 Diolch, any ideas here? I'm sort of triaging some of them as they come in and then neglecting most of the ones that we already have. I'm not going to hold a lot of help with that. Maybe we need to do a better job of getting new contributors into the kernel team. Maybe we need to have a better way of Why? Is sharing responsibility for the bugs there, like tagging them. I don't know. But as it is, we are doing a really bad job. So any ideas? Please take your mic from here. Tell us some ideas. So it seems either we need more people or we just need to be more strict aboutllen o unrhyw i gyflawni ar hyn o fyfanyl ac wedi'w rhaid i ni'r unrhyw o'r hub warchas i chi rydym ni'n gwych. Rhyw聲 yn ymddangos i'r fan. Rhyw pwysig mewn hwer o gwath, mae hynny i ni'n dryf yn ni i chi. Rhyw pwysig wedi ei hunu a'r thyrmyn o'r hyffordd mewnóg neu rhan o'r tŷ iawn. Rhyw hwn o'r flach nhw'n gweld i'n llwg llwg o bwysig ffrygedig hwn o'r hwn o'r llwg arno'r hunu. ac oeddaid yn eich gwirio wrth i fynd, o'r gweinio eu gwirio. Gymryd fallsonIFD yn fydd iawn, Ond rydych chi'n cael ei gwybod trys i ni ddaeth, ond o'r gweinio, ond o'r gwirio, dwi'n cael ei gweinio, oherwydd a'r gwirio, ond mae'r gwirio alliannol fod yn ystredig ar gweinio. Felly, mae'n fyrde i gweinio. Efallai yn gymuned, wydai chi'n gweinio ythig ar yr wydd, zi mor bwysig cyngor. N? No response? I don't know if they've been interested in that kind of metadata. Yes. Possibly we're probably not quite automatically because sometimes we forget to take off the more info tack. Right. But yes, I think anything that is, it really is more info, no response for several months should be close. I'm not sure about how old are the bugs. If a bug, for example, is for a kernel that was in DBN Lenny or something like this, it probably the best way to deal with it would be to ask, does it still exist and if not close? I bet there might be many very old bugs in it. Yeah, we have had a few people used to do that, I think, Dan, Frazier and TBM used to go through the old bugs, but I think you know, I've been doing that for a while. Well, when we had a package renamed from Linux 2.6 to Linux, most bugs that were filed against Linux 2.6 got a ping or closed saying, please reopen if it really still applies. Yeah, that's why I was going to say for old, you know, old stable and older, I think, closed assuming it's fixed and then they can reopen. Yeah. Rather than maybe for newer stuff you might want a ping. Is it worth having something like a bug squashing party looking at kernel bugs only and then sharing the hard questions? Yeah, I think so. Coming up with template mails and things maybe. Martin asked me if we wanted to do a maybe summit bug squashing at the Linux hotel. We could do that. Martin. Right. So that's a few ideas. Anymore? Or should we move on to the next? I think we might already do this. Do we have our own set of mappings to priorities? Cos most users would think that anything that caused a computer not to break. There is something in the, that went into the handbook. Yeah. Maybe just making sure we all know about that and apply it more aggressively. And maybe include it in the report bug template. Actually, maybe an idea would be to, there's this, what is it, helper tag that you can put to us. So maybe if you think, well, I don't have the time right now to figure it out. But maybe it doesn't really need a lot of kernel knowledge to actually figure it out or actually just try to reproduce. So sort of request for reproduction sort of tags where if it just has a simple instructions of how you got there, just ask other people to try and reproduce. So this is the newcomer tag. The newcomer tag? Yeah, I think so. Something like that. There are probably some that we would use that for. I suspect not that many, unfortunately, but yeah. Of course, because then you would have fixed it yourself, right? All right then. So kernel flavours. It's been quite a while cutting down the kernel flavours so that we get pretty fast build on most architectures. The mips has been very slow but Aurelian just deleted several flavours there. So that won't be too bad. And for MB64, PPC64 YL, RMHF, RM64, they're fast enough but maybe we could have had different flavours if there's demand, if there are good reasons for adding different configurations. So one interesting thing I've been hearing about is the idea of using KVM to make KVM with a minimal VM has a stronger form of containerisation. And Intel showed something called clear containers and there's now something called LKVM, which you can use for that. Should we be shipping a tiny kernel that would enable something like that? Do they not also rely on reworking the user space boot process and things to be optimised? Probably. That doesn't mean it's not worth having such a kernel. And some people have been asking, I think someone asked about RT for the RT feature set on RMHF. Can't remember who hasn't happened yet. I never