 Hans up, who here already runs Debbie'n, or Ubuntu or related system on arm? You guys probably know as much about this as I do then, that's good. There's been a huge amount more discussion in the last couple of months on the on mailing list than in the last few years. I assume that again probably most of the people here have already been watching that. Mae'n meddwl? Mae'n meddwl. Mae'n meddwl. Hector, rydyn ni'n meddwl? Rydyn ni'n meddwl. Mae'n meddwl. Mae'r document erbyn yn gofio, erbyn gofio.debian.net. Mae'n meddwl. Rydyn ni'n meddwl. Mae'r cyntaf. Rydyn ni'n meddwl. Rydyn ni'n meddwl, dwi ddim'n meddwl, i'w meddwl, o'r cwmence, ..a'r ddweud o'r amser o'i gwybodaeth. Rydyn ni'n hoffi'r trach, o'r hoffi'r trach a'r hoffi'r trach.. ..yna ni'n hoffi'r hoffi'r hoffi'r hoffi'r hoffi'r hoffi'r hoffi'r hoffi'r hoffi'r hoffi. Yn y gweithio? Yn y gweithio eich cyfnod o gyfnod dda, mae'n unrhyw o'r gweithio? Felly mae hi wedi'i gweithio bob amsgrifion. Mae'n gweithio, ond mae'n adrwm ddangos ei gallu bydd yma yw ddweud cyfnod yma. Mae'r proses yn y pethau'r gwybodaeth, yw'r cyfnod yw'n ddweud cyfnod yma yw ddweud cyfnod yw ddweud cyfnod yw ddweud cyfnod yw ddweud cyfnod yw ddweud cyfnod. Yn ddechrau, mae gynhyrchu yn mynd i weld i'r cefnod ac mae'r cwm yn ein bodi'r bobl. Felly, mae'r architechur yn enwy. Mae'n siarad iddo i'r ddechrau, mae'n gweithio. Mae'n gweithio chi'n dda'r cyhoeddaeth i'r adeiladau. Felly, mae'n gweithio arweithio i ddweud y gydag. ymnym i gwaith llawer o'w'r cyfarfyn i'w gwylio sy'n rhan o'ch gwir procedent yn gyrach arhos. Mae'n gweithio eu PROS. pan aeth ni'n wneud. Felly, rwy'n gweithio. Rwy'n credu yw hwn yn cyfeirio. Rydw i ddim yn rhoi gilydd. Yma yw'r hyn sylch. Mae'r rhaglen o'i good chyf meth. Rwydw i ddim yn ymwy Silk, neu mae'n fideo yn y dda. Mae yna du'n fydig prIG, mae'r rhaglen o'i gwyst i'r newydd, online sometime either by the end of this year maybe only next year hopefully v7 year that's not official it's not a done deal but especially if you guys all give a nice housing cheer then I can go back to my management and say look people really want this stuff right if in theory one wanted to contribute for example a beagle board that one had to a arm v7 built or a port how would one go about doing that in theory get the mic please the big part of beagle board and I've you I have on myself is there's just not enough ram even with and with usb based drives it's too slow I am building open office or in our large package can be measured in days or weeks versus something with 512 ram can do it in about two days flat I think 512 is the reasonable expectation for our v7 port with email memory because that hardware is coming available there's the beagle xm has that much has that publicly released yet I think so there's an early spin yeah oh okay so my information slightly out of date so beagle xm would be acceptable yeah absolutely now of course the mobile boards that we have these on v5s they have a gig and a half of ram on board I don't know if it's all Courtney in use I may have screwed up building the kernel for them on the rally showing a gig obviously that's something it's something that we found over the last few years on buildings is memory is probably even more important than CPU if you if you are going to build open office or firefox or or something similar if you're trying to link in swap you've lost and they have satire as well which is quite important absolutely they have satire on board yet another problem with beagle boards is that they only have usb which makes it like 30 megabytes per second which is quite important you need many of them if you want to do that the mobile boards we have at the moment which Marvell boards they're one of the dev boards I can't tell you off the top of my head okay because yes okay yeah okay because there's some boards that shall go nameless that have the SATA data thing on the USB box which goes very interesting lance issues we're not using any more abilities the other thing we probably should need to take account I get maybe getting a little have ourselves for the so you need to decide how we're going to handle devices that can't help the installer specifically I know this is true for some of the new omap stuff coming out I'm not sure it's true for the XM but most of these boards have a single SD slot and that is their input output we saw this problem in Ubuntu by implementing pre-installed images which is basically what exactly what says in the 10 but Debbie installers borderline useless and parted is even useless because of the special partitioning required by the beagle board because parted was written in an error where CHS partitioning was already obsolete yeah that you could fix a parted probably to be able to handle mtd device and it's been somewhat fixed I mean that's partly been done last year and that's something that needs pushing in so Debbie installer actually works but my understanding is that we've done most of the base yeah after Beth did did somewhat specifically for doing MTD base to my knowledge the XM doesn't have NAND anymore that went away oh it's just totally SD to all the SD in the USB port yeah that's not sure how much I can say but that seems to be the standard on quite a few boards so I think we may actually need to come to the possibility that Debbie installer is not going to be useful anymore for these type of boards and may we may need a new type of image I mean that even the net boom installer can't be directly used right now due to the requirements of partitioning that parted okay someone wants to say something why can we install to SD yeah we can generally because daddy installer has an issue of installing over itself that could be fixed the second problem is that parted blows it can't be used on these devices because of the partitioning scheme at least an omap okay