK 8 will be EOL before the end of the Debian 10 release cycle. We still have a couple of packages that will not compile with OpenJDK 11 hence we also thought about shipping OpenJDK 8 for building those packages but declaring it as unsupported at runtime. However, as this bug report shows, people tend to assume that OpenJDK 8 is supported as well. So we either end up without OpenJDK 8 in Buster at all and with less packages, that are only there to build other packages, or we declare OpenJDK 8 as unsupported at runtime in our release notes which carries the risk of nobody reading them. signature.asc Description: OpenPGP digital signature
Bug#897945: #897945 still present/breaks with Java 8
Le 29/01/2019 à 14:19, Per Lundberg a écrit : > Another question: if this is the case, should the openjdk-8-jdk package > (and friends) be removed altogether in sid? We can't remove openjdk-8 yet, it's still required to build some convoluted packages. But Buster will only support OpenJDK 11, this will be documented in the release notes. Emmanuel Bourg
Bug#897945: #897945 still present/breaks with Java 8
Am 29.01.19 um 13:54 schrieb Per Lundberg: [...] > But, the approach you suggest is fine. Markus, will you reopen this bug > or should we file a new one about it, what do you prefer? I think we don't need a new bug report for that. I will change that later today. Regards, Markus signature.asc Description: OpenPGP digital signature
Bug#897945: #897945 still present/breaks with Java 8
Hi Markus, On 1/29/19 1:32 PM, Markus Koschany wrote: > >> I am sorry but OpenJDK 8 will not be supported at runtime in >> Buster. Another question: if this is the case, should the openjdk-8-jdk package (and friends) be removed altogether in sid? I looked briefly but couldn't find any bugs mentioning this.
Bug#897945: #897945 still present/breaks with Java 8
On 1/29/19 1:32 PM, Markus Koschany wrote: Hi Markus and Emmanuel, Emmanuel - the problem is that VisualVM will display its splash screen, then die with an error in the log about java.lang.NoSuchMethodError: java.nio.ByteBuffer.position(I)Ljava/nio/ByteBuffer; >> I am sorry but OpenJDK 8 will not be supported at runtime in >> Buster. If it works with OpenJDK 11, which it does, then we should >> spend the remaining time before the release on more urgent issues. > > I should have added that we compile for Java 9 bytecode with the > - -Dpermit.jdk9.builds=true switch in visualvm and in > libnb-platform18-java. No surprises here that it won't work with > OpenJDK 8. I am fine with this; we should just try to make the user experience a bit smoother. > What we can and probably should do is to force OpenJDK 11 > by using a stricter dependency on default-jdk (>= 2:1.11) | java11-sdk > and by also updating the 03-launcher.patch and remove the alternative > search paths for OpenJDK 7 and OpenJDK 8. The result is that visualvm > simply will not start anymore and quit with "No jdkhome found" if > someone uses an unsupported JDK. That sounds like a reasonable path: the effort should not be huge, and the user experience will be improved. The problem right now (with Java 8) is that it just "dies" as described above, no error message printed. You have to more or less _know_ that you should go digging in ~/.visualvm/1.4/var/log/messages.log for it. This is not so great. But, the approach you suggest is fine. Markus, will you reopen this bug or should we file a new one about it, what do you prefer? -- Best regards, Per
Bug#897945: #897945 still present/breaks with Java 8
