 Welcome everybody, we have, next up is Michael Bank. He's a Debian GNU herd porter and he's going to talk about the GNU herd port and how it's going and what comes next for it. Michael Bank. Okay, yeah. Thanks everybody for showing up. So I was a bit worried that nobody would be interested in the herd anymore, but at least there are some people who want to see how bad it's going. So, sorry? So, well, actually I wanted to run this on the herd, but I checked it out just in front of the audience because my notebooks have a high SVG, as you saw I have to change the resolution and I didn't know how that worked, but I had X running, I had this presentation running just in the wrong resolution. So basically it still works natively. It works pretty well in QEMU and in Zen, but yeah, not today. So, it's been a long time. That's a quote from like, I don't know, ten years ago maybe. Somebody might remember Jeroen Deckers. He made a big flame war on Debbie and DeVal when Adam Heath said that the top level herd directory should be removed because it's not an FHS and then there was a lot of big flame war and Marcus gave that advice. So, I want to talk about a bit about the past, the present and the future because some of you might not know, I mean probably all of you are quite aware about that it hasn't been released yet and that kind of stuff, but not really what the real status is. So, that was a pretty good summary on the age recently. I don't know where they have seen it, about how it came to be. So, you can read that up if you didn't. I mean we thought it was pretty accurate. I don't know, I mean how they researched it because we didn't hear anything from them, but it was pretty good. So, but in general the herd was designed by Thomas Bushnell BSG who's I think still a Devin developer, but not that involved anymore and I don't think he's around. And with some input by others including Richard. And, well, there was a lot of thought about what to do, what kernel to base it on and stuff like that because as you might know the herd is not a kernel, it's a user space system so it needs a micro kernel to run on and it was at that time decided that it should run on Mach which Thomas apparently didn't like that much, but he craved in and then did it. He implemented the core system along with Ronald McGrath who is maybe not a Devin developer anymore, at least he's also not very active anymore, but he's still a Glypsy maintainer and he's original Glypsy author. Yeah, Richard didn't really code on the herd as far as I know so he just gave some input every now and then. In general the architecture, as I said, it's GNU Mach micro kernel which includes the virtual memory management and the device drivers and also the network layer which is, yeah, both the device drivers and the network is Linux code actually so it hasn't been written directly for the herd and at this point it's still, it's 2.0 Linux device drivers so we are really happy that it works in Zen and QEMU because they emulate such old hardware that it easily works with Linux 2.0 but it's quite difficult with new stuff so I'm happy that it works and I think that. On top of it there is the POSIX layer so actually G-Lyp C for the herd is much bigger than for Linux because you have just the Mach micro kernel which does all this Mach messages and on top of that you have the herd servers and then on top of that all the POSIX stuff is implemented in terms of herd and Mach messages so it's quite a bit bigger and so all what you know as a system call basically which is implemented in the kernel and Linux is implemented in G-Lyp C on the herd and on top of that there are several herd servers who provide authentication networking file systems in a herdish way and you have the famous or infamous herd translators which are a bit like fused file systems but you can basically do anything with them they're just like docking things in the file system you cannot just do file system operations on them because you can with fused I believe you can basically do any kind of stuff you want with them which is provided by the herd libraries like the diskfs library or the storeio library and stuff like that which as they're all implemented in user space makes it in principle quite nice to develop against if people are interested so Thomas and Roland are still around I mean if somebody posts an interesting question on the main mailing list sometimes it just takes 20 minutes for Roland to comment on it and he always has good advice and he commits the herd-specific G-Lyp C stuff that our porters or developers are sending to libc-alpha so as I said, yeah, he's still around but sometimes he just has some... yeah, make it up again but he's a great guy then there was the second and third generations well the second generation, I don't know maybe in the mid-90s there was Way Before My Time there was Miles Bader and Mark Katniss they joined in as GNU people there was a couple of other people I started developing more user-friendly stuff like HostMux or I don't know how it's actually called FTPFS which already in the mid-90s was like a Fuse file system like as it's HFS you see today so with HostMux and stuff you could just write cd-slash-ftp.gnu.org-slash and then it would immediately go on the network and you could just use FTP transparently 15 years ago but, well, Linux has all that these days as well but just to mention that I think it was them who basically did that they're no longer around I mean I see Miles every once in a while in Hashtavian but they are not around in the community anymore and then in the late 90s I would say Marcus Brinkman and Neil Wallfield I think they also both stepped down as Devin developers not so long ago they joined, I mean Marcus started the Devin port did a lot of effort in that he designed the Hurt console back then there was only the Mach console which is pretty weird