have any idea how many people are using RT on a real-time patch set. Any other ideas? Anyone really feel that the current flavours aren't suiting their needs? They need a custom configuration or patch set. Don't say GL security because they don't want to work with distro. They don't want to combine GL security with distro patches, so not going to happen. Probably. Everyone's happy with standard kernel packages. Really? It's a little bit large, but it works. We could also start to split large build into more than one packet to remove the parts that are not often used. I noticed that I think Ubuntu has a split between the most common and less common drivers, the latter being optional. I'm not sure how that works or how they decide. What I found with Ubuntu, I don't know whether it's really necessary to do it this way, but they have a special server kernel, I think, in an extra package. I don't know whether it's really sufficient differentiation between this one and the other one. One probably would have to look at the configuration of these to see what's changing. I think the differences they turn preempt off on there. There may be other differences. One thing I was wondering is, as we move to D-Dabs and all that stuff, would it be possible to kill the kernel package, the debug package, because that's a huge package? Oh, yeah. Well, let's turn it to D-Dab. It is part of the spec that we can create our own D-Dab and replace the current debug package with it. Then we should be able to build it for all flavors or architectures without overflowing the mirrors. As far as I know, the deck patch actually landed some days ago. We could, it may work in the short timeframe. It is going to make builds slower. Compressing that debug package is probably the single longest step of an MD64 build. Yeah, it has ended its current. It does not really concurrent. Right. Building separate packages, building separate binary devs is parallelised, I think. Compressing the debug packages takes so long that they end up being the last things running. Some suggestions for splitting out are the meter drivers, which take rather a lot of space. The media drivers. For example, DVP drivers. Sorry, I'm still not... DVP for looking TV. Digital video broadcast. It's not used very often and not by many people. So maybe those who need it can install it. Stacking is also a lot of space. Stacking drivers could also be an extra package. I wouldn't want to... We already do this splitting out into lots of tiny packages for the installer. It's kind of a pain. I'd rather have fewer little packages. I don't think we really want to cope with more than maybe three. Maybe even only two. Maybe stacking is a good thing to leave it out of the default because some people don't like stacking drivers. We have to check which part of the driver tree really uses the space. Then you can decide which one you could split out without producing a lot of outcry by people. The biggest driver directory is net. About 27 megs, 26 megs. Media, then media, 16 megs. Staging, 14 megs. Scuzzi, 11 megs. It does seem to be 8 megs. While it would be nice to split them up, it's one of those sort of infinite manual bike sharing tasks. If you move something out, someone will complain and then you have to have an argument about it and repeat without very clear objective criteria. Which is hard to do. I think we can take the example of the UDAP when we have split in very, very small UDAPs of the drivers. For example, the NIC driver, the names, the one in extra are kind of the usual from now and the one in the main are kind of the one from 10 years ago. 20 years ago in some cases. Yeah, maybe 20 years ago. So splitting is possibility but we should update that regularly and we have to do that. I'm not sure we will continue to have someone looking at that regularly. With the UDAPs, what I'd actually like to do is firstly teach Colonel Wedge to pick all modules under a particular directory. I haven't been included in some other UDAP. Secondly, I'd like to actually look at the DI image configurations which groups of module packages are always placed together and then combine those because there's no point having them separate if they're always used together. I mean that's completely off this topic. Any more thoughts about Colonel Flavours? So testing. There are several. A growing number of self-test programmes included in the upstream source. I've started work on getting those to run under auto package test and that's done. So auto package test has several different ways of running tests. The default I think is to use a container and that's what CI.debin.net uses at the moment. But you can ask to run in a QMU VM and that's what I'm doing for Colonel. My idea is to pull in the Linux image packages and then boot each Colonel using KX and then run the build and run the self-tests. So I have that working in as far as the self-test run but a lot of them are failing. And