 Ie! Cyn allan y medсе nes yr hynny sy'n cael cwysig o ymddangos ymddangos yn y Cysigol, gan yr hyn? Roedd dim, a bod yn rhywbeth o newid ymddangos, i'r ddweud o ac yn unig o'r holl ddydd i'r bobl yn yr Yng ngyseidol. Yn cael ei ddim yn script? Yn cael ei ddim yn hyn feddwl yn hyn i ddechrau. Nid yw'n ymddangos yn y G siŵr? Rwy'n meddwl ymddangos? I was having a lunch with my friend and he was talking about the company. It is not really a good experience but we are not using stable without many additional snow, so we will move with it. Mae chi'n rhaid i fod ymlaen, maen nhw yma yn geisio fel Gwyduchol, fel Llyfrgell, fel Llyfrgell yma mewn unig o 답 experton. Mae'r ddweud i'r ddaf i gyd yn gweithio. Mae'r ddweud i gael bod maen nhw'n differentiate, i gyfligafol hefyd, ac mae'n ddyddol yn ddiwedd, yn cyd-ddiwrnEEFygiad ac mae hynny, dyna nhw. Mae'n ddysgu'r dyma'r ddysgu. Felly, yna'r ddysgu? Yn ddysgu, bydwch yn ddysgu. Fe'n ddylch chi'n ddysgu'r ddechrau, ddysgu'r 3.10 yn gwneud drosyngau, gyda'r llyfr. Felly, fy oedden nhw'n fyddwch i'w dwydd yn ystau'r ardalau mewn gwleiddoedd. Rwy'n da'n siarad y weld fy oedden nhw'n hwylodd yr Omnolol? Rwy'n ddim fyddwch i ddweud â'r gwleiddoedd. Felly, mae yna gwell o ybyd ar hyn o'r ystafell. Ydwch chi'n gwybod i'w'r wneud ddafodd yw'r gweithloedd AMSM, a ydwch chi'n ei gwneud yn gweld ddim i chi'n gweld ddangos? ac mae'r pethau yn unig ond gan ddim, ond le'n gweithio ni'n gweithio ddim yn burgers yn gyfodol. Yn gweithio gyfan rhaid, maen nhw'n ddim podod yn gweithio hwn ar y cyflau', cael eu bacti drwyddo gan'r newydd yn ei brifysgol, ac mae'n ddim ddim yn cymdangad eich tyme, oherwydd o'r newydd yn newydd 10. Ac that seems like a potential middle drum between … Let's update the whole kernel in our stable retrieve and instead let's just pull up all the new graphics I've installed because you have no idea which pieces we need and which bits we don't need because you probably need all of it Would that be an acceptable solution? Would it might be, but then we have to be very careful Mae'r trwm yn wneud ei wneud yn meddwl gwaith croesol cyfnod yn y ONE. Rwy'n meddwl cyn nhw, maen nhw'n frech i'w ddiwch cyfnodd. Roeddwn i'w ddweud ei wneud yn cael eu bod yn gallu gweithio gwaithau. Roeddwn i'w ddweudökol eisiau gael ei wneud pan yn bod chi ddim yn cael eu gwneud yma? I'm Eric Enholton, the Intel Graphics Driver Team. So, one solution we've been talking about implementing in the user-state side of the Graphics Driver is, you know, there's this conflict between, we really want to be able to get the current version of our driver for the new chip sets into old versions of distributions, but no distribution wants to take the new driver because, yeah, you know, you broke the old generation on some application and you're not getting around to fixing it yet. So, one of the ideas we have was, okay, let's take the livegeo loader, the loader of the Graphics Driver, have it look at the PCID of your chip set and decide which file to open based on that. And then you could have, you know, Debian could choose to have the current generation of chip sets load the current version of the driver that already supports them, and then when newer chip sets come along later, you would just update to say, oh, yeah, and all these new chip sets load this new driver off to the side that was built using a newer version of Mesa, and then you don't have regressions because otherwise it didn't work at all. Yeah. Actually, I never understood why I915 that those drivers haven't been split so that the refactoring to support the new chip has no effect on the new chip. So, we actually did just split off the remainder of the 915 code from 955. We used every pile of shared code about 15,000 lines. We split that into, it's down to 10,000 lines. In Mesa, yeah. And we talk constantly about, well, you know, is there another split we want to do, but the question is always which generation, because no matter what you end up duplicating, once we do that basically you wouldn't end up fixing anything on older drivers, and we do have contributors. But that's fine. We have contributors improving older drivers, which is pretty cool, like being able to run new applications on your old Iron Lake system from three years ago now. You know, that's neat that we're doing that in open-source software, I think. I think we want to make that possible, but yeah, for a stable