[03:38:28] --- abo has become available [03:38:36] --- reuteras has become available [03:38:41] --- haba has become available [03:40:36] * haba and all other @kth.se participants back (I guess abo restarted jabberserver and it now can talk to the world again) [03:42:30] mdionne, Simon: Sounds like a good explanation. No, we have not seen any corruption (IMHO). [03:44:59] --- reuteras has left [03:45:49] --- reuteras has become available [03:51:51] --- reuteras has left [03:53:27] --- reuteras has become available [03:55:59] --- stevenjenkins has become available [04:05:23] --- reuteras has left [04:05:41] --- reuteras has become available [04:28:10] --- haba has left [05:28:39] --- haba has become available [05:38:34] --- deason has become available [06:04:33] --- Simon Wilkinson has become available [06:06:45] --- Simon Wilkinson has left [06:19:12] --- reuteras has left [06:19:39] --- Simon Wilkinson has become available [06:20:10] --- reuteras has become available [06:38:11] --- Simon Wilkinson has left [06:43:00] --- meffie has become available [07:13:33] --- mho has become available [07:17:56] --- Simon Wilkinson has become available [07:22:31] --- meffie has left [07:32:39] --- Simon Wilkinson has left [07:34:56] --- meffie has become available [07:40:59] --- Simon Wilkinson has become available [07:42:12] --- Simon Wilkinson has left [07:42:51] --- Simon Wilkinson has become available [07:43:22] Hmmm. If IMA ever does get enabled in a distro kernel, then we're a bit scuppered. [07:43:56] that bad? [07:44:42] We'd need to move to doing open by pathanme. [07:44:47] pathname, even. [07:45:13] That's the only code path that's accessible to non-GPL'd code that correctly does IMA accounting. [07:46:03] uh. the code to do that is mostly there [07:46:14] where by mostly i mean "didn't i write that already" [07:46:29] yeah, moving to open-by-pathname would make less linux-specific stuff [07:46:37] You did, for Mac OS X. But for Linux, we dealt with the loss of iget() by using file handles instead [07:46:51] solaris uses it, too [07:47:02] (Essentially, we use the NFS exporter interface to open cache files on Linux) [07:48:44] Arse. Fedora 12 ships with IMA enabled. This is a problem, then. [07:48:53] --- reuteras has left [07:50:08] --- reuteras has become available [07:50:24] > You did, for Mac OS X. But for Linux [07:50:36] no, pretty sure i wrote it for linux but then someone rewrote away my code [07:51:14] Oh well, looks like we're going to need it back, at least for the latest kernels. [07:51:45] Unless we can convince them that, in effect, making anything the opens a file in the kernel GPL-only is a bad thing. [07:54:02] The problem is that doing filp_open, rather than dentry_open, means that we can no longer work around the AppArmor/SELINUX issue, as filp_open doesn't take operation specific credentials. [08:02:15] I am so glad I always build my own kernels. Originally it was because Red Hat was horrible about keeping up with new hardware in a timely fashion. Nowadays that's less of a problem (not completely gone, when people buy bleedy stuff), but building my own lets me undo dumb decisions. [08:03:52] The way the ima counts have been integrated is just wretched. There are lots of places in the kernel which look like they're broken (David's cachefiles module, for one). [08:04:41] Basically, when you _open_ a file using dentry_open() you're expected to explicitly increment the ima counts. But, when you close the file, the counts are implicitly decremented. [08:05:53] --- meffie has left [08:22:22] Oh well, I've tried emailing the IMA maintained to see if he'd be prepared to help us. [08:25:04] --- Simon Wilkinson has left [08:55:21] --- Rasmus Kaj has become available [08:57:14] --- reuteras has left [09:06:00] --- Simon Wilkinson has become available [09:35:18] --- Rasmus Kaj has left [09:45:00] --- dev-zero@jabber.org has become available [09:57:35] --- Simon Wilkinson has left [09:59:35] am I the only one who finds it odd that we force CC (and other vars) to certain values after a bunch of autoconf tests have run? [10:00:11] like, checking $GCC is bogus if we force a non-gcc compiler in osconf.m4 [10:01:12] It's a part of our build system that has not yet transitioned. [10:01:59] In part because it's complicated -- the compiler found by autoconf is not always the right one for everything we build. Often autoconf will find gcc, because it prefers that or because the user said so, but that compiler can't be used to compile kernel code on most platforms. [10:02:27] no. lots find it odd. no one ever finished the autoconf glue to force compilers that work for platforms that needed it (old solaris for kernels, aix for kernels) [10:02:35] see src/cf/*cc.m4, iirc [10:03:01] none of the people who found it odd ever did more than finding it odd :) [10:03:45] I don't find it odd, because I understand why it is that way. It's unfortunate, and could use fixing, but fixing it is not more important to me than, say, getting my new machine room online. [10:04:23] i have the right compilers when i have a machine at all. i don't care. the people who care the most all care enough to have someone else do it. go them. [10:05:03] I noticed 867 breaks AIX for this reason; I was wondering how much effort is involved in moving the CC-forcing to before autoconf's CC check stuff [10:05:07] sorry. i think i am being excessively