Re: 3.25.90 will probably be delayed
On Mon, Aug 14, 2017 at 3:48 AM,wrote: > Whew, it's done! I wound up doing a minimal number of workarounds: > > * Held colord at a previous version > * Built PackageKit without Vala support > * Built totem without plugin support > > This is a normal amount of workarounds. There's always a few hacks we need > to do to get everything building. So in the end I'm pleased with the level > of quality. > > Better late than never, > Thanks Michael (and everyone else!) for doing the hard work of whipping everything into shape :) Cheers, Nirbheek ___ desktop-devel-list mailing list desktop-devel-list@gnome.org https://mail.gnome.org/mailman/listinfo/desktop-devel-list
Re: 3.25.90 will probably be delayed
Whew, it's done! I wound up doing a minimal number of workarounds: * Held colord at a previous version * Built PackageKit without Vala support * Built totem without plugin support This is a normal amount of workarounds. There's always a few hacks we need to do to get everything building. So in the end I'm pleased with the level of quality. Better late than never, Michael ___ desktop-devel-list mailing list desktop-devel-list@gnome.org https://mail.gnome.org/mailman/listinfo/desktop-devel-list
Re: 3.25.90 will probably be delayed
On Sun, Aug 13, 2017 at 10:33 AM, Richard Hugheswrote: I can do this tomorrow, today I have family things to sort today. Thanks to everybody offering PRs for the colord issues, I've merged all of them and I think we're in good shape now. Yeah that's fine, enjoy your Sunday! Unfortunately I did just file [1] for you to look at tomorrow. It's not clear to me if that will require changes in meson or not. I'm going to try holding back to colord 1.3.5 for this release. I bet that will work. Michael [1] https://github.com/hughsie/colord/issues/57 ___ desktop-devel-list mailing list desktop-devel-list@gnome.org https://mail.gnome.org/mailman/listinfo/desktop-devel-list
Re: 3.25.90 will probably be delayed
On 13 August 2017 at 15:35,wrote: > Now if I can get a new colord tarball, I think this might work. I can do this tomorrow, today I have family things to sort today. Thanks to everybody offering PRs for the colord issues, I've merged all of them and I think we're in good shape now. Richard. ___ desktop-devel-list mailing list desktop-devel-list@gnome.org https://mail.gnome.org/mailman/listinfo/desktop-devel-list
Re: 3.25.90 will probably be delayed
Thanks a bunch. Ernestas! Now if I can get a new colord tarball, I think this might work. I can disable vala support in PackageKit because vala ships its own PackageKit vapi, but that wasn't an option for colord Michael ___ desktop-devel-list mailing list desktop-devel-list@gnome.org https://mail.gnome.org/mailman/listinfo/desktop-devel-list
Re: 3.25.90 will probably be delayed
On Sat, 2017-08-12 at 17:47 -0500, mcatanz...@gnome.org wrote: > On Sat, Aug 12, 2017 at 4:52 PM, Bastien Nocera> wrote: > > What usually happened in the past was that the older version was > > used > > when such a problem happened, which would hopefully have > > streamlined > > the process somewhat ("I used this old version because that new > > one...). > > Yes. > > The problem this time is that I don't know what components are even > to > blame for these issues. E.g. in the PackageKit bug it's not clear > if > the underlying issue is in PackageKit, or gobject-introspection, or > glib itself. Since we don't know where the bug is, I don't know > which > package I can hold back. Right, that's a fair assessment. Might be useful to check afterwards why the problem wasn't hit on C-I, or jhbuild before you ran into it. > > Quite a bit of time was also spent on e-mailing maintainers (myself > > included) that didn't make releases after porting to projects to > > Meson > > but changing jhbuild modulesets. I really think that somebody needs > > to > > spend the time finding and implementing a solution for this > > problem. > > We're migrating away from jhbuild and this is a one-time issue, so I > don't think it's worth spending time on. The solution is to make a > new > release sometime before the release is due if you switch to meson at > the same time that you delete the Autotools build. > > It's annoying to do for 15 modules, which is why I sent out mails, > but > it's also easy to work around by just changing the build type in the > release moduleset. If that's not something that you spent a lot of time on and didn't block the release on, then don't mind me. ___ desktop-devel-list mailing list desktop-devel-list@gnome.org https://mail.gnome.org/mailman/listinfo/desktop-devel-list