distribution I think you want to avoid pulling in those newer feature updates. So, yeah, you know, is matching based on PCID going to be enough for people to say, yeah, we could pull in new Mesa for new chip sets and keep the old stuff for the old stuff? If you can make that work, yeah. I don't know how that would work in the kernel level, how you deal with having two 9, 915 drivers. I think the worst part for the kernel is not going to be the 915 driver, which you could presumably, you know, do some said job to subset all your symbols, but there's also the shared DRM core, which is, additionally, these days doing communication between the different DRM drivers. I don't think you could do that split in the same way in the kernel. Right, yeah. Is that implied to user-space Mesa versions installed if you're going to choose pick a runtime when you run? Yeah, so you would have Mesa 9.1, you know, today with the Intel, up to Ivy Bridge driver and the Radeon drivers and the new Bo drivers, and then, you know, stable would also end up with Mesa 9.2 for Bay Trail, the next platform that's coming out that we're still building the code for. So does anyone know whether you're already packaging Mesa so you can install more in one version or not? So the, I don't know how that would get managed from the, like, get package version control side of things, but in terms of the installed files, it's just these DRI.so file names, and all we would, what we need to do is add this code for having to load or look at your PCID and decide a different name based on that. So that's my job. We didn't really get an answer there about how we would deal with the DRM core changes there. I've got nothing on that one. With new network cards, how dangerous is that to get the compatibility issues? Sporting new Nick drivers, less dangerous as new drivers? That seems to be easier because there aren't so many changes in the network driver API. And since that's my day job as my tenure network driver, I know all about the gotchas for back and forth compatibility there. So that's, I won't say that we easy it up to date with ethernet drive support, but that's something I can think is easier to support, easier to do. So the compact drivers project does cover some ethernet drivers, I think. I'm not sure exactly which part to see. No, I don't have a list. So could we perhaps have compact drivers as an optional, as an extra separate package? And what would that, what would that, how could that work? As a package of the newer drivers with some compatibility codes so they can build against an older kernel headers, they would build those out of tree. Well, that's, that's how it's meant to work. So if we would package that, that would, I suppose those drivers modules will get installed into the updates directory in us, they would override the drivers part of the Linux image package. And you'd need to have the, I suppose you'd need to have the installer work out, but if you have, if you have such and such PCI IDs present, I guess you'd need to have discover, know that with these PCI IDs, you need, you'll need compact drivers. Do that again? Would that mean having a new compact driver package at each point released? Yes, so yes. Would you update people to, to install with one version of compact drivers to the new one? I suppose so, yeah. So obviously there's a potential for regression there. The same regression for that. Yes. Yes. But there's no, there's no regression for people who didn't need it in the first place. So I don't know, it's certainly no worse than what they would have at the moment, which is they will have the option to install a new kernel from, from backports, and then if they want any security support, they have to keep updating that as well, for anything close to security support. Because of course backports doesn't have real security support for what it hasn't done. So would you get one massive compact module package, or do you get kind of compact security and compact net and compact flour? Well it hasn't, so far so no, it hasn't been, no one's attempted to package this yet. Potentially you could do that. I guess that would give you some, the compacts could be, or whatever. Eat the mic over there, okay? Come on, speak up. How are we going to do this? Or raise your hand. I'm not waiting for the mic. Any bright ideas? I'll have a go at packaging compact drivers. That was quick.