Re: Build and test failures with Qt 5.6 and Qt 5.3
On Wednesday, November 11, 2015 10:22:13 AM CET Jan Kundrát wrote: > On Tuesday, 10 November 2015 08:34:02 CET, Martin Graesslin wrote: > > Yes, most tests don't require a WM. Especially the Net* tests > > simulate being a > > window manager. Only the KWindow* tests need a window manager. > > Interesting; it was a Heisenbug, apparently. I cannot reproduce it anymore, > but I've added code which at least prints out a warning if the Xvfb and/or > openbox dies for some reason. > > > Well for me there is not much to do. I don't get them failing > > on my system, so > > I have no clue what I should fix to make them work again. > > I believe that the CI system and the tests together should be designed to > provide enough diagnostic information. If you as a developer say "I cannot > reproduce it on my system, therefore I cannot help you", what can I do as a > CI administrator to improve the situation? Should we perhaps reduce the > number of CI-specific scripting to make the deployments closer to what the > developers run? Would you like pre-built VM images? Well I did what the CI system does. I started Xvfb with exactly the same command as in the build output and also used openbox on that Xvfb and run the unit test on the Xvfb. So I think I reproduced the setup. I really tried hard to get to the broken state and tried also a few different setups which all did not result in a broken test. It takes a long time till I give up on trying to reproduce a test, in this case it has happened. (I also must point out that I think there is something somewhere horribly broken if we need to adjust our tests because Qt changed - currently I need to adjust the tests for each release, I'm rather pissed by the "ABI stability" Qt provides). I wouldn't know what further information the tests could provide to debug it. I could tell if I were able to reproduce them and see what goes wrong. At the moment all I have is "wm might have crashed" which is nothing the tests can check for. In the long run I want to spend some time in getting our "Xvfb + openbox" setup replaced by kwin_wayland on the virtual backend. It would mean KWin gets automatically more testing (that's the selfish part) and we have a known base for our tests and don't depend on openbox. Cheers Martin signature.asc Description: This is a digitally signed message part. ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel
Re: Build and test failures with Qt 5.6 and Qt 5.3
On Tuesday, 10 November 2015 08:34:02 CET, Martin Graesslin wrote: Yes, most tests don't require a WM. Especially the Net* tests simulate being a window manager. Only the KWindow* tests need a window manager. Interesting; it was a Heisenbug, apparently. I cannot reproduce it anymore, but I've added code which at least prints out a warning if the Xvfb and/or openbox dies for some reason. Well for me there is not much to do. I don't get them failing on my system, so I have no clue what I should fix to make them work again. I believe that the CI system and the tests together should be designed to provide enough diagnostic information. If you as a developer say "I cannot reproduce it on my system, therefore I cannot help you", what can I do as a CI administrator to improve the situation? Should we perhaps reduce the number of CI-specific scripting to make the deployments closer to what the developers run? Would you like pre-built VM images? Cheers, Jan -- Trojitá, a fast Qt IMAP e-mail client -- http://trojita.flaska.net/ ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel
Re: Build and test failures with Qt 5.6 and Qt 5.3