but the Hurt console is quite nice you can catch it, it has Unicode and all kinds of stuff Neil contributed important fixes some of them haven't been applied yet but mostly he contributed a project spreading library which boosted porting of applications quite a bit because before we only had a Mach threading library which you couldn't really use from user space it wasn't used so everything which had relied on pthread was not going and due to that we were able to port a lot of stuff that was maybe 10 years ago and then in the like I don't know 2002, 2003 they decided they wrote a paper an academic paper on problems with Mach and decided to start a new well first they decided to port Hurt on L4 then it was called HurtNG for next generation because there were problems with L4 so to give a sneak preview actually HurtNG has been discontinued we're still using GNU Mach we're not terribly happy about that but it turned out that L4 had well they believed that it had issues for security stuff so sending messages wasn't as secure as they wanted it to be then they thought about going for another microkernel that Professor Shapiro implemented but that didn't work either and Neil he's still in academia he's working on a microkernel for marine management starting to fade out I think this works can you still hear me? so they basically removed they basically stopped doing that at that point I mean as I said Neil is still there but he's sort of moved like into research operating system research and at this point well we or the Hurt community believe that in the next couple of years we will move to another microkernel it's just not there so yeah by 2005 they were working on HurtNG and while Roland and Thomas had basically disappeared from general development they were still doing development until the early 2000s but it was pretty bad we didn't have a built-in running I remember I was going over my old slides five years ago and the status there was pretty bad yeah we had frequent crashes it wasn't so perfect and even then Marcus said when I joined the Hurt in 1997 there was little doubt about the Hurt state at that time it was just about at the end of a dying a slow death which is probably still the case today or at least at 2005 but then on those three guys showed up and they did a lot of work I mean it's going really slow because the community is quite small but Samuel Thibault basically rescued the Debian port and the GNU Hurt on it's on his own so that's just one of the few stuff he did M68K porters know that implementing TLS is important he did it I mean otherwise Gelip C is really difficult to have nowadays and he ported GNU Mach to Zen he was doing a postdoc at Sensor at that time I think and I think he also implemented PAA physical address emulation or what's it called extension thanks so that basically meant that we were able to run GNU Mach on regular IA32 hardware which could be as fast as we want because basically before then Mach wouldn't boot on AMD64 I mean the Centrino is about the last thing that works but everything else was just not booting anymore we had lots of trouble with the interrupts and the hardware was getting just too new and Mach was getting outdated so it was really bad we had to I was running a build team at that time and it had to be I don't know Athlon 800 or something like that anything else wouldn't wouldn't really work anymore so having it on Zen really was great I mean we could run it again and of course it was also great in so far as it was easily monitorable you could just reboot it remotely no problems you don't need serial console or anything like that you can basically you can also well debug it much better and yeah I mean he fixed countless of GNU Mach crashes I mean I would see on the build team that it's crashing and I just give the instruction pointer to Samuel and usually like 10 to 15 minutes later he would commit the fix to CVS it's just amazing so he fixed most of the GNU Mach crashes so it's much better these days he fixed and ported G-Lib C whenever Mr. Drapper implemented something new and didn't care about porting or portability that much Samuel would check in and do it he also fixed and ported XORG several times so X still works at least with the VESA driver I mean at this point we don't have kernel mode setting so there's some more work to be done if we want to stay on top of it but at least the VESA driver still I believe still works I just tried it and oh that's a type way he initiated basically the Debian installer port and so far as he like made some research on it and then got it to some limbo state for a while where you could work on from and also he maintains the auto builder I was doing it for a while but now he's basically the main guy I mean I'm backing up he's on vacation so I'm backing up for him but he maintains the auto builders he keeps them running we have a couple of them now and yeah he's doing Trimman's job he's like the one guy why Debian Hurts still exists so you have to kill him if you want to kill the port Olaf Bodenhagen he did a lot of community work basically he organized all this google sum of code stuff for the GNU site so we had I don't really remember but we had about 10 projects now overall over the last couple of years and one year he even managed to get the hurt being it's proper organization the other years we had one or two usually one slot from the GNU project so he organized all that he put a lot of effort in there he kept hitting people to do stuff and he also always is a lot of questions about what you could do what you could really do with the hurt because at this point we're just porting POSIX applications but in reality there is a lot of features which are still not really explored on the hurt all this user level user space stuff you could do basically could be tremendously useful but somebody needs to go out and have the