Re: 3.25.90 will probably be delayed
On Sat, 2017-08-12 at 15:40 -0500, mcatanz...@gnome.org wrote: > Hi developers, > > I'm still struggling to get buildable release modulesets for > 3.25.90. > As you know, tarballs for that release were due Monday and the > release > was due Wednesday, but it's Saturday now and I haven't delivered it > yet. Normally I spend about one day working on a release and it's not > a > big deal, but this time around there have been such a huge number of > failures that it's taken all week. I can't understate how much worse > this release has been than any I've ever worked on before. > > I was hoping to finish up today, but it's just not going to happen. > Normally all releases have an app or two that fails to build and we > just mark them as skipped in the jhbuildrc that we release and roll > with it. But right now, we have tricky outstanding build failures in > low-level components [1][2] that I don't understand and which I have > spent *far* too much time on already. This is a judgment call, but I > think we're just not in a state to make our .90 beta release right > now, > since if we can't build it ourselves, distros probably won't be able > to > either. Please help fix these tricky issues and then we can see > where > we're at and decide how to adjust our schedule when they're fixed. > I've > uploaded tentative jhbuild modulesets [3] for smoketesting, so you > can > build with the exact same tarball versions and build flags that we > release. You can use a command like this: > > $ jhbuild -f sample-tarball.jhbuildrc -m gnome-apps-3.25.90.modules > build > > The good news is that a huge number of other build failures have > already been fixed. In 80% of these cases the problem was already > fixed > in git and just needed a new tarball release. It's becoming too much > effort to track down maintainers when releases are needed. When you > make incompatible changes to your module or commit build fixes, it's > important to follow the GNOME release cycle as otherwise it creates > a > big problem on release day when we can't build our tarballs against > each other. > > Michael > > [1] https://github.com/hughsie/colord/issues/54 > [2] https://github.com/hughsie/PackageKit/issues/212 > [3] I've uploaded tentative jhbuild modulesets [3] for smoketesting, > so > you can build with the exact same tarball versions and build flags > that > we release. > > ___ > desktop-devel-list mailing list > desktop-devel-list@gnome.org > https://mail.gnome.org/mailman/listinfo/desktop-devel-list I’ve created a pull request for the colord issue. That, along with the commit you referenced, fixes the build for me. ___ desktop-devel-list mailing list desktop-devel-list@gnome.org https://mail.gnome.org/mailman/listinfo/desktop-devel-list
Re: 3.25.90 will probably be delayed
On Sat, Aug 12, 2017 at 4:52 PM, Bastien Nocerawrote: What usually happened in the past was that the older version was used when such a problem happened, which would hopefully have streamlined the process somewhat ("I used this old version because that new one...). Yes. The problem this time is that I don't know what components are even to blame for these issues. E.g. in the PackageKit bug it's not clear if the underlying issue is in PackageKit, or gobject-introspection, or glib itself. Since we don't know where the bug is, I don't know which package I can hold back. Quite a bit of time was also spent on e-mailing maintainers (myself included) that didn't make releases after porting to projects to Meson but changing jhbuild modulesets. I really think that somebody needs to spend the time finding and implementing a solution for this problem. We're migrating away from jhbuild and this is a one-time issue, so I don't think it's worth spending time on. The solution is to make a new release sometime before the release is due if you switch to meson at the same time that you delete the Autotools build. It's annoying to do for 15 modules, which is why I sent out mails, but it's also easy to work around by just changing the build type in the release moduleset. Michael ___ desktop-devel-list mailing list desktop-devel-list@gnome.org https://mail.gnome.org/mailman/listinfo/desktop-devel-list
Re: 3.25.90 will probably be delayed