On Monday, November 9, 2015 6:00:43 PM CET Jan Kundrát wrote: > On Monday, 9 November 2015 08:21:21 CET, Martin Graesslin wrote: > > I'm sorry, but I'm not able to reproduce the failure on my > > local Qt 5.6 setup. > > The failures look like there is no WM running on the CI system, > > maybe openbox > > crashed? (We had that before that our tests were able to trigger crashes > > in > > openbox). > > All Qt versions are hitting the same pool of VMs, and the failure only > happens on Qt 5.6. Unless there's something in Qt 5.6 which makes openbox > crash, I don't think that it's a plausible explanation. > > Anyway, if you look at the test log, you can see that the remaining tests > all passed. Would they still pass even if there was no running WM at that > point? Yes, most tests don't require a WM. Especially the Net* tests simulate being a window manager. Only the KWindow* tests need a window manager. > > I can add some extra diagnostics to the build scripts if it's really > needed, but let's figure out other possibilities first, please. Well for me there is not much to do. I don't get them failing on my system, so I have no clue what I should fix to make them work again. Cheers Martin signature.asc Description: This is a digitally signed message part. ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel
Re: Build and test failures with Qt 5.6 and Qt 5.3
On Monday, 9 November 2015 08:21:21 CET, Martin Graesslin wrote: I'm sorry, but I'm not able to reproduce the failure on my local Qt 5.6 setup. The failures look like there is no WM running on the CI system, maybe openbox crashed? (We had that before that our tests were able to trigger crashes in openbox). All Qt versions are hitting the same pool of VMs, and the failure only happens on Qt 5.6. Unless there's something in Qt 5.6 which makes openbox crash, I don't think that it's a plausible explanation. Anyway, if you look at the test log, you can see that the remaining tests all passed. Would they still pass even if there was no running WM at that point? I can add some extra diagnostics to the build scripts if it's really needed, but let's figure out other possibilities first, please. With kind regards, Jan -- Trojitá, a fast Qt IMAP e-mail client -- http://trojita.flaska.net/ ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel
Re: Build and test failures with Qt 5.6 and Qt 5.3
On Saturday, November 7, 2015 11:59:48 AM CET David Faure wrote: > > - [5.6] kwindowsystem: KWindowInfoX11Test fails tests > > Martin, can you take a look at > http://ci-logs.kde.flaska.net/0c/0c4039dc8d4c3fa9eed18a4f5604a90648fe3e84/re > builddep/rebuilddep-kf5-qt56-gcc-el7/d8b3c13/shell_output.log ? I'm sorry, but I'm not able to reproduce the failure on my local Qt 5.6 setup. The failures look like there is no WM running on the CI system, maybe openbox crashed? (We had that before that our tests were able to trigger crashes in openbox). Cheers Martin signature.asc Description: This is a digitally signed message part. ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel
Re: Build and test failures with Qt 5.6 and Qt 5.3
On Sunday 08 November 2015 15:06:51 Jan Kundrát wrote: > >> [root@ci-el7-a-4 ~]# grep -c opendocument /usr/share/mime/subclasses > >> 0 > >> [root@ci-el7-a-4 ~]# rpm -qf /usr/share/mime/subclasses > >> shared-mime-info-1.1-7.el7.x86_64 OK, clearly CentOS patches the application/zip inheritance out of freedesktop.org.xml I hate it when distros think they know better than upstream and make their own changes, which are not distro-specific, but rather their own view on how things should be. Yes this inheritance is arguable, and shows the lack for a "private inheritance" mechanism (to avoid .odt opening up in ark if no office suite is installed), but this "argumentation" should be done on the XDG list and solved everywhere, rather than by hacking s-m-i and making it incompatible with other distros :( I'll make the test pass on CentOS, but I'm not happy about such hacks. Feel free to forward this to whoever maintains s-m-i at CentOS BTW. -- David Faure, fa...@kde.org, http://www.davidfaure.fr Working on KDE Frameworks 5 ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel
Re: Build and test failures with Qt 5.6 and Qt 5.3