ideas pushing the concepts and I believe he's also been working on a kernel graphics interface driver for Mach on his massive features but that's not integrated yet and thirdly Thomas Schwingem he pushed things on the GNU site so it was always very very difficult because the hurt and Mach repositories would still run on CVS on Savanna I believe and only Roland really had access well Neil and Marcus too but they basically had disappeared so it was very difficult to get fixes in because Roland wasn't there all the time and Thomas was basically not around I mean Thomas Bushnell was not around at all at that time so it was very tedious to get fixes in and Thomas Schwingem basically managed to get everybody going I think he lobbied also with Richard that there were no new committers now Wable so he got commit access for Samuel L so now it's getting much better I mean things are getting committed much much better much faster and he also did this cross-building stuff so people could more easily cross-build GNU hurt toolchain and in the end like couple of months ago he also migrated the repositories to Git so we're on Git now as well finally so yeah they basically became hurt hackers and now I want to talk a bit about the presence so that it was a bit the presence but the status so the community on one side it's very small so I don't know how many Linux kernel hackers there are there's about a dozen hurt hackers right now it's still a nice community I mean that was the main point why I sticked around because just the community was so nice and people like Neil and Jeff Bailey also who did quite some work for a while and Marcus Brinkman they're just lovely people and it was just great to hang out with them unfortunately they're not that involved anymore but still the community is nice people are great and also we managed that former Google sum of code students are still partly active I mean a couple have disappeared but quite some are still active so one of them even is this year he's a mentor and some of them doing new stuff and lately there's also two new porters Emilio Pozzuilo Montfort I hope I pronounced it accurately he's been partying a lot of GNOME packages unfortunately he's not here I mean you probably know him from there both dev and developers right Pino Toscano and him he's been doing a lot of GNOME work and he's been doing and this year he's working for the GNU project on Google sum of code for fixing well I'm talking about that in a bit but basically those two they've been partying a lot of GNOME packages KDE packages you know so it looks quite good I mean most of GNOME and KDE has been imported so the community is great and yeah I want to talk a bit about Google sum of code because this year I think it's really they're doing quite some great work so as I said Emilio he's working for the GNU project and he's fixing test suit failures so that's maybe doesn't sound so interesting but so far he's been fixing the GLIP test suit and basically what we did we were just labeling the test suit so we get the packages in the archive and we never really managed to have a really deep look at them but he's been doing it and he found lots of bugs he implemented new stuff like he found out where our implementation is not really POSIX or at least is well if it's if what Linux or GLIP is doing is beyond POSIX then he was fixing it up for them so it works and he had to fix and port GNOME and HURT and GLIP-C quite a bit for that he's now doing core utility and I hope he's moving on to Python and Perl as well still hoping he does we can get him to fix all the database test suits because they failed due to missing POSIX record locking there's a patch which is 10 years old but it needs fixing that would be great so we can only lock the whole file at this point we can lock regions of files so all the database test suits are failing Karim Ala Ahmed is improving the virtual memory management which is also great because nobody has really looked at our GNOME which is the fork that we are using for the last 20 years I would say or maybe 15 years at that level so he's working on that he's been doing some stuff I mean I talked to Samuel a while ago and he was really euphoric about it that it's improving quite a lot on the build demons he was a student from last year so this is a bit misleading he had an application but in the end he was rejected but he's still working on things with a mentor so he was motivated enough to still do this project even though he didn't get accepted for Google of some of code officially and what he's doing he's porting the DDE as a device driver environment I think it's from the University of Dresden or something basically better than the current glue layer that we're using I'm not that much into the technical stuff but the point is that we can use current Linux 2.6 device drivers unmodified with that and he's we now got at least the network drivers for Intel working so that that is working I haven't tried it on this one yet but I heard it is working so it seems to be great so that's that's one of the things that really always had a problem because there were no device drivers that were up to date the glue layer that we had in Gnumach from 2.0 to 2.2 or 2.4 or 2.6 at the time that new Linux releases were done and it was just a bit of a problem but now with this DDE it seems that we finally have a resolution for that so we can stop thinking about device drivers and finally Jeremy Koenig who is around he's porting Demian installer for Demian that's the Demian project and I'm going to talk a bit more about that here actually it's Jeremy here okay well he's around at DeppCon maybe not in the room so he did quite a lot of stuff I mean as I said we had a project about like 5 years ago and the guy just disappeared basically so that was a failure we were not very happy about that but now as I said Samuel did some