On Sat, 2017-08-12 at 15:40 -0500, mcatanz...@gnome.org wrote: > Hi developers, > > I'm still struggling to get buildable release modulesets for > 3.25.90. > As you know, tarballs for that release were due Monday and the > release > was due Wednesday, but it's Saturday now and I haven't delivered it > yet. Normally I spend about one day working on a release and it's not > a > big deal, but this time around there have been such a huge number of > failures that it's taken all week. I can't understate how much worse > this release has been than any I've ever worked on before. What usually happened in the past was that the older version was used when such a problem happened, which would hopefully have streamlined the process somewhat ("I used this old version because that new one...). Quite a bit of time was also spent on e-mailing maintainers (myself included) that didn't make releases after porting to projects to Meson but changing jhbuild modulesets. I really think that somebody needs to spend the time finding and implementing a solution for this problem. Cheers ___ desktop-devel-list mailing list desktop-devel-list@gnome.org https://mail.gnome.org/mailman/listinfo/desktop-devel-list
Re: 3.25.90 will probably be delayed
On Sat, Aug 12, 2017 at 3:40 PM, mcatanz...@gnome.org wrote: [3] I've uploaded tentative jhbuild modulesets [3] for smoketesting, so you can build with the exact same tarball versions and build flags that we release. Sigh, Geary Here they are: https://download.gnome.org/teams/releng/3.25.90/ ___ desktop-devel-list mailing list desktop-devel-list@gnome.org https://mail.gnome.org/mailman/listinfo/desktop-devel-list
Re: 3.25.90 will probably be delayed
Hi developers, I'm still struggling to get buildable release modulesets for 3.25.90. As you know, tarballs for that release were due Monday and the release was due Wednesday, but it's Saturday now and I haven't delivered it yet. Normally I spend about one day working on a release and it's not a big deal, but this time around there have been such a huge number of failures that it's taken all week. I can't understate how much worse this release has been than any I've ever worked on before. I was hoping to finish up today, but it's just not going to happen. Normally all releases have an app or two that fails to build and we just mark them as skipped in the jhbuildrc that we release and roll with it. But right now, we have tricky outstanding build failures in low-level components [1][2] that I don't understand and which I have spent *far* too much time on already. This is a judgment call, but I think we're just not in a state to make our .90 beta release right now, since if we can't build it ourselves, distros probably won't be able to either. Please help fix these tricky issues and then we can see where we're at and decide how to adjust our schedule when they're fixed. I've uploaded tentative jhbuild modulesets [3] for smoketesting, so you can build with the exact same tarball versions and build flags that we release. You can use a command like this: $ jhbuild -f sample-tarball.jhbuildrc -m gnome-apps-3.25.90.modules build The good news is that a huge number of other build failures have already been fixed. In 80% of these cases the problem was already fixed in git and just needed a new tarball release. It's becoming too much effort to track down maintainers when releases are needed. When you make incompatible changes to your module or commit build fixes, it's important to follow the GNOME release cycle as otherwise it creates a big problem on release day when we can't build our tarballs against each other. Michael [1] https://github.com/hughsie/colord/issues/54 [2] https://github.com/hughsie/PackageKit/issues/212 [3] I've uploaded tentative jhbuild modulesets [3] for smoketesting, so you can build with the exact same tarball versions and build flags that we release. ___ desktop-devel-list mailing list desktop-devel-list@gnome.org https://mail.gnome.org/mailman/listinfo/desktop-devel-list
Re: 3.25.90 will probably be delayed
On Thu, Aug 10, 2017 at 6:25 PM, mcatanz...@gnome.org wrote: Unfortunately I don't see anything wrong with the generated enums files. It turned out those are actually distributed in the PackageKit tarball, which was briefly a subject of suspicion since there are no problems when building from git, but I don't see anything wrong with them. Since this seems to be a major problem we do need to figure it out now instead of proceeding with the release and expecting distros to build a broken result. But I am stumped. Help welcome from anyone interested in this problem. Michael Obvious workaround is obvious: distros can just build PackageKit with --disable-vala. (Heads-up distros: you'll probably need to do this!) Bug report: https://github.com/hughsie/PackageKit/issues/212 Moving on. Michael ___ desktop-devel-list mailing list desktop-devel-list@gnome.org https://mail.gnome.org/mailman/listinfo/desktop-devel-list
