[02:52] <R0b0t1> Hello, how are ARM packages built?
[02:53] <R0b0t1> I ask because with my distribution I have experienced various issues related to cross compilation that go away when compiled on-device. Are packages for ARM devices compiled on ARM devices, or does it depend?
[04:02] <TheLexx> is this where I ask questions about casper booting system
[15:21] <cjwatson> R0b0t1: We build ARM packages natively, on 64-bit ARM server hardware (in an appropriate chroot - 64-bit ARM hardware can run 32-bit code).
[16:01] <R0b0t1> cjwatson: AhHA!
[16:01] <R0b0t1> So Ubuntu has unobtainium.
[16:04] <cjwatson> It is admittedly not the most easily-available hardware on the planet.
[16:05] <cjwatson> AFAIK we did get it off the shelf rather than via any partnership deal or whatever, although I believe the particular hardware we have is no longer available :-(
[16:07] <R0b0t1> That's unfortunate. Do you know what it is, exactly?
[16:07] <JanC> is it really that important to use server hardware?
[16:07] <R0b0t1> JanC: Yes, unfortunately. I'm fighting an uphill battle trying to get every package to compile in a crossdev environment. There are issues with autoconf and some harder to troubleshoot issues with compiling and linking that only show up in the cross compiler.
[16:08] <R0b0t1> Some of it does seem to be poor support for SIMD acceleration in things like firefox and ffmpeg.
[16:08] <JanC> I mean server vs. some other ARM SoC
[16:08] <cjwatson> R0b0t1: HP ProLiant m400
[16:09] <R0b0t1> JanC: Well, no. But on any other SoC most packages take half a day and there is insufficient IO speed to saturate the processor.
[16:09] <cjwatson> You might still be able to find it somewhere, I don't know
[16:09] <R0b0t1> i.e. SD cards are too slow and sometimes end up as the compilation bottleneck.
[16:09] <R0b0t1> cjwatson, cheers.
[16:09] <cjwatson> (I just operate the Launchpad end of it - I wasn't involved in the purchasing)
[16:10] <R0b0t1> cjwatson: Well, with no information whatsoever it was very hard to find companies producing ARM server boards. There's the other issue though, which is that they're still thousands of dollars.
[16:11] <R0b0t1> So this gives me something to go on but it probably has to wait. Again, thanks.
[16:11] <cjwatson> np
[16:54] <JanC> R0b0t1: https://softiron.com/products/overdrive-1000/ might be useful
[18:40] <Unit193> cjwatson: ...Speaking of, bazaar.launchpad.net hostkeys are 1024 RSA.
[18:42] <cjwatson> Unit193: bug please, it'll be a little while before I can do much about that
[18:43] <cjwatson> Unit193: or actually, don't bother, we already have a bug for ecdsa/ed25519 support
[18:44] <cjwatson> which you know about because you've commented on it :)
[18:44] <Unit193> I know, I'm subbed and have commented.  Though that's generally more client and less about host keys.
[18:44] <cjwatson> effectively the same thing
[18:45] <cjwatson> Unit193: the path for making progress on that is to finish the upgrade to xenial, then convert our build system from buildout to pip, and *then* we'll be able to upgrade twisted in a reasonable way
[18:45] <Unit193> Yep, just depends on which side generates the key.  And hopefully not making too much noise there.
[18:45] <cjwatson> until then it's a nightmare
[18:45] <Unit193> Oouch.
[18:45] <cjwatson> (I've tried without that and given up)
[18:56] <Unit193> Heh, no rush as ed25519 isn't finished yet.  And you're already doing great work on LP, so thanks.
[19:37] <R0b0t1> JanC: A friend linked me to that earlier. Something I couldn't ascertain was whether or not the A53 used in it was an actual server component or an SoC. Reading it again it seems like a server component, but I still can't tell.
[19:39] <R0b0t1> Like I said most of the bottleneck seemed to be IO, and it seems like it is saying the A53 version supports SATA. I'll give it another look.
[19:40] <R0b0t1> It seems to be made for development, so...
[19:40] <R0b0t1> JanC, cheers.
[19:53] <JanC> R0b0t1: as far as I know many SoCs support SATA actually, even if many boards/devices don't actually expose/use it
[19:56] <R0b0t1> JanC: Well, I know of the Banana Pi, but it seems to use a USB SATA controller. As far as I know most other boards are the same way as the highest speed interface the phone SoCs tend to support is USB3, and they may only have one USB3 interface.
[19:56] <R0b0t1> I hope I'm wrong with newer devices of course.