that could be due to something silly like we haven't turned on the necessary features in the Colonel configuration. But I don't think that's it. I fixed a number of those cases. So this implies that upstream isn't actually running those or at least they're not paying my attention. And if they're not then it's probably not that useful to run these ourselves because we're just going to take the upstream, broke it and not that we did something wrong with the packaging there. And then if CI.debin.net becomes able to run things in a VM, this then prevents the Colonel from ever transitioning to testing. So maybe it's possible to go through these and fix all of them. I probably won't have time to do that. But I'll probably spend a little bit more time getting these. Seeing if there's anything easy and then I'll commit the... I'll push the changes and then maybe someone once would like to look at the test failures or not. There are other test suites that are out of tree that would also be good to run to look for regressions. I don't know how we would include those, whether they would need to be included in the source package or whether they would be purchased separately. And then because auto package test allows you to specify extra effectively build dependencies but for tests. So we could package these test suites and pull them in for the tests. Anyway, interesting working on that. Anyone good ideas for test suites that would be good to run? Does this work? Yeah. It might be worth to look into OpenQA by OpenSusa. Right. They can do quite a lot of stuff. They can even run stuff in a VM. They can take screenshots. So if you want to check the graphical installer, for example, you can also use that. It's quite feature-rich. Right. There are people working on the installer but I just want to be able to test the kernel more or less in isolation, although I suppose there are some things that would... I mean, so what does... Is this OpenQA meant to test the full system? Or is it for testing the... Is this for testing individual packages or is it for testing the whole system? As far as I understood, it's mainly targeting testing the full system. So they also use it for Android, for example, or for the Fedora stuff also. I haven't looked much into it yet. Okay. That probably isn't suitable for a package test, although it's definitely something we should have. That sort of thing is... We should have that sort of thing in Debian. Probably going to be something that they install... We should be part of the installer tests. You know, install the system, try these things, do they still work? So... Holder was working on that at one time and I think Kibbie is now... Talk to them if that's something that you want to work on. Anything for package tests of the Linux package and checking for regressions? There's someone back there, deliver a microphone. Someone passed a microphone. Hello. When we upgraded, for example, for the ARM architecture, when we do an upgrade from Wisi to Jesse, a few platforms broke or couldn't boot, there were a few problems. I'm aware of upstream, now they have a site called kernelsci.org and they're running lava and they're running upstream kernels checking bootability. Maybe it's something we can also try to integrate and test our ARM kernels. They already have a machine farm. So we're just writing the tests there so we can test the kernels boot on some of their devices. Yeah, lava sounds good. This is a tiny separate framework, right? Maybe the thing to do then is to be able to run the auto package test suite in lava. That may be something entirely separate, because that's all about testing that the hardware initializes right, that we have all the right drivers and drives, tree and so on. Yeah, it's a smoke test that machine boots was fine. I mean, you can write your own tests if you want to do something especially. That's probably something separate from the auto package test, but does... I know Neil Williams has been working on lava. I know he was talking about testing deviant kernel packages on lava. Is that in place? That's not in place, but we also discussed about setting the lava service for deviant, but we found that what we did was like a small corridor discussion. We discussed it over lunch. But it seems like it's going to be a lot of maintenance overhead for setting up all these services, because you need to maintain the server, but you also need to maintain a lot of little devices and run the tests and keep track of the results. So, see if somebody already have the farm, they're already maintaining this, why not use their resources and a few tests running our kernels? It might make more sense. I don't know, maybe we can do the effort and try to set up the service if people think it's interesting, but it's probably going to give us more work of maintenance. So, we now have packages, all the packaging files in Git, which is quite convenient, solves a few problems. We still