so we need to teach parted how to do these that's not ins ins ins amountable surely yeah there's certain architectural issues of part that makes this difficult but yeah it could be fixed okay thank you so many well from parted side I'm calling are going to to add the patch the patch that has been worked for AMD support so it should be fixed soon for parted itself and after that install should come with it cool right well could you note there some somebody that the devian installer issue so we can yeah lyrics during it thank you then this is the official army help or I think we have the building is ready we have porter boxes we might need to work on devian installer to provide ways to install more more devices and be able to handle external kernels because there is also a lot of kernels that don't go into mainland so maybe devian we don't want to support those kernels but maybe we should have a way to easily yeah one of the things we inherited in in bwntu when we bootstrapped our on port was essentially every board has to be whitelisted in flash kernel and lab devian installer and what's currently being implemented in bwntu and I'm planning to push back into devian although it may not land until squeeze releases because I didn't anticipate the freeze to happen is we implemented generic architect sub architecture support basically if the board can boot with a stock kernel it will work without us having explicitly whitelisted there's a blueprint for design this probably won't land until devian until Sid is usable for general uploads again but it will help resolve a lot of our issues because it also defines a mechanism where we can essentially add a new sub arch on the fly if necessary that part's not completely baked yet but I'd be glad to throw it on to devian arm and let people smack play with this back car enough so that sub architecture is within the kernel configuration is basically what you're talking about there well okay gang to di guts which I didn't want to do essentially what the installer currently does is it looks for a specific sub architecture by looking at proxy pu info and seeing that board exists in its whitelist and it doesn't it does army L unknown and there's no code to actually properly handle no kernel and no flash kernel so the installer actually will error out after a while on something I've liked to fix but never been high on my radar what we did in bwntu is if the there's no entry in the whitelist it froze up a big scary warning and then continues it looks at the kernel version string and tries to pull the sub arch out there because of the kernels built correctly at both debian and bwntu it'll be version number hyphen sub architecture so we figure it's a relatively safe assumption that if we can boot a kernel with a sub arch flag it is that sub arch I mean granted a user could break that pretty badly we also implement most arm architecture we's we've done both this is true for omap three dove and omap fours we've implemented a generic boot mechanism so it will work across multiple boards so you could swap out the custom kernel sorry you could swap in a custom kernel flash kernel will still do the right thing and then you just have to swap out the installed binary there's still some places where this needs to be fixed but as we're also getting dev tree support coming from lanaro this situation is getting less and less painful but still have to deal with for now so that's actually a good point to dive in because you I'm hoping that most of the people here have heard of lanaro anyone not heard of lanaro wait wait for the mic yeah well in fact I'm not going to give a description I'm going to hand over to Wookiee all right okay to give people a bit more information lanaro is a project to make all this stuff that's a bit of a problem in arm work better essentially so there's better support for current CPUs better support for different flavours whether you've got vfp or not and neon processors better support in the toolchain for all this fancy shit and so at the moment at least most of that work is operating around to be low level toolchains kernels getting more but an awful lot of kernel support which exists in various manufacturers trees but isn't mainline properly and try and make that situation much less broken so that's all good there's also going to be work on things like all the new graphics GPU chips but they're all non free so that's going to be a right pain in the ass but at least open GLES and the graphics stack and all that stuff should work better because we're all getting shiny fancy fast displays with our shiny new socks which is obviously interesting so yes there's a great deal of stuff going there they're spending serious money which is quite good you know employing me and him several people in this room so yeah that's basically that and we can use quite a lot of that stuff now exactly how we're going to use it is one of the things I guess we want to try to work out today any any questions about that yeah okay I do I have a couple specific questions um traditionally it seems like there's a lot of interesting companies doing support one of the maybe code sorcery I seem to be doing a great deal and in GCC will that stuff become free software these are ready so good so three releases twice a year and that was a release is completely open source in terms of toolchain they have proprietary software as well but what's in the toolchain is free software and they also work with Leonardo so I I think three for four concentrate engineers working on it now right now about to be five and it's about two third of the toolchain working group in the narrow which delivers toolchain and we integrated their toolchain tree into the now toolchain tree so that it's permanently public because their tree was not public until then only their terrible releases