-BEGIN PGP SIGNED MESSAGE- Hash: SHA512 Am 29.01.19 um 12:06 schrieb Markus Koschany: > I am sorry but OpenJDK 8 will not be supported at runtime in > Buster. If it works with OpenJDK 11, which it does, then we should > spend the remaining time before the release on more urgent issues. I should have added that we compile for Java 9 bytecode with the - -Dpermit.jdk9.builds=true switch in visualvm and in libnb-platform18-java. No surprises here that it won't work with OpenJDK 8. What we can and probably should do is to force OpenJDK 11 by using a stricter dependency on default-jdk (>= 2:1.11) | java11-sdk and by also updating the 03-launcher.patch and remove the alternative search paths for OpenJDK 7 and OpenJDK 8. The result is that visualvm simply will not start anymore and quit with "No jdkhome found" if someone uses an unsupported JDK. -BEGIN PGP SIGNATURE- iQKTBAEBCgB9FiEErPPQiO8y7e9qGoNf2a0UuVE7UeQFAlxQOa9fFIAALgAo aXNzdWVyLWZwckBub3RhdGlvbnMub3BlbnBncC5maWZ0aGhvcnNlbWFuLm5ldEFD RjNEMDg4RUYzMkVERUY2QTFBODM1RkQ5QUQxNEI5NTEzQjUxRTQACgkQ2a0UuVE7 UeRmiw/+Nxkz+4R+Uk6qgYyFtkTnM6zLvOOrb8QD+jrKpsomm6qstnAJuj0sPogG Rz0q/NUjYT4YYzR7pROjXI6287E3G0fmYyVGep7lQWyEPCWMsKAHpN+Ya7S81mpD PI6c2CQ3MvkdG38v5E+Jt0ILzFGmzE7eNtpnXBPVDrMi7+hLsxULQe+NR5GOdL15 c2kADaexqvxwgoRD75OeWtFapboAaMzm5Lj16q6wC3mygU8d0uBHuAbhMfW8AFIH nHOLoGYj7Dxkhb7dILMCuTqL208oGGhQT/d5sX8xXRLV/WYH7bHCvTlABPMAgjlB IqAXGfSZ70GY2hp1gN3iElAyS8vRjXrlAyTzar4ulx38WqD79qeKtan90+rcL3eA mxZjtrHQtvjDW6JaPuEDkfRbT7gUOa1AldzxjprK/p7DwTpqLWAIGax+xnvWgWC0 V6w5/iCYprDzVSJQhQXDt5VyMOGgDkNyZ/0J2yJAADQ/4uFnsuSnrAw4YO/rlm4k +fYXoBEdGqYHGe3l0BmTdvSOX279n6YzZCQUWX6IbuMeW0kNLtYD8CI355gS/2cG CBNVE50YQZ95+h84tlmaL5Ce+xmEpxy14PNyKoFVfQwQA10ZR2eTvU8SODUtzkRs yed/BEwbFZShUH42d98I/lJj86gmNE2hA2vwjuQQHZEqzKktxDU= =p+tq -END PGP SIGNATURE-
Bug#897945: #897945 still present/breaks with Java 8
Am 29.01.19 um 10:58 schrieb Emmanuel Bourg: > Le 29/01/2019 à 10:40, Per Lundberg a écrit : >> FWIW, version 1.4.2-1 works correctly with openjdk-11-jdk version >> 11.0.2+7-1, but it _does not_ work correct any more with Java 8 >> (openjdk-8-jdk version 8u191-b12-2) > > Thank you for the feedback Per. What issue did you see when using > visualvm with Java 8? You tried running visualvm with Java 8 or you > tried attaching visualvm to a Java 8 VM? > > Emmanuel Bourg I am sorry but OpenJDK 8 will not be supported at runtime in Buster. If it works with OpenJDK 11, which it does, then we should spend the remaining time before the release on more urgent issues. Markus signature.asc Description: OpenPGP digital signature
Bug#897945: #897945 still present/breaks with Java 8
Le 29/01/2019 à 10:40, Per Lundberg a écrit : > FWIW, version 1.4.2-1 works correctly with openjdk-11-jdk version > 11.0.2+7-1, but it _does not_ work correct any more with Java 8 > (openjdk-8-jdk version 8u191-b12-2) Thank you for the feedback Per. What issue did you see when using visualvm with Java 8? You tried running visualvm with Java 8 or you tried attaching visualvm to a Java 8 VM? Emmanuel Bourg
Bug#897945: #897945 still present/breaks with Java 8
FWIW, version 1.4.2-1 works correctly with openjdk-11-jdk version 11.0.2+7-1, but it _does not_ work correct any more with Java 8 (openjdk-8-jdk version 8u191-b12-2) This worked correctly with 1.3.9-1 after downgrading the libnb-*-java packages as suggested by Ben - visualvm worked fine on Java 8 with this combination of packages. So, we should either: - Reopen the bug and fix the problem (likely by compiling visualvm with openjdk-8, which should make it work on newer JDK versions also) - Accept the breakage and indicate this somehow, preferably in the package metadata. (I don't know if we can use dpkg dependencies in this case? It's perfectly legal to have both OpenJDK 11 and 8 _installed_ on a machine; it's just that the latter is unusable with visualvm from now on.) Because the resolution here is not perfectly clear, I will refrain from just reopening the bug until it has been discussed a bit more. Best regards, -- Per