have only the Debian directory in there, which is kind of weird given that upstream is Git as well. But on the other hand, we don't simply... Well, first of all, there are a lot of different tools for working on Debian packages in Git and managing your patches. We have quite a lot of patches, some of which have to be applied to the upstream source before building a ridge.todd.xz, and sometimes we have Fee-Uset patches and they are only applied for some of the bills rather than for all of the bills. So, I don't know which, if any, of the existing Debian packages in Git tools would work with those. I haven't had time to actually try any of them yet, though I may do. Does anyone want to advocate for any particular tool? Does anyone have negative experiences with one of these tools? I'm managing a large patch series. Fagrant? It's a much smaller patch set, but I've basically... It's a much smaller patch set, but for Uboot, we just apply patches... I basically merge the upstream versions and then apply... and then rebase... Well, no, I don't rebase it. But I just apply the patches as a merge commit. So, in your Debian.todd.xz, is it just a single patch for all Debian-specific changes? No, we just merge them in... There aren't a separate series of patches. I just merge them into the upstream branch because we're producing a DFSG-style upstream. Well, that includes the Debian changes. Yeah. But I mean, it keeps track of the new upstream version. I merge it and then that Debian-der corresponds to that new upstream version. But it's a manual workflow. But I imagine you have to rebase the patches using whatever patch system you've got now. Here you'd have the advantages of Git to... Shell, yeah. I've got Git DPM. It works with the patched source. You can just have an upstream branch that's patched. But... An upstream branch that's patched? Yeah. Why not? Okay, it's the name of the branch. It doesn't tell you what in which condition it is in. So it's patched with what? With the DFSG patches. Okay, yeah. I've got one package where I need that, actually. But it won't support the feature sets. Right. Because you only have one source. The other thing I would say about Git DPM is it's not that great if there's multiple maintainers working on the tree at the same time. Because of the way it merges, rebases into things. The unpicking if you're the one who loses the race is quite annoying. It's a bit inconvenient because it uses a little bit unusual merges. It merges the patched tree into the Debian tree, and the later only contains Debian patches as directory, but not the patches itself. Shall I say that again? The tree in the end includes Debian patches. I think that's going to be a common problem with anything for doing patch queue stuff in Git. It's relying on clever merging rather than a quilt-like series file or something. I think they're pretty much all going to have some sort of potential for rebase nightmare. I've got no experience with anything on the scale of the kernel. Or DFS Git patches. But Git build package makes you easy to switch between patched and unpatched. Git build package makes it easy to switch between patched and unpatched sources with its patch queue management. If it is Git, it should cope with a large patch queue, I guess. How much help do you get with merging or rebasing on to a new upstream version in Git build package? It has the support for invoking standard Git rebase for the patch queue branch on upstream imports, over upstream imports. I don't know whether it can be used for rebasing with respect to the upstream Git commits. It does it with respect to the upstream merge commit. I mean a single commit. I don't have experience with Git tape cherries, is that it? Or there are others. Digit? Anyone think Digit is good for maintenance? Wait a minute. Isn't Digit the stuff where you can see the archive as Git? It's kind of orthogonal. It's more like Dput and sBuild and that sort of thing. Although it does let you keep your Git history in a more Git-ish form and expose that to people. Python 3. I don't have problems with Python 3, but Weezy, which is still supported, only has 3.2. That's a little bit unconvenient. Yeah, but we're not... I suppose no one's doing backpawls from said to Weezy. Yeah. Otherwise it should work mostly. What's wrong with 3.2? What would be the problem with 3.2? There are some things missing between 2.7 and 3.2, stuff like ArcPars and other stuff that is used. With 3.3 and newer, you can even build a source that works with both 2.7 and 3. You can't with 3.2. Okay. I think you are actually backpawling from said to Weezy. I know that it doesn't work. Python 3.2 doesn't work. Is that a concern? Okay. Let's do it then. Those are all the topics I came up with. Anyone else got some issue with the kernel maintenance that we should discuss now? Where is that? Looks like we're out of time and topics. I've got that exactly right. Thank you all for coming.