=== timchen1` is now known as timchen119 [18:15] cyphermox, cjwatson: would you please review live-installer changes? http://people.canonical.com/~xnox/live-installer/ [18:15] git formatted patch series. [18:15] 0004 -> fixes all the typos and mistakes. but otherwise it's in logical order. [19:26] xnox: ok [19:26] cyphermox, the idea is that one can do: $ sudo mount -o loop xenial-server.iso /var/www/html/ [19:26] cyphermox, netboot [19:27] I figured [19:27] cyphermox, choose-mirror $thatmachine -> and the install would work without any extra packages, or having a full mirror. [19:27] (as in packages that are not on the iso) [19:28] err [19:28] oh yeah, sorry, EPARSE [19:29] * xnox is waiting for s390x image to be respun to retest it, was lacking kernel modules. [19:31] might wanna fixup 0003-Whitespace-tab-changes-only.patch for line length [19:34] and I'm a little concerned with 0004-Use-prefix-throughout.patch; we might want to check harder for a bad mirror and whatnot, before running fetch-url [19:34] oh, but otoh the mirror should usually be good anyway, otherwise bad things might happen [19:34] what does the debian installer team think of your patches? [19:54] cyphermox, these are brand new. and as far as i know they don't use live-installer by default. [19:54] cyphermox, i will post them to them, in the new year. at the moment rushing to get s390x working. [19:55] cyphermox, what do you mean by "check harder for a bad mirror"? at the moment we try to download, if fails we skip. [19:55] if it does download, hopefully it's something usable. [19:55] and we are garded with check for .disk/base_installable which no mirrors to date should have, unless they are are ubuntu iso exports.... [19:58] Nah, don't bother. [19:58] cyphermox, so... ok for me to upload this? [19:58] =) [19:58] Well, I wouldn't, because people could also fail to copy dot files [19:58] they are not meant to copy them, they are meant to mount the whole iso. [19:58] Yeah, it looks sane, but I really think it should be in debian too [19:59] ok. [19:59] You could just as well copy the files though [19:59] they will notice when things don't work and explode ;-) [19:59] Why do you need this though? [20:00] You should already be able to live-install with a URL [20:00] That's what live-installer/net-image is for [20:00] correct, however that's at the moment preseed only option, and this patches make a computed attempt. [20:01] and i'm trying to make it computed without any preseeding. [20:01] OK [20:01] so s390x cannot boot ISO and cannot mount ISO, hence we will provide an option to use ISO as a "near by" mirror, when ports.ubuntu.com is not accessible. [20:02] so given an .iso, one would be able to mount & export it over e.g. http on one mainframe z/VM. [20:02] then ipl boot "s390x netinstaller" point it at that iso, and voila by magic squashfs is fetched and things work as if we did a normal boot off an iso. [20:12] Hum. [20:13] We already keep the squashfs on cdimage rather than ports though, cdimage isn't in the mirrors [20:13] this is not to install from cdimage, nor from ports. [20:14] it could be, but that's not the current goal. [20:14] I don't think it would be such a great idea to duplicate the location of images [20:14] OK [20:14] when pointed at ports mirror, simply debootstrap is used. [20:14] and the full base-installer is chosen / run, rather than live-installer. [20:20] yeah [20:20] well, I have no objections to your patches [20:36] xnox: I'll do the live-instlaler merge on top of your upload later [20:37] ok. let me upload them then. [21:16] cyphermox, with a fixed kernel finally figured out a bug in my code. [21:16] cyphermox, http://paste.ubuntu.com/14056501/ [21:16] i'm not allowed to return 1 from a function, cause otherwise set -e kills me. [21:16] so i'm making it a global var, which works. it is afterall just a state flag, which goes from 0->1 only once. [21:17] i think it's ok, unless you can think of a better way to return things from a function. [21:17] set a global RET variable? [21:20] or echo it and call it RET=$(process_places ... [21:21] or echo the place you've found [21:21] hm, not sure i want to invoke a subshell $( [21:25] why not? [21:25] it's used regularly in all d-i components already [21:26] well, I have no opinion on how specfically to achieve return the value as you want, either method works [21:35] ok. [21:57] + * And add a cheeky abi dependant build-dep for the kernel ABI, such that [21:57] + d-i is attempted to build once appropriate linux is fully built, [21:57] + published, accepted. This is similar to the linux-meta automatic [21:57] + build-deps. Possibly further var replacements should be taken from [21:57] + linux-meta package to just bump/define a single $kernel-abi variable. [21:57] ^ ugh, yeah. we shouldn't add places to modify version numbers and get things wrong, we should reduce them [21:58] cyphermox, it's a mass sed anyway. and at the moment we modify N places, where N is number of arches. [21:58] now there is one more. [21:59] cyphermox, are you doing a d-i upload? [21:59] sure, but N+1 is still larger than N [21:59] no, I'm not [21:59] just noticed it while looking at merges overview [21:59] cyphermox, and well that dep was not quite sufficient. cause it also needs an arch:all dep. [21:59] cyphermox, ack. [21:59] yeah [21:59] ideally we would indeed have a single debian/kernel-abi file or some such, and everything source that. [21:59] but in some cases you also need other kernel ABI version changes for other things [22:00] +1 [22:00] hence the comment ;-) [22:00] heh [22:00] I'll bring it up to KiBi too, that would certainly be good, and we still have yet to merge d-i [22:07] xnox: FWIW, I was going to revert that change on my next upload. It's not super helpful. [22:07] And yes, I agree that a single place for abi substitutions would be nice, and I might commit that both upstream and to Ubuntu.