cynical today. [10:05:25] if you want to force CC *only on aix* it's probably not that hard. [10:05:49] i take it setenv CC before running configure isn't enough? [10:06:22] I assume it is; I'm pretty sure it would fix that specific issue since it correctly thinks it's not using gcc [10:06:37] I was just wondering about a better fix [10:07:15] well, aix-cc.m4 needs to be updated but them could be used [10:07:46] alternately, if you have more aix than i do at the moment, you may be able to test whether we need to override anymore. some is relic from aix 4.3.3 [10:07:55] --- abo has left [10:07:59] --- stevenjenkins has left [10:08:02] (which i no longer have since the machine died) [10:08:49] --- abo has become available [10:11:35] --- stevenjenkins has become available [11:00:38] I remember vaguely that I wrote config tests to check of AIX and Solaris compilers are there and "good enough". Since that, our usage of said platforms has gone to epsilon, even if Solaris is coming back with ZFS. [11:26:44] --- Simon Wilkinson has become available [11:29:55] How does 867 break AIX - I thought the change there only affected things when gcc is in use? [11:31:25] --- Simon Wilkinson has left [11:33:45] because autoconf autodetects gcc, and later on we force CC to cc [11:33:52] so we think gcc is in use when it's not [11:34:00] (easily circumventable with CC=cc) [11:52:26] --- haba has left [12:07:28] --- Rasmus Kaj has become available [12:14:21] --- Rasmus Kaj has left [12:14:26] --- Rasmus Kaj has become available [13:18:50] --- Russ has left: Replaced by new connection [13:18:50] --- Russ has become available [13:37:34] deason: The problem with moving the compiler forcing to before Autoconf's check for gcc is that Autoconf uses the compiler to figure out the host system in many cases, so we don't have enough information before the compiler check runs to figure out what system we're on. [13:37:59] It does sound like we need to run the compiler setting stuff earlier than we are at present, though. It should be run as early as possible from configure.in. [13:38:54] is there any way to get it to check again, or something like that? [13:39:33] any way for $GCC to go back to 'no', or do we perhaps just need another way of checking "are we really using gcc" [13:40:31] There is not any supported way that I know of. [13:41:07] The right (long term) way forward is to separate the kernel compiler from the userspace compiler and actually build all of OpenAFS on AIX except for the kernel with gcc unless the person compiling overrides CC. [13:41:19] Or at least, I think that's the right way forward -- it's what Autoconf very badly wants to do. [13:41:21] And fighting it is hard. [13:41:42] (Of course, if gcc isn't installed / not in the path, it will use AIX xlc.) [13:42:00] what about the configure tests for the kernelspace compiler? isn't that still a problem? [13:42:31] Those configure tests need to explicitly set CC during those tests to KERNEL_CC and unset it afterwards. [13:42:54] Hopefully none of them will care about whether $GCC is set. [13:43:09] Generally, the kernel tests are just for whether a particular symbol is there or not, so I don't think $GCC matters, does it? [13:43:10] I was imagining $GCC is not the only var of its kind [13:43:42] Also, do we actually have a lot of Autoconf kernel tests for AIX? I know we have tons for Linux, but Linux is something of a special case. [13:44:14] no, I didn't mean to imply anything like that; I'm thinking of the general case [13:45:33] Yeah, I see your point, but I suspect that there isn't a good general-case solution to this. Kernel code is weird and Autoconf isn't really set up to handle it very well. [13:45:44] But I suspect we'll find that it won't matter. [14:48:17] > Linux is something of a special case. And we have macros to handle configure tests involving Linux kernel code, and they fairly easily do the right thing, which is to use Linux's build system, which could care less what compiler we're using. [14:50:50] Right. [14:51:09] So it's only really a problem for kernel configure tests on non-Linux platforms. I'm not sure if we even have any of those right now. [15:00:18] --- mdionne has become available [15:00:56] Me either. I'd sort of doubt it; that's really not something autoconf is good at or even terribly appropriate for -- kernel code is really an extension to existing software with particular architecture, and feature tests can't substitute for knowing what the architecture is. [15:01:07] Yeah, exactly. [15:20:00] --- haba has become available [15:32:00] --- mdionne has left [15:59:39] --- Russ has left: Disconnected [16:17:25] --- haba has left [16:17:25] --- dev-zero@jabber.org has left: Lost connection [16:26:17] --- dev-zero@jabber.org has become available [16:38:27] --- abo has left [16:38:58] --- abo has become available [17:08:32] --- jaltman has become available [17:32:31] --- Russ has become available [18:07:08] --- Rasmus Kaj has left [18:08:02] --- Rasmus Kaj has become available [18:14:38] --- jaltman has left: Replaced by new connection [18:14:40] --- jaltman has become available [18:47:53] --- jaltman has left: Disconnected [20:29:02] --- jaltman has become available [20:36:24] --- deason has left [21:38:44] --- Jeffrey Altman has become available [22:02:23] --- reuteras has become available [23:36:46] --- dev-zero@jabber.org has left