and they now is ended up doing a release every month of the toolchain so that you get more continuous integration and better stuff I think he was trying Leonardo is up already free software and and the code sorcery but I think you mean that is this going to be integrated into deviant beats probably I do mean that as well yeah so that's a completely different question so in integrated the linear toolchain in the buntu GCC because it was discussed at UDS and there was a desire to have it and because they now bases a lot on a bunch of but I know that the toolchain is also being used for the harm outflow port which is the the next point on their gender it's been used because it's the only toolchain which was supported and and I mean we actually intend to sport supporters in fixing bugs in the toolchain whenever there are two chain bugs and at the same time it's the only one which also supported our flow dbi right now so we we intend to sport debian arm outflow port for that but I think Matthias wasn't sure he wanted to use the linear toolchain on all architectures it's is there is a desire to use it on some debian architectures and I don't think it's very complex technically it just makes it harder to forward issues upstream you have to forward them to now instead so perhaps that's a question for Matthias perhaps do you want to do you want to express your opinion on the you know diff in in debian and why it should be included or not or I can proxy for you if you prefer because we've had this conversation so Matthias's comments to me were that the the linear toolchain is is great and happy and the most awesome software ever but as far as its use in debian that's that's pretty much a decision for the arm porters to make who are people in this room so if if if there's a thought that the linear GCC provides benefit to the debian arm port the the impetus for that needs to come from other people in this room and not not the toolchain side of things it's it's not something that Lenaro or the debian toolchain maintainer any of us are interested in enforcing debian arm to use this is Tom marble I just wanted to let you guys know that I'm on the open JDK IRC chatting with Rob Savoy and Andrew Haley and they say hi okay I'll tell them hi back Rob is saying that he was trying to get on a bigel board exam but he had some kernel troubles and he had to drop back to the C4 which is I guess the general bigel board also said that there is another board the IE GSV 2 which apparently has some good support so I just thought I'd share those comments I guess the question really on toolchains is do we think there's breakage in the other architectures if we start using all the stuff in there the the Lenaro toolchain you know are we going to break anything elsewhere in debian that's the big question do we have a faint idea do we just have to try it well I have supervised a lot of the rebuilds that have happened since the Lenaro toolchain went live and then went to it's generally helped across the board fairly well we had a couple of rough spots right when it initially landed but those got fixed relatively quick basically we were getting I think it was an ice from the kernels building but other than that we didn't have any major issues on the flip side while the Lenaro toolchain source has you know support for all architectures GCC does I don't think anyone's ever tried to build it for anything short of arm and I3A6 so no doubt we've probably got some bugs working out there it may be a good I I personally think that if it's feasible we should move to the Lenaro toolchain I find it's better maintained than well I should say better maintained I should say it's got a faster turnaround time for fixes and it's also resolved a lot of an archive ice internal compiler hours so I think it may be prudent that we bootstrap compiler for several architectures that are fast probably I3A6 AMD64 arm powerPC anyone else want to throw an architecture at me spark itanium built non-dead architectures and just do an archive rebuild and see if the end results usable I figured that's the only way we're going to actually know if we're going to stress test this architecture for architectures that are not fast enough for an archive rebuild I think we'll have to talk to their porters in advance and see what we can do see what we can do but if we can get it working on six architectures and the six I named could probably rebuild stable or on a snapshot unstable and probably about on a week two weeks that would give us a good basis to go off and it's not particularly difficult to rebuild the deputy an archive of a new toolchain I know we've done it with a lot of inbun to we probably use the same infrastructure and I know they're specialized tools to do it yeah I know Martin's been through trying with you to see snapshots of various points any comments some comments about the different GCC versions currently we have three well branches of GCC that's the FSF branch the upstream branch we do have the code sorcery branch and the linear branch and at the moment the linear branch and the code sorcery branch don't defer that much for 4.4 but that will change for 4.5 so the narrow is not identical to code sorcery and for 4.5 I assume we only will have IX86 and arm changes in the linear branch no other changes so it's a bit difficult then to use this kind of toolchain which is only tested and only developed for arm and eye 386 namely 64 for other architectures which are in Debian so that's my concern I which well well I do not want to to have that kind of toolchain for all Debian architectures would it be possible the question remains that we could use the narrow toolchain on some architectures but not on others it's okay for me to do that for for one architecture like like arm but I doubt it will benefit Debian to to use the narrow