On Saturday, 7 November 2015 11:59:48 CET, David Faure wrote: Having this for all of KF5 would definitely be useful. And it's done now: http://ci-logs.kde.flaska.net/matrix.html - [5.3] ballo, missing QList::join - [5.3] knewstuff, QTimer::singleShot and new slots - [5.3] modemmanager-qt, checks for Qt>=5.4 - bluez-qt fails to install because it ignores the install prefix when installing the udev rules, and tries to modify a live system - [5.6] frameworkintegration, test failure through some font stuff - kwidgetsaddons, unstable tests [1] [2]. They tend to fail, but the failures appear to be perhaps timing-related? - The karchive, kconfig, kcoreaddons, kio, plasma-frgamework have the same issue as previously reported -- just for the completeness. Cheers, Jan [1] http://ci-logs.kde.flaska.net/89/8956b5338b3daf9e2e53f7fc44b23e879f3844a8/rebuilddep/rebuilddep-kf5-qt54-gcc-el7/e1103bb/shell_output.log [2] http://ci-logs.kde.flaska.net/89/8956b5338b3daf9e2e53f7fc44b23e879f3844a8/rebuilddep/rebuilddep-kf5-qt56-gcc-el7/4c4af2c/shell_output.log -- Trojitá, a fast Qt IMAP e-mail client -- http://trojita.flaska.net/ ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel
Re: Build and test failures with Qt 5.6 and Qt 5.3
On Sunday, 8 November 2015 12:17:19 CET, David Faure wrote: Any chance you could do some debugging around there? I'm afraid I don't have time and motivation to fix this, sorry. I'll be happy to spin a VM with an identical setup for you, though, if you're OK with IPv6-only access. If so, please send me your SSH pubkey and I'll get it running. [root@ci-el7-a-4 ~]# grep -c opendocument /usr/share/mime/subclasses 0 [root@ci-el7-a-4 ~]# rpm -qf /usr/share/mime/subclasses shared-mime-info-1.1-7.el7.x86_64 That is really broken. The inheritance of opendocument.text from application/zip has always been there in s-m-i (since 0.18). Do you actually have /usr/share/mime/packages/freedesktop.org.xml at all ? Yup, that file is present. If yes, then I don't know what could have gone wrong; could you maybe send me a .zip of all of your /usr/share/mime ? http://jkt.flaska.net/tmp/centos7-usr-share-mime.tar Cheers, Jan -- Trojitá, a fast Qt IMAP e-mail client -- http://trojita.flaska.net/ ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel
Re: Build and test failures with Qt 5.6 and Qt 5.3
>> For how long are we going to support [Qt] 5.3? > > Good question. > > So that we don't need to reopen this discussion regularly, I would like to > suggest a general rule/policy: > > We will always support the last 3 Qt x.y releases. > > Right now that means 5.3 + 5.4 + 5.5. > > When 5.6 is out, we can drop 5.3. That sounds ok. Cheers, Ivan ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel
Re: Build and test failures with Qt 5.6 and Qt 5.3
On Saturday 07 November 2015 21:42:33 Jan Kundrát wrote: > Hi David, > thanks for looking into this. I'm happy that you find the additional > coverage useful. At this time, the infrastructure cannot easily send out > automated e-mails only upon a change in the state of a build job -- if > someone has some time and is willing to improve this, I'll be happy to walk > them through. Actually, I'm not sure I want more CI noise by email :-) (because there are of course false positives, i.e. setup-related failures) I'll just try to remember to keep an eye on your matrix in addition to build.kde.org, but feel free to poke me when you see red again, after we make it green ;) > >> - [5.6] kcoreaddons: different number formatting > > > > This test sets a C locale. Could it be that on your system, the C locale > > doesn't include having the comma as thousands-separator? > > The build jobs appear to be running with LANG=en_US.UTF-8. My ssh setup > apparently forwards these LANG and LC_* variables, so I cannot guarantee > 100% correctness, but here's how the number formatting works in root's > bash: > > [root@ci-el7-a-4 ~]# LC_ALL=C printf "%'.3f\n" 12345678.901 > 12345678.901 > [root@ci-el7-a-4 ~]# LC_ALL=en_US.utf8 printf "%'.3f\n" 12345678.901 > 12,345,678.901 > > Is that