some initial work on that and Jeremy has been picking it up and it works now basically so he added in it Rem this support to Gnumach which has been merged yesterday by Samuel and uploaded what a microphone Colin has the command that was always the blocker and I was trying to do this experimentally originally I could see how to port all of the components getting the damn thing to boot was the hard bit well I'm not sure how actually how they did it but he implemented it he had to fix some stuff for busybox like for example for some reason something and Demian also wants to make deer slash and that just failed because nobody thought that somebody would ever do that on the herd so he had to fix that by I think fixing the remote procedure calls and stuff he ported part man so you can basically partition things he integrated user space partition stores in lib parted that's quite nice so you can just use the partitions in user space you don't really have to think about offsets and stuff like that for busybox or he finished porting it I think Samuel did quite some work on that but I think he now has most of that stuff upstream and he fixed crap installer that was also one of the big problems all the time I mean maybe to reiterate how important that is so we had these installed CDs that Phillip Childs has been doing who recently resigned he's been doing it for ages and they were based on bootfloppies if somebody remembers what that was that was what Devin was using before Devin installed it so he was using bootfloppies on linux to boot it and then he was extracting base trouble basically and yeah didn't even well back then it would have probably been impossible but it also didn't have any grub installation feature so people had to set up that themselves and actually it's a bit more involved on the herd because you have to multi-boot a couple of modules so that was always a big problem and for a long time you couldn't really install grub from the herd and that was always going to kill the project basically you could just do it on linux and then because you had to run grub at some point but now he also fixed that so that's working you can check out the roadmap and yeah so I did an installation on this one it worked everything well the one problem is that it installs grub too and that works perfectly in QEMO so you can just boot it afterwards it works everything I mean you don't there's no hacks involved or anything it just works except on nature hardware it just rebooted but I mean I now installed it I installed grub1 on it and that that boots yeah as a grub2 developer I'm interested in looking into that if you wanted to get in touch with me sure couldn't do that and the rest of the status that means the install is working it was a report from Samuel a while ago which he filed so well there's a couple of things for example it just sets the QEMO static internet configuration so you can use it we can use it transparently in QEMO but if you want to do something else you have to set up networking yourself and stuff like that but basically it works you don't have to switch on the console and do some weird command or stuff like that it just works I believe Bach well Samuel has been dealing with that most of the time I believe Bach is usually building experimental and one of them is continuous building the failed packages in hope that some of them might have been fixed in the meanwhile because as I said Emilio and Pino are quite working on porting stuff so I believe the current number is 67% which doesn't sound so great but it's much better than a couple of years ago and it's really difficult it's not like you just I mean quite a lot of stuff needs quite a lot of thinking and working and takes ages to get these things going well it's going upwards slowly and we're happy but could have been could have been faster and the great thing is that with all the work that Samuel has been doing on Bach we now have about well according to him because he's been doing it basically the uptime of the built demon continuously building is one week which is much much better than what we had a couple of years ago and I had to reboot this thing every couple of hours and then file system check all the time I don't see any file system corruption anymore it's really much better we have to send Dom you working as I said it's not working out of the box I believe Samuel has been working on that over the weekend it would be really cool if we could just put it into the Debian GNOME package but with a bit of fiddling you can get that going there's QEMO images available and most of GNOME KDE has been boarded and built and there's a potter box available if somebody's interested in checking out their packages it's used to get your LDAP wired so you can just lock in with your Debian SSL key and it works I mean you have to probably ask us about the build dependencies and stuff but there's a change route yeah so the future get more packages get Debian installer fully working Jeremy I hope can do some more work maybe get it working on the graphical installer as well integrate Zen that seems to be working already and explore hurtish concept for Debian problems I mean that's a bit wishy-washy but well I don't know this whole user space stuff and translator stuff might be interesting for some things we have to rethink about it so far as I said we're just playing catch up but you could do quite some cool things for example we had some stuff where people were stacking translators and using network translators to talk to each other and stuff like that you can do quite cool stuff with that if you just think about it and have nice ideas well what really the problem is right now and what also killed Phillip Charles with doing these snapshot releases is it's just unstable in Debian unstable so he was always having much much problems getting everything in a state that he