toolchain for for Intel because that's the platform we use the most and it's the reference for for other ports so having something different than on any other port maybe except arm I think it's it's not an option sure yeah I think in long term the best option would be to linear or to concentrate in getting the changes to upstream GCC as fast as possible and Debian to keep using it whatever we're using for short term doesn't matter we can usually narrow for the hard flow port getting it booting it booting it up but on the other hand it doesn't really need a much of discussion anyway okay my big concern with multiple toolchain packages is this is quickly going to lead to the path to madness especially if we are building different packages on different architectures plus that's going to require adjusting some of the base dependencies to pull on the right toolchain and it just gets very ugly very fast right now we are committed for squeeze we're not changing the toolchain the closest that this is going we're probably half a year to a year away before we'll be able to upload it before we'll be able to change toolchain so but it may be prudent to build take GCC Lunaro and throw it in experimental so if we have something and then it's available there as a package and for Debian ports where I'm assuming we're going to be bootstrapping arm FL it can pull it straight off experimental pet peeve of mine nothing that you're talking about would imply that GCC dash Lunaro itself would be unreleasable so it should go in unstable instead of having the overhead of trying to do stuff in experimental right okay I guess move on from toolchains obviously we can carry this discussion on afterwards so there's been quite some discussion about exactly what arm CPU features we should be supporting I mean God forbid I remember being in the equivalent discussion to this three years ago which is when we actually decided yay let's go for army L let's go for V4T because that's well almost the future there's been some interesting discussion since some private some public the V4T is a reasonable base target for lots and lots of software it supports the vast majority of arm hardware that's out there in the real world today but of course on the significantly newer hardware it really really is not optimal there is a quite a lot more performance to be had by optimizing for the newer hardware but of course we then have the difficult decision of well who do we leave behind thoughts two thoughts first while I definitely agree that the performance benefits from having a new port will be worthwhile we needed to realize that quite a few prod boards and processors have shall we say flaky implementations of some of their features and this has hit us pretty hard before in boom to answer and developing boards which have to remain nameless of course yeah but then the vendor will rip my head off and I like my head second point second point is that I remember as being conditioned of the army L port gang bootstrapped is that the arm port had to go away eventually yes I think it's a bad idea if we get rid of the army L port because they're still going to be armed five devices under in the market and I don't see that no one at this point is suggesting that we throw away army L or move it irrevocably forward too far yet there's been discussion about having a secondary unofficial port which may become official later on of course especially mean we've only just commissioned a whole load of V5 build these things like the guru plug the shiva plug are still I think V5 T it would be a really really bad thing if you know especially after the discussions we've had this week about yay freedom board freedom box and what you mean Debbie doesn't support it anymore just to get back on your FPU bug I think even the buildings which we have a stiff FPU bug which is just a cowl patch yeah no they're not the same board they're all the mavels and they don't have some too they're not the same anyway that's a cowl patch we can discuss later what I find great about having two ports that we can have different requirements so we can have a fast port and a slow port which is more compatible and it's kind of similar to ice 86 and MD 64 where you can run ice 86 on your on your MD 64 and it will support very old VR CPUs or what not but modern laptops have all moved to 64 bits and I really can run the fast port and in the same way I feel the market is kind of segmenting and there was always multiple options but we get the the newer hardware which is rv7 and it's more expensive and very often comes with a FPU and it's in what's in an A-class CPU and then you have the cheaper rv5 which are going to be produced for a long long time still to come because they are so cheap and they are produced in great masses so it seems sensible to have kind of two ports like we have for ice 86 and MD 64 however the side effects not to be it's not obvious that the current state of the toolchain gives us performance benefits on our outflow however the fact that we would move to for instance v7 or something like that would give us some significant performance benefit and also lost my train ourselves because I've lost sight to the projector we can we can access it afterwards but yeah essentially I'm happy if we have a two oh yes the other thing is for ISVs we try to produce binaries which target ARM it's going to be an issue if there are two ARM ports widely popular and used in devices and they don't exactly know for which one day they should build their code so I mean is having two ports the best way to go it seems very likely that I don't think we can avoid the fundamental thing is I think is is vfp support is the thing that we pretty much got to have and then probably you might as well have arm 7 while you're at it and then there's a difficult question of do we want to