a correct behavior? I get the same, at least. > > Or maybe the system's locale still interfers, i.e. KFormat > > format(QLocale::c()); > > isn't enough to -really- use the C locale? > > Note that the failure is specific to Qt 5.6, which IMHO suggests that > there's some behavior change in Qt. OK, good point. And looking at bash's behaviour, maybe this is a bugfix, i.e. the C locale isn't supposed to get en_US-like thousands separators Someone needs to dig into Qt changes and/or to make a qt-only testcase. (don't have time right now). > >> - kio, in all versions: test failures in KNewFileMenuTest::test(text file > >> with jpeg extension) > > > > Also a mimetype problem. kcoreaddons' kde5.xml adds "*.doc" as a pattern > > for text/plain, but surely it doesn't mean for it to become the > > main extension. > > Can you the value of > > QMimeDatabase().mimeTypeForName("text/plain").preferredSuffix() > > on your system? Here's it's "txt", I suspect it's "doc" on your > > system. Not sure why yet > > though, but let's first check that. > > It prints out "txt". OK so why does this code give you ".doc" ? kio/src/filewidgets/knewfilemenu.cpp:556: chosenFileName += QLatin1Char('.') + wantedMime.preferredSuffix(); Any chance you could do some debugging around there? > >> - kservice, in all versions: test failure > > > > It appears that your system doesn't know that opendocument > > inherits application/zip. > > Do you have this line in /usr/share/mime/subclasses ? > > application/vnd.oasis.opendocument.text application/zip > > That line is missing: > > [root@ci-el7-a-4 ~]# grep -c opendocument /usr/share/mime/subclasses > 0 > [root@ci-el7-a-4 ~]# rpm -qf /usr/share/mime/subclasses > shared-mime-info-1.1-7.el7.x86_64 That is really broken. The inheritance of opendocument.text from application/zip has always been there in s-m-i (since 0.18). Do you actually have /usr/share/mime/packages/freedesktop.org.xml at all ? If yes, then I don't know what could have gone wrong; could you maybe send me a .zip of all of your /usr/share/mime ? -- David Faure, fa...@kde.org, http://www.davidfaure.fr Working on KDE Frameworks 5 ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel
Re: Build and test failures with Qt 5.6 and Qt 5.3
On Saturday 07 November 2015 12:55:45 Ivan Čukić wrote: > For how long are we going to support [Qt] 5.3? Good question. So that we don't need to reopen this discussion regularly, I would like to suggest a general rule/policy: We will always support the last 3 Qt x.y releases. Right now that means 5.3 + 5.4 + 5.5. When 5.6 is out, we can drop 5.3. -- David Faure, fa...@kde.org, http://www.davidfaure.fr Working on KDE Frameworks 5 ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel
Re: Build and test failures with Qt 5.6 and Qt 5.3
Ah, for the *** sake. My fault, I expected that the 'since' tag is missing from that one like it is for setClearColor and didn't check. I've removed the call in question when compiling against Qt 5.3. It will manifest as an issue for people that use dark colour scheme, but since the activities KCM is used only by plasma which requires Qt 5.4, we should be ok. Thanks for the reporting. Cheerio, Ivan -- KDE, ivan.cu...@kde.org, http://cukic.co/ gpg key id: 850B6F76 ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel
Re: Build and test failures with Qt 5.6 and Qt 5.3
On Saturday, 7 November 2015 22:38:16 CET, Ivan Čukić wrote: - [5.3] kactivities: won't build due to QQuickWidget::setClearColor which is 5.4+ only. It should be ok now. I've added a work-around (essentially how the method is implemented in 5.4 and 5.5). kactivities/src/workspace/settings/imports/dialog.cpp:77:15: error: no member named 'quickWindow' in 'QQuickWidget' -- Trojitá, a fast Qt IMAP e-mail client -- http://trojita.flaska.net/ ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel
Re: Build and test failures with Qt 5.6 and Qt 5.3