could freeze so he was basically doing release management on unstable and trying to get a set of packages going and build CD images and that's why he burned out basically because it's just so difficult so we hope to get at least some kind of testing after the squeeze release obviously because right now we're keeping up quite well I mean it's not like there's like hundreds of packages not being built so the auto builders are not a problem out of date packages finally I'm trying to get that going and then try to get some testing going probably we need to dispatch somebody to release team if they want to deal with that kind of stuff so it shouldn't be the release teams issue but if they are okay with it we might think about it and well I don't know what happens in the end we will see so that's the end of my slide so far I can try to show you how it's looking if I'm getting this right that's not good so yeah that's can you see that that's the the installer booting natively on the herd it's a well it's a CD image and that's called booting I don't know whether you can see that but it works basically showcasing so while it's running is there any questions yes hey my name is I'm fascinated by the project and I actually have a boatload of questions but I'll just start with one don't take this the wrong way this is going to sound horrible but what's the point I will accept because it's cool because we need another free alternative but what's the motivating factor actually is there a Minix talk I mean Yalla has been doing quite a lot of work is he presenting that I was faster than you I got it sorry yes I'm actually also working on a port to the Minix micro kernel operating system as well and my motivation is basically because it's there so okay go on please so I'm not particularly a hard developer except in that I occasionally make something for it when I notice it but but I can offer one useful reason why GNU hard to continue to exist even though most of us are using GNU Linux and that is that there is considerable pressure on there is considerable pressure on GNU Linux to remain stable neither so many people are using it and it's becoming quite difficult in many ways to do serious research you can't any time you try to make an extensive change it breaks somebody's real world situation and I think it is healthy for there to exist I don't really think it matters what it is but it is healthy for there to exist a system which has enough traction that people are working on it and keeping it going that's sort of a minority interest so that it has the freedom to do more way out to research and everything else and I'd offer that as a reason for the project's existence that's a quote from Linux when he announced Linux it says do you pine for the nice days of Minix 1.1 when men were men and wrote their own device drivers are you without a nice project and just dying to cut your teeth on an OS you can try to modify for your needs are you finding it frustrating when everything works on Minix? no more all-nighters to get a nifty program working then this post might be just for you so yeah that's basically a point trying to check back on this there we are and actually the cool thing is the Herk console does U2F8 so you can basically I don't know actually can you do that on the console on Linux? we have to use the user's best console emulator for it so Jeremy was that because of the micro kernel architecture that makes it easier somehow to do development work? yeah I was asking is your name Jeremy? no I'm Colin oh Colin I'm sorry like I say I don't think it matters much exactly what it is I think that the GNUV K3BSD port serves a similar purpose in some ways that is an experimental system in some regards but I do think that it's healthy for the Debian project not to restrict itself simply to GNU Linux yeah right so one thing to always keep in mind is that when you use the space part this whole I mean I haven't really talked about that all the porting issues all the porting issues we have like there's no pathmax macro defined on the Herk but that's not actually the Herk I mean that's the GNU project it's the GNU system in that sense and our Debian port we always have to think about well does it make sense to continue working on GNU HURD or does it make sense to continue on Debian GNU HURD that's like two different questions and um yeah I don't have why I don't have I don't know anyway okay going well so one thing is that as I tried to say so in Linux these days it's really hard to get big changes in it's difficult that's also why I mean we've been talking with the new guys and they say it's just great I mean you can hack around in the kernel and you immediately see any progress because it hasn't been optimized in the last 20 years so everything you do basically might get a 10 or 20% boost in system performance I mean I don't know whether you want that but it has some instant gratification to it and while Mach is not that great from I mean it's a microkernel so it's not that easy to hack on the HURD itself because it's a really nice code I mean it's the same thing like the GNU C library and you can do it in user space it makes debugging quite a bit well on one level easier on the other hand all this inter server communication also makes it quite a bit more difficult but yeah it's hard to say but I mean for Debian I believe it's just it makes it a bit more portable in one hand we have manpower we have people who are knowledgeable enough to fix things in a way that they just don't define Pathmax to 4096 or something so it works but usually what you have is what you rather do I mean you have like a final name or something and then usually you would just allocate a static buffer of Pathmax to be sure and if you do that all the time it's just like waste memory and we have people who can now allocate memory dynamically at the right size