make it hard flow or soft flow and thereby keeping it compatible I know whether that really matters or not if it's a whole new architecture I guess it doesn't really matter so the question is merely is it faster but the other thing I think that's actually useful in discussion is we need to work out what to do about so architecture stuff it will be nice if we could do some of this stuff without rebuilding the entire archive in a new way because it actually only applies to a relatively small subset of stuff 90% of the things people want to do and I guess Steve has opinions about that so I think one of the things you said was actually precisely opposite of what you should have said you you said that if we're doing okay I'll stand up if you're if you're doing on a new port anyway then you can make it incompatible I think if we're if we decide that hard float is the right thing to do then it's a new port if you don't need to do hard float then you don't need to make it a new port and therefore shouldn't in order to not cause incompatibility absolutely so the the real question of of of do we make a new port or not is do we need to make a new port in order to get the the ffp benefits that people are looking for it does does hard float make enough of a difference and is that what people are after that seems depend on who you talk to so there is another option which we didn't really name but the concept of sub-architectures or having a rebuild of RML faster with different optimizations is another option which is for which we don't have a name because we never did that in Davion I think the difficulty with doing that as a sub-architecture of RML is the fact that the calling convention is incompatible on across an arbitrary set of libraries yeah sorry well exactly that's hard float is an incompatible calling convention yes but what I'm saying is that we have another option which is to have another army archive which is built with high optimizations like vfp v3 and Rmv7 or something like that just like just like Ubuntu archive and that option is something which we never did in Davion but it might be the most sensible thing if we want to benefit from new architectures but without having incompatibility between ports so ISVs would decide whether the the optimized form to be supported on any device or whether they optimized to be to get the best performance and work only on Rmv7 and their packages would not be portable to all the devices but they would work on all new devices and they could produce a single package which targets whatever market they want to target and Genesi hello this working yeah Genesi has provided Davion developers with Rmv7 hardware they've donated 14 boards for developers and and five to six for a cluster for build demons and we Constantine of Margaritis has already been working on the hard floating point and we already have like a bootstrap of the port which is running here in this device is this one and it's a it's located at freebag.org slash repository you have minimal tarbol you can play with it make benchmarks and just send them send them to the devian arm mailing list and we're going Genesi has also donated two disks which are two terabytes for devian ports so we are going to install those disks into devian ports machine our alien is going to do it but he couldn't be here today and then we'll start building devian for hard float support so it's just a matter of deciding if we want to adopt that in devian and maintain two ports officially or just keep with our RML port so a very useful thing for I guess for anybody here is if you have your own applications which obviously you have currently running on RML please try and get hold of one of these or grab somebody else who is and do some benchmarks with your code because obviously the best benchmark is the real code if we can find out exactly how much of a performance improvement we might see if it looks to be like you know there's a 50% performance improvement or something then you know I think the general feeling is probably we will go with this if I'll average however it's more like 5% really it's not worthy effort I plan to set up border boxes for our main chef so people devian developers can play with this port and make their own test and benchmarks so please send the send results to devian arm mailing list and also it'd be nice to compare this is our Cortex A8 but Tegra board is Cortex A9 the pipeline has changed and it's there's not real this we don't know if it's real benefit or not of maintaining this port or keep with with something similar what Ubuntu has now with this soft FP which is compatible with with current RML and maybe we have other ways which we could mix libraries and with HW cups way or I mean this this is on the Bobby document there is a question about about that that how could we mix and support optimised libraries about that so yeah as you said the two catches when you testing that hard-float port with your applications first it's depending on your CPU it might be a much faster or not so Cortex A8 is known to have a slow of the FP I don't know if it's still loading or something like that so as result it will benefit of a huge speed up on to use hard-float which is probably not same speed up as on A9 class or other ARM CPUs so we have to be careful about that that we are aware that it's only going to affect A8 CPUs that the benefit is only on A8 and the other thing is the current devian apport is ARM v4t so it's really old so you're going to see a huge increase anyway the question is whether there is a difference between soft-float and hard-float that sort of things a major question yeah so how can we come to a conclusion on whether we want a port or not now there might be some slight confusion here about soft-float and soft-fp and hard-float there is basically a way of using the