> - [5.3] kactivities: won't build due to QQuickWidget::setClearColor which is > 5.4+ only. It should be ok now. I've added a work-around (essentially how the method is implemented in 5.4 and 5.5). Cheers, Ivan ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel
Re: Build and test failures with Qt 5.6 and Qt 5.3
Hi David, thanks for looking into this. I'm happy that you find the additional coverage useful. At this time, the infrastructure cannot easily send out automated e-mails only upon a change in the state of a build job -- if someone has some time and is willing to improve this, I'll be happy to walk them through. Having this for all of KF5 would definitely be useful. I'll see into doing this. - [5.6] kcoreaddons: different number formatting This test sets a C locale. Could it be that on your system, the C locale doesn't include having the comma as thousands-separator? The build jobs appear to be running with LANG=en_US.UTF-8. My ssh setup apparently forwards these LANG and LC_* variables, so I cannot guarantee 100% correctness, but here's how the number formatting works in root's bash: [root@ci-el7-a-4 ~]# LC_ALL=C printf "%'.3f\n" 12345678.901 12345678.901 [root@ci-el7-a-4 ~]# LC_ALL=en_US.utf8 printf "%'.3f\n" 12345678.901 12,345,678.901 Is that a correct behavior? Or maybe the system's locale still interfers, i.e. KFormat format(QLocale::c()); isn't enough to -really- use the C locale? Note that the failure is specific to Qt 5.6, which IMHO suggests that there's some behavior change in Qt. - kio, in all versions: test failures in KNewFileMenuTest::test(text file with jpeg extension) Also a mimetype problem. kcoreaddons' kde5.xml adds "*.doc" as a pattern for text/plain, but surely it doesn't mean for it to become the main extension. Can you the value of QMimeDatabase().mimeTypeForName("text/plain").preferredSuffix() on your system? Here's it's "txt", I suspect it's "doc" on your system. Not sure why yet though, but let's first check that. It prints out "txt". - kservice, in all versions: test failure It appears that your system doesn't know that opendocument inherits application/zip. Do you have this line in /usr/share/mime/subclasses ? application/vnd.oasis.opendocument.text application/zip That line is missing: [root@ci-el7-a-4 ~]# grep -c opendocument /usr/share/mime/subclasses 0 [root@ci-el7-a-4 ~]# rpm -qf /usr/share/mime/subclasses shared-mime-info-1.1-7.el7.x86_64 Cheers, Jan -- Trojitá, a fast Qt IMAP e-mail client -- http://trojita.flaska.net/ ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel
Re: Build and test failures with Qt 5.6 and Qt 5.3
>> - [5.3] kactivities: won't build due to QQuickWidget::setClearColor which >> is 5.4+ only. > > Ivan, can this be worked around? I'll have to check. (if nothing else, I'll disable that part - it is for the settings module, not really a part of the framework) For how long are we going to support 5.3? Cheers, Ivan ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel
Re: Build and test failures with Qt 5.6 and Qt 5.3
On Friday 06 November 2015 17:21:33 Jan Kundrát wrote: > Hi folks, > please take a look at the CI status matrix at [1]. TL;DR: There are build > failures with Qt 5.3, plenty of test failures, and some test failures > specific to the still unreleased Qt 5.6. > > Oh, and a cmake 3.4-git regression that I'm not going to bother you with. > Reported, known, downgraded for now. > > This testing is more or less a side effect of kio's and plasma-workspace > presence in Gerrit. If you like this CI coverage, if you appreciate the > range of covered compilers and Qt5 versions, or if you would like it to > extend the coverage to all of KF5 and/or plasma or some other apps, please > let me know. Having this for all of KF5 would definitely be useful. Clearly it spots additional issues compared to jenkins, the more CIs we have the better for quality ;) However I admit that I keep forgetting to take a look at http://ci-logs.kde.flaska.net/matrix.html in addition to jenkins, so your email is very useful. > A quick summary (please click the "console" link in the matrix overview to > see full log): > > - [5.3] kactivities: won't build due to QQuickWidget::setClearColor which > is 5.4+ only. Ivan, can this be worked around? > - [5.3] kcoreaddons: QDebug::noquote Fixed. > - [5.3] kwidgetsaddons: QTimer::singleShot and the new slot syntax Fixed. > - [5.3] plasma-workspace: QString::splitRef You mean plasma-framework. Nick, can you ifdef this depending on the Qt version? > - [5.6] karchive: wrong number of items in the archive I just committed more debug output in case of failure, to find out more. The result is staggering, null bytes everywhere. > - [5.6] kconfig: error which I don't understand Possibly also related to null bytes, let's dig more after fixing karchive. > - [5.6] kcoreaddons: different number formatting This test sets a C locale. Could it be that on your system, the C locale doesn't include having the comma as thousands-separator? Or maybe the system's locale still interfers, i.e. KFormat format(QLocale::c()); isn't enough to -really- use the C locale? > - kio, in all versions: test failures in KNewFileMenuTest::test(text file > with jpeg extension) Also a mimetype problem. kcoreaddons' kde5.xml adds "*.doc" as a pattern for text/plain, but surely it doesn't mean for it to become the main extension. Can you the value of QMimeDatabase().mimeTypeForName("text/plain").preferredSuffix() on your system? Here's it's "txt", I suspect it's "doc" on your system. Not sure why yet though, but let's first check that. > - kservice, in all versions: test failure It appears that your system doesn't know that opendocument inherits application/zip. Do you have this line in /usr/share/mime/subclasses ? application/vnd.oasis.opendocument.text application/zip > - [5.6] kwindowsystem: KWindowInfoX11Test fails tests Martin, can you take a look at http://ci-logs.kde.flaska.net/0c/0c4039dc8d4c3fa9eed18a4f5604a90648fe3e84/rebuilddep/rebuilddep-kf5-qt56-gcc-el7/d8b3c13/shell_output.log ? > - plasma-framework, all versions: wrong dialog dimensions/offsets Same as what we can currently see in Jenkins, so that one is more expected than all of the others... > [1] http://ci-logs.kde.flaska.net/matrix.html -- David Faure, fa...@kde.org, http://www.davidfaure.fr Working on KDE Frameworks 5 ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel
Build and test failures with Qt 5.6 and Qt 5.3
Hi folks, please take a look at the CI status matrix at [1]. TL;DR: There are build failures with Qt 5.3, plenty of test failures, and some test failures specific to the still unreleased Qt 5.6. Oh, and a cmake 3.4-git regression that I'm not going to bother you with. Reported, known, downgraded for now. This testing is more or less a side effect of kio's and plasma-workspace presence in Gerrit. If you like this CI coverage, if you appreciate the range of covered compilers and Qt5 versions, or if you would like it to extend the coverage to all of KF5 and/or plasma or some other apps, please let me know. A quick summary (please click the "console" link in the matrix overview to see full log): - [5.3] kactivities: won't build due to QQuickWidget::setClearColor which is 5.4+ only. - [5.3] kcoreaddons: QDebug::noquote - [5.3] kwidgetsaddons: QTimer::singleShot and the new slot syntax - [5.3] plasma-workspace: QString::splitRef - [5.6] karchive: wrong number of items in the archive - [5.6] kconfig: error which I don't understand - [5.6] kcoreaddons: different number formatting - kio, in all versions: test failures in KNewFileMenuTest::test(text file with jpeg extension) - kservice, in all versions: test failure - [5.6] kwindowsystem: KWindowInfoX11Test fails tests - plasma-framework, all versions: wrong dialog dimensions/offsets [1] http://ci-logs.kde.flaska.net/matrix.html -- Trojitá, a fast Qt IMAP e-mail client -- http://trojita.flaska.net/ ___ Kde-frameworks-devel mailing list Kde-frameworks-devel@kde.org https://mail.kde.org/mailman/listinfo/kde-frameworks-devel