because usually you know how long your buffer will be and stuff like that so it might also that the code in Debian or in the upstream projects might benefit from that quite a bit sorry it was well well and you first is is SMP working on herd or is it not relevant to no we unfortunately that's not working I mean so I mean herd in general is multi-threaded multi-threaded but SMP on the Mach is not working at this point I mean I think we had to I'm not sure about it I don't believe it ever worked really I mean there was one guy hacking on it for a while and he got an experimental thing going which I believe went in the right direction but he didn't really continue so no there's unfortunately no SMP that's certainly one thing which is problematic so where can we get that CD image it's a good question this is working it's on Jeremy's website photo time being probably just should go to that there it is is it? yeah no that's Samuel this one there it is mini mini.iso jk.fr.eu.org Debbie and herd installed on mini.iso and then just runs in QEMO other questions you had one yes yes one problem I've been having with the minix port is running into like linexisms in software apart from the pathmax thing you mentioned what other like major porting hurdles do you have when you're bringing software signal stuff that's because our signal that's because our signal implementation is quite old so Roland said has been saying it for 10 years or well until he was not actively working on it anymore that he has to rewrite signaling stuff but he never did so that's one thing then there's a couple of these well the main point of the GNU system is of course that there should be no limits so that's the whole pathmax isn't defined the whole max host length isn't defined stuff like that I mean that's a double-edged sword on the one hand everybody likes the GNU utilities there are much more well some people like the GNU utilities they're more featureful than the old BSD ones or proprietary unix ones but on the other hand due to linux going with the classical unix kernel it also means that other stuff in GNU is just bit of a hassle on the hurt because as I said it's a policy question of the GNU system to do that it's not a problem in that sense other things there is a website called porting problems on the hurt wiki I believe so there are some open issues well you can find it somewhere basically signal stuff then well lots of other stuff we have fixed so far I mean the most things I had actually I didn't have any analysis on all the failed stuff on the build even a while ago and quite a lot of what this path makes stuff and then it's just like random well of course there's a lot of a lot of programs but they have to really say that the k3bsd port was tremendously useful because a lot of programs will just include linux slash fs or some header which is linux specific because they think well it's linux I'm not caring about that it basically has the same issues and there are now release ports so things are getting actually fixed and then catched up on that so we're much better doing in that and then there's a lot of like glibc not defining one part of the structure but it's defining it so for example our networking stuff and glibc is more modeled on bsd because that was what the people knew at that point and linux wasn't I mean even when it started linux wasn't that big and stuff like that so yeah is there any further say are the major impediments left to having a usable system for a user so for instance like drivers I would imagine it's a big thing network drivers wi-fi storage devices yeah I wanted to use her on my laptop wi-fi is a big problem we don't have we don't have sound there was one guy Richard Brown he managed to get sound working for a bit but he never integrated it so there's no sound in and hurt there is well now we have the network drivers at least so I'm not sure how easy it will be to get wireless drivers ported but there was one guy called Stefan Siegel who ported PCMCIA so basically with the old PCMCIA wireless card you could have working wireless but okay that's been a couple of years so you can't use wireless on that and then there was a recent thread about what's missing and for most people it's that USB is not working so I have to really have to say at this point I think it's much more a developer system in the sense that developers can use it they can use it in QEMU or on the boxes or even natively to hack on stuff but there's only one guy right now I believe or maybe a couple but one guy that's Olaf Bodenhagen he has his home directly NFS mounted and he does most of his work on the herd but if he wants to listen to music or kind of stuff or basically watch a YouTube video or something then he has to well do all the boot and that's not so great but yeah there's quite a couple of things missing so I don't recommend it running as a user system it's really for hackers I would say at that point okay I think we're running out of time is there any last question otherwise thanks again everybody for attending and yeah I hope at some point yeah it will get better or even better I mean I'm quite glad at the current state it's much better in the last year or so yes two years it's getting very slowly because it's so few people but it's getting better and better and it's not I mean at some point we were thinking that it would just starve because we're moving so fast and we can't keep up it seems to be that we can keep up and we're getting better and also we're getting it's I hope I mean the main reason for me is that I hope that it's getting less trouble for Debian so it's always like oh there's outdated herd binary so oh this is not working on a herd and I hope that that is getting also better so it's less work for Debian other people we don't really want to hassle other people in Debian with a herd point okay well thanks everybody