[00:00:53] --- kaj has become available [00:10:39] --- Russ has left: Disconnected [00:22:59] --- reuteras has become available [00:29:59] --- jaltman has left: Disconnected [05:52:58] --- kaj has left [05:52:59] --- kaj has become available [05:54:50] --- meffie has left [05:56:08] --- meffie has become available [06:16:02] --- meffie has left [06:32:55] --- reuteras has left [06:34:55] --- jaltman has become available [06:40:20] --- jaltman has left: Disconnected [06:41:26] --- rod has left [06:53:24] --- pod has become available [06:59:35] Yay. Looks like the way we call truncate_inode_pages has been wrong since around the dawn of time... [07:06:28] --- deason has become available [07:32:56] --- cudave has become available [07:47:44] --- jaltman has become available [08:36:42] --- jaltman has left: Replaced by new connection [08:36:43] --- jaltman has become available [08:47:54] --- jaltman has left: Replaced by new connection [08:47:54] --- jaltman has become available [09:05:48] --- kaj has left [09:28:11] --- jaltman has left: Replaced by new connection [09:28:12] --- jaltman has become available [09:37:11] --- jaltman has left: Disconnected [09:38:03] --- jaltman has become available [09:58:52] --- kaj has become available [10:01:26] --- rra has become available [10:03:08] --- jaltman has left: Disconnected [10:05:53] --- jaltman has become available [10:07:27] --- jaltman has left: Disconnected [10:28:18] --- kaj has left [10:31:12] * rra writes long things in Gerrit that should probably be on openafs-devel. *heh* [10:31:31] Yeah. We have that problem... [10:31:38] I do it all of the time. [10:31:50] I'll resend the bit about Autoconf philosophy there as soon as I get through my morning mail. [10:49:07] --- matt has become available [10:57:07] fwiw, I don't have much of an opinion on the fuse change; it was just one of my todos from the meeting [11:01:16] Yeah, it was after I'd left. [11:07:50] --- kaj has become available [11:28:38] --- kaj has left [11:34:04] --- tkeiser has become available [11:40:28] --- jaltman has become available [11:40:54] --- jaltman has left: Replaced by new connection [11:40:55] --- jaltman has become available [11:46:16] --- tkeiser has left [11:55:05] --- tkeiser has become available [11:55:59] --- kula has left [11:56:11] --- kula has become available [12:03:37] --- kaj has become available [12:59:33] --- meffie has become available [13:15:45] --- tkeiser has left [13:44:21] --- tkeiser has become available [14:21:14] --- Kevin Sumner has left [15:30:43] --- meffie has left [15:33:19] --- meffie has become available [16:01:39] --- kula has left [16:17:53] WRT fuse, my opinion is that options definitely should be in one central place. Scattering them amongst multiple .m4 files just leads to uncertainty and doubt. [16:18:20] I generally agree with --enable. [16:18:23] I don't agree with --with. [16:18:48] Definitely for --enable. Less sure about --with [16:19:13] > uncertainty and doubt. No fear? [16:19:13] --with flags are often conceptually part of a macro with a well-defined interface where you can override its default location probes. If we want to reuse any Autoconf macros from any other source, they will often define their own --with flags. [16:19:37] If the macros have a reasonably defined interface, I think that is fine. [16:19:48] See, for instance, src/cf/krb5.m4. [16:21:27] In the long run, of course, --enable-fuse should become --with-fuse and move into the macro file itself, and we should do the normal thing of building with FUSE if we have it, ignoring it if we don't, failing if we don't have it and --with-fuse was explicitly given, and ignoring it regardless if --without-fuse was explicitly given. [16:22:06] Yeah. [16:22:25] I think the issue is that --enable is seen as conveying certain assumptions about maturity. [16:22:51] Also, while I think collecting OpenAFS's --enable flags in one (or, really, two, as mentioned in my mail) places is a good idea, I'm not horribly worried about it since grep AC_ARG_ENABLE configure.in acinclude.m4 src/cf/*.m4 is not very hard. [16:23:44] We're going to get some that we can't do much about, like --enable-dependency-tracking (although we only have that because we claim we're using Automake when we're not) or --enable-reduced-depends (which comes along with krb5.m4). [16:25:05] (Surely --enable-bigendian and --enable-littleendian can go away in favor of just using AC_C_BIGENDIAN?) [16:25:23] --- matt has left [16:25:57] I'd