Re: 3.25.90 will probably be delayed
Can you attach your two versions of the .vapi? On Ubuntu 17.10 they are almost identical and would compile against either one. On Thu, 10 Aug 2017 at 10:06wrote: > On Tue, Aug 8, 2017 at 10:45 PM, mcatanz...@gnome.org wrote: > > OK, I'm done sending emails. If you don't have a mail from me, then > > your module is probably fine. > > > > Michael > > Another update. Thanks to lots of help from lots of people, I'm down to > just three known build failures. (There might be more hidden behind > modules I failed to build last night, but probably not many.) Two are > easy and just need new tarballs. There is only one real problem right > now: simple-scan. Here is the build error: > > > ../../../../releases/gnome-apps-3.25.90/simple-scan-3.25.90/src/app-window.vala:1441.95-1441.100: > error: The name `code' does not exist in the context of `Pk.Error' >result_text = _("Failed to install drivers > (error code %d).").printf (e.code); > > ^^ > Compilation failed: 1 error(s), 0 warning(s) > ninja: build stopped: subcommand failed. > > The problem seems to be that in GNOME we have two different, competing > packagekit-glib2.vapi files: one installed by PackageKit and one > installed by vala. The one installed by vala has Error.code, but the > one installed by PackageKit seems to take precedence, and it does not > have Error.code. The two sets of bindings are really > significantly/worryingly different. I'm really not sure what to make of > this. > > Since this might indicate a major issue somewhere in either the > gobject-introspection or vala stacks, I don't think we should proceed > until simple-scan is buildable. > > Michael > > ___ > desktop-devel-list mailing list > desktop-devel-list@gnome.org > https://mail.gnome.org/mailman/listinfo/desktop-devel-list > ___ desktop-devel-list mailing list desktop-devel-list@gnome.org https://mail.gnome.org/mailman/listinfo/desktop-devel-list
Re: 3.25.90 will probably be delayed
On Tue, Aug 8, 2017 at 10:45 PM, mcatanz...@gnome.org wrote: OK, I'm done sending emails. If you don't have a mail from me, then your module is probably fine. Michael Another update. Thanks to lots of help from lots of people, I'm down to just three known build failures. (There might be more hidden behind modules I failed to build last night, but probably not many.) Two are easy and just need new tarballs. There is only one real problem right now: simple-scan. Here is the build error: ../../../../releases/gnome-apps-3.25.90/simple-scan-3.25.90/src/app-window.vala:1441.95-1441.100: error: The name `code' does not exist in the context of `Pk.Error' result_text = _("Failed to install drivers (error code %d).").printf (e.code); ^^ Compilation failed: 1 error(s), 0 warning(s) ninja: build stopped: subcommand failed. The problem seems to be that in GNOME we have two different, competing packagekit-glib2.vapi files: one installed by PackageKit and one installed by vala. The one installed by vala has Error.code, but the one installed by PackageKit seems to take precedence, and it does not have Error.code. The two sets of bindings are really significantly/worryingly different. I'm really not sure what to make of this. Since this might indicate a major issue somewhere in either the gobject-introspection or vala stacks, I don't think we should proceed until simple-scan is buildable. Michael ___ desktop-devel-list mailing list desktop-devel-list@gnome.org https://mail.gnome.org/mailman/listinfo/desktop-devel-list
Re: 3.25.90 will probably be delayed
On Tue, Aug 8, 2017 at 4:37 PM, mcatanz...@gnome.org wrote: Hi, Unfortunately 3.25.90 is going to be delayed indefinitely until all the tarballs build for me. It hasn't been going well so far. As I mentioned last week, there are a much higher number of failures than usual due to the meson switch. I'll be reaching out to individual maintainers as needed. OK, I'm done sending emails. If you don't have a mail from me, then your module is probably fine. Michael ___ desktop-devel-list mailing list desktop-devel-list@gnome.org https://mail.gnome.org/mailman/listinfo/desktop-devel-list