fpu with still using the old soft-float abbey in this case we move the fpu or the float arguments from the fpu registers to the general purpose registers when we make a function call this way we can keep compatible with the soft-float port it is somewhat slower than having a native hard-float abbey in which case the arguments are just moved in the floating point registers now of course we've had suggestions for doing the for doing the optimization for new stuff of doing just simple optimization on the libraries that people use say like we at the moment we have a libc-686 package you know we could go through and just have specific so that doesn't scale we try that in in a window and it's only doable for a couple of libraries but you have so many issues with that first you have to you have the human time of actually changing the packages and making the builds slower and more complex and then you actually have to make sure that these packages are pulled at installation time so that an app to get installed would actually pull user the vfp version or whatever and then you have the problem of actually space this disk usage simply because you might be using you you're only using the vfp version but you still pay the price for the non-vfp version or vice-versa so in any case it's um it's usually not very nice unless it's for something like a fan pack which is going to be huge anyway and it's on desktop system something like that but in general it's not scalable it's only rare cases probably less than 10 packages where we can do that okay but I think the concept of having alternates that you could install I think is something we should consider I think if yeah exactly I think you know this problem is only going to get worse we have a lot of options and you know if you actually want to be able to supply what people want to use you know we don't need every possible conceivable combination but we need at least two obviously and you know maybe more than that if people start producing things that have got neons but not fps vfps or you know there's a lot of things I've got vfp chip but not a neon chip and so on and everybody wants to do fancy graphics on their boards all this stuff matters so I'm not entirely clear I guess Steve might be the best person what exactly will be necessary to make it possible to have you know the equivalent of m player I686 but without saying that we just have different m player flavours and you know part of the multi-arch support we've been working out gives you some of that capability but we also need deep package mechanisms and you know so oh so you you don't like my idea of having the arm EL-libs package that's a 600 mc tarball that we upload every time we need to change it maybe we can think something better than that multi-arch is something we've thought of that's better than that but as far as the actual mechanics for what you're talking about I there's so much up in the air right now I hesitate to go into a bunch of detail honestly yeah once we get multi-arch in place then we can start looking at options for I guess I guess we'd like to move in this direction but you know there's quite a lot of infrastructural stuff and you know there's things that the narrow might do for us and there's things that a bunty might do and you know if we can all avoid running in opposite directions which I think in practice is going to happen because there's quite a lot of people here involved in all three of those things so hopefully that will just kind of work but in the meantime we basically got to have a new port because that's all we can do right so I think you have you make a very good point in the timeline of things we have to think about how much time it will it will take us to implement something and at which point it will be usable if it takes us three years before we can use some architectures and it's not worth investing because we probably will have moved on to something else at that point so multiarch has taken a long time to be designed but it's about to be rolled out and I think the design is actually converging to an actual implementation now and that's really good. Doing a hard float port is also something which we understand I mean we understand how to do a deviant port and even if it's a very time consuming process and I'm very heavy and I don't know it consumes a lot of disk space and it will take more images and it's going to be hard to release with one more new port but it's something which we know how to do and I very much like the concept of subarchitectures but nobody has a plan for them yet so I question whether it's sensible to go in that direction if we don't have any idea of how we want to implement them like for instance where they store in the pool, how do you pull them on the system, whether they are a complement to the existing packages or whether they are completely replacing packages so subarchitectures is a nice concept but I think we are too far from an actual design that it's reasonable to bathe on that. Does someone have a design for subarchitecture? The other problem with multi-arch especially with this is you're going to need a unique triplex for this to work properly and ideally we need to change the first part which there is support in upstream autocomph for actually doing. The autocomph libraries essentially, okay I'm sorry I haven't had time to follow the mailing list discussion. Yes there was quite a there was a little bit of bike shedding on the mailing list and eventually people agreed on a route to go for it. There are two different discussions one is a triplet for Halfloat which is one discussion and I think that one is settled down that we're going to use the same triplet which is weird. However on the multi-arch side we decided that new triplets were not suitable for using in pass names and this has been discussed