really hope so. [16:30:25] --- deason has left [16:30:51] --- tkeiser has left [16:32:01] * rra rips out all of our code for that in favor of using AC_C_BIGENDIAN. [16:32:49] Hm, the existing code smells like it tries to handle the case where the kernel and the userspace are of different endiannesses. [16:32:51] That's impossible, no? [16:33:27] We need Autoconf 2.63 for fully correct handling of Mac OS X universal binaries with endianness, but I bet we were losing in that edge case already. [16:34:12] We certainly weren't doing anything special to handle it, so unless people were passing --enable flags for Mac OS X builds, we should be fine. [16:38:00] Ah, hm, no, we were reproducing the workaround in a different way that Autoconf now does. [16:38:12] I suppose people are going to be sad if I say that we should really use Autoconf 2.63 or later. [16:38:34] I think we've already made that call. [16:38:38] Certainly 2.60 or later. [16:58:40] --- kula has become available [17:02:46] Is going to 2.63 okay? There was some trouble over 2.60 and I don't know if the higher version causes even more trouble. [17:03:02] Ah, yeah, 2.63 will be more of a problem. [17:03:09] Too new for Debian stable, so probably too new for just about everything. [17:03:13] I'll sit on this patch for a while. [17:20:21] --- rra has left: Disconnected [17:44:51] --- Russ has become available [17:49:52] --- JSund has left [19:16:24] --- deason has become available [19:36:08] --- tkeiser has become available [20:34:32] --- jaltman has left: Replaced by new connection [20:34:32] --- jaltman has become available [20:38:19] > I suppose people are going to be sad if I say that we should really > use Autoconf 2.63 or later. i'll kick and fuss [20:39:04] > We need Autoconf 2.63 for fully correct handling of Mac OS X universal > binaries with endianness, but I bet we were losing in that edge case already. we build kernel modules. autoconf can't help us [20:49:03] How so can't Autoconf help us? [20:49:44] I'm pretty sure that Autoconf 2.63 does exactly the right thing for kernel module builds on Mac OS X. I'm pretty sure of that because it does exactly what we're doing right now. :) [20:49:55] Except does it for us and does it with somewhat less cruft. [20:50:38] If it detects that you're using a compiler with universal binary support, it uses the __BIG_ENDIAN__ define at build time to decide what endianness you are. [20:51:22] You may have missed the bit though later where I figured out that we do indeed handle that case now. [20:51:45] So switching to Autoconf's native support doesn't buy us any correctness, just dumps the configure options and means less code to maintain. It can wait until 2.63 is more widespread. [20:51:53] the kernel module stuff has to be glued; you have to build each object and lipo them together. the problem is the flags you pass to cc for one object type are not the same as the next even excluding architecture [20:52:28] does it do that, or does it do what we do for every non-kernel object? [20:52:38] (which works for everyone but us, basically) [20:52:56] --- jaltman has left: Disconnected [20:53:05] --- jaltman has become available [20:54:14] If it detects the compiler has universal binary support, it always defers to the __BIG_ENDIAN__ define. [20:54:20] I'm not sure if that's what you mean. [20:54:28] --- abo has left [20:54:55] --- abo has become available [20:55:45] It does this in afsconfig.h: #if defined AC_APPLE_UNIVERSAL_BUILD # if defined __BIG_ENDIAN__ # define WORDS_BIGENDIAN 1 # endif #else # ifndef WORDS_BIGENDIAN /* # undef WORDS_BIGENDIAN */ # endif #endif [20:56:10] (With the #undef bit replaced by #define by configure's processing. [20:57:11] it's not what i mean. [20:57:22] like, we need that but it's not sufficient [20:57:36] look at the gross crap in src/libafs/MakefileProto.DARWIN.in [20:58:28] when new splufty autoconf will solve that, hey, yay. until then it's just another thing i need to upgrade on every machine i do development on. you might guess i'm not very interested. [21:00:53] Ah, that's what you're talking about. Yeah, that's not the problem I'm trying to solve. The problem I'm trying to solve is to get rid of the Autoconf cruft we already have, including the stray --enable options that really no one should ever need or want to use. Not a big deal right now. [21:57:53] --- tkeiser has left [23:05:09] --- deason has left [23:17:01] --- Born Fool has become available [00:00:00] --- Born Fool has left