over the last week at that comf and I think Steve is a better person to cover that part. Man you're not going to let me get any hacking on this are you? So very briefly new triplets are not suitable because we figured out that they do the wrong thing for Heartflote in the sense that they do not accurately define a self-compatible set of ABIs so they're not suitable for putting all of the libraries together in that directory because we may need to be able to co-install two different directories and it doesn't make sense to call one of them to be like an HW caps version of the other because it's not HW caps it's it's not an optimization thing it's an ABI difference. So what are you going to use instead? There is an action item to get the LSB to do it for us. So and actually it's not just ARM that has an issue with triplets we have a similar but slightly inverted problem on x86 in that we use I486-Linux-Gnu and Ubuntu previously used I586-Linux-Gnu and has now moved to I686-Linux-Gnu and all of these things should actually be the same path and since you know other distros outside of Debbie and Ubuntu are going to look at I486 and say well what's that for anyway we need to have something that's actually persistent across time within each architecture family and so yeah there needs to be an upstream standard we are going to float a proposal to the LSB as soon as I can get around to implementing a proof-of-concept tool to spit out the names and hopefully make some forward progress there. I guess a related point to this is as like suggested we could just build the same thing call it army L for different options but we don't really have any mechanism to control that effectively that's what we've got Ubuntu so army L is built to a different set of options a different higher base effectively than Debian's and the only problem with that is that people randomly install things called army L and they don't work certain amount of breakage might happen and you know that's where we're at already. It would be nice not to make that even worse. Anthem 2 of course which hasn't actually been mentioned but you know that's a similar issue yes I'm not sure I have a point to make but I guess it would be nice if we had a mechanism to control the you know the compatibility the hardware capability options basically and we haven't quite got that done yet but it's going to become urgent I suspect if people are building lots of random repositories all called army L that in fact have different requirements. Does everyone know what Sam 2 is? So on army Sam TPUs and I don't even know all the CPUs but on army Sam TPU you have a parallel set of instructions which is 16 bits I think which is Sam 2 and it allows having smaller code which does the same thing as ARM code otherwise and it's turned on by default in Ubuntu and we've noticed that it would usually give better performance because you don't use as much CPU cache or memory and so or even disk base and so it seems to be a good win to use Sam 2 by default that's only in army 7 but in army 5 they also have Sam 1 mode which is basically the same thing but it has some drawbacks because it doesn't have as many instructions but if we do an army Sam port I think would be a good idea to do it some 2 unless someone has compelling reasons not to. Yeah and just to add even more confusion of course you can interwork which is specifically you can have some functions in your code written using ARM instructions you can have some of them written in some instructions and you can happily switch from one to the other you know on-the-fly and this scares people to only debug code. I think ARM's general belief is having done this thumb implementation twice now the second time around they've worked out that fundamentally it's just better than ARM mode and I think they'd quite like everybody to move to thumb 2 eventually that's the way things are going to go so and then there's any reason to fight that we should probably just go with the flow that's that's the way things are headed. Yeah and if anybody isn't aware of course when we say that the current Debian Army L port is on V4T the T on the end means that that supported thumb 1 thumb 2 came in our later versions of the ARM architecture. Correct yeah. Constantinos from ISE is saying that quads it's also implemented in the package so maybe that could be using to solve this part of the puzzle. Quads as opposed to triplets. We only have triplet support for the names but we have vendor tags which is a quad so we could use this vendor tag. That does indeed supply a halfway house solution to this problem for the time being but I don't think we were convinced that it was a sufficient and complete solution and we should do this properly you know and get agreement out in the real world from everybody so we could all agree rather than reusing the vendor tag in the fourth thing which you know I could be made to work it's only a string doesn't matter exactly how it's done but the point is that there's more than that there's the top level and then there's a secondary level for compatibility so there's actually two things here. Yes so you have to roll back the discussion a bit the reason why we were looking for different triplet was to support multiarch and also to distinguish in package builds and the reason for that is that the toolchain was lacking defines for hard float mode but we fixed that so we've sent to the upstream toolchain and it's in our toolchain you'll find no new defines which are used to test for whether you're on on hard float mode or soft float mode so that's one thing which we solved and for multiarch we decided to move out of the triplet so there is not this requirement anymore that all the vendor structures might have different triplets so we can use the same triplet for our might chef. Right and looking at my watch I mean we're going to be running short of time very soon so quickly running through the West Africa. This is the Africa donations donated by Genesi they gave us to Debian ports gave us to terabyte hard disk and some major projects in Debian or people already have some of the efficacy to play with them. So wonderful of being very generous and please everyone help me in thanking them that's excellent stuff. Yes well done Genesi he's really really rather good. They are also providing a building the cluster and providing boards to try to get kernel into mainline and do things properly like everybody should do. Yeah and unless there's anything else I'll leave it this highlighted. Wookie and I have booked with us I hope Wookie's got his with him as well. I'll give him one away already so unless someone thinks of a better reason so we can take it back. Okay okay we have a working Nvidia Tegwa dev board here right now given to me by one of the marketing guys at home as I left last week it's currently running as a standard ish lucid install and there's instructions to go with it obviously if people want to put Debian on it if what people want to go and do other weird stuff with it it's entirely up to them. Does anybody have a have an argument why they should have a Tegwa board? It's has... It's got LVDS, HDMI, VGA, lots of USB, SD, USB, ethernet, audio, MMC, some other weird connector I haven't worked out what it is yet, PCI Express it looks like so and multiple point the Nvidia GPU thing so it can do super high-speed fancy yeah it is quite cool hardware and multiple it's Cortex-09 multi-core thing so armed particular interest is people testing all the multi-core stuff because obviously multi-core hardware is actually quite thin on the ground at the moment. That I don't know if somebody would like to. I'd really like to see Debian run on that so that'll definitely be something and the reason I'll tell you why I'm helping out the Geneva Alliance it's Alliance of Automotive OEMs original equipment manufacturers as well as silicon vendors free-scale etc and certainly Tegra is going to be considered so you know if Debian can run on that another free software can run on that I think the Automotive Alliance would certainly consider that for their next generation of head units. Okay cool I read on the planet Gen2 that the Gen2 people are working on supporting this Nvidia Tegra thing yeah. I guess one thing that occurred to me that I didn't say earlier is about Leonardo is it's a big change for ARM in terms of actually doing things in the open they really have been resistant to that over the last few years they haven't given us and there has been help behind the scenes there are people in there doing you Richard Earnshaw has been doing GCC for a long time and you know but technically he was doing that in his spare time and you know I guess this is a major improvement in terms of how much stuff is actually going to get done and how much easier our lives collectively should get. Literally in the last couple of years ARM has got free software major style I mean from board level down there is buy-in as they realise that okay ARM is called of course is a wonderful CPU architecture whatever the fact I work for them has nothing to do with me saying that but they realise that if they to carry on especially as into more and more smartphones and higher end systems they need to be able to show that there's actually there's a proper software story and so since that message started rattling around in the inside ARM we've now hired lots and lots of free software hackers to basically to make free software work better on ARM there's lots of us now trying to do exactly that if there's any things that people need help with if there's any suggestions from people then we're listening. It's been thoroughly infiltrated so this is all good. With a bit of luck at some point soon I mean those of us who are working on the inside might be able to fix some of the longer standing problems I mean we're still running on exchange for example. But you can't have everything immediately but we are hopefully making a big difference already and that's going to carry on. So, Wicca. Hi. Marcos from IRC said just high-end that the ARM HF port has been added to Debianports.org or at least the support for it is there now so the unofficial port is on the way. So, yeah I guess we should also congratulate Marcos for doing a particularly fine job on that. It hasn't taken him very long. I wish we were all that good. So, yeah basically it's now trivial to just try that out so we can all have a go at this you know how much does hard float matter and what the hell are we going to do next. I guess we haven't got any FTP people here have we? No. So, I'm not quite sure how much resistance there is to having four more flavours of arm in the archive. As there's no one here I suggest we declare it done and see what happens. That's the same conversation at the end of a gunshot. So, I guess any more questions or should we wrap up? The TO-2 boards, what was their intended purpose? Yeah, the TO boards was the previous release for this hardware. I'm aware. The TO-2 board neon bug is a bit unfortunate because it hardlocks the board and it doesn't need root to trigger. Which could be a potential denial service attack if we use them as porter boxes. So, I'm not sure how. No, we don't use it as porter boxes. We use the TO-3 but TO-2 are just building these to build. Okay. That's fine. Not for a deviant board. I just want to make sure that those neon boards are not going someplace where someone can accidentally or intentionally bring them down. They are only useful for building and stuff. Just want to throw that out there. That's all I got. Just do not put it in the garbage. We could do something useful. Okay, I guess. Thank you all for coming.