[00:08:22] --- Russ has left: Disconnected [00:41:18] --- jaltman/FrogsLeap has left: Replaced by new connection [00:41:19] --- jaltman/FrogsLeap has become available [01:00:40] --- jaltman/FrogsLeap has left: Disconnected [01:00:51] --- jaltman/FrogsLeap has become available [02:35:37] --- sxw has become available [02:53:30] --- sxw has left [07:37:14] --- deason has become available [07:47:24] --- pod has left [07:51:19] --- pod has become available [08:41:15] --- meffie has left [09:04:23] --- Russ has become available [10:51:22] --- jaltman/FrogsLeap has left: Replaced by new connection [10:51:24] --- jaltman/FrogsLeap has become available [11:12:12] --- jaltman/FrogsLeap has left [11:14:47] Hm. I wonder why Gerrit isn't sending out mail about the change that I just submitted. [11:42:44] Weird. The build-bot message about it went out fine, but no notification about my submission or my verified mark. [11:43:14] Hm, maybe Gerrit is too "smart" and doesn't send me notifications of my own stuff? [13:57:35] 1.6.0pre1 fails when linking aklog on natty. Is this known? /usr/bin/ld: aklog.o: undefined reference to symbol 'error_message' /usr/bin/ld: note: 'error_message' is defined in DSO /lib64/libcom_err.so.2 so try adding it to the linker command line /lib64/libcom_err.so.2: could not read symbols: Invalid operation collect2: ld returned 1 exit status make[5]: *** [aklog] Error 1 [14:05:33] I guess this is with --enable-reduced-depends, which might be relevant. [14:17:32] Yeah, only with --enable-reduced-depends. [14:29:38] > maybe Gerrit is too "smart" Since the upgrade, I don't remember getting email about new changes that I push. (I do get email about comments, &c.) So it seems likely that your hypothesis is correct. [14:30:21] Anders: It sounds like you're getting afs' com_err and not the system one? You might try 3547. [14:31:40] That doesn’t explain why it works without --enable-reduced-depends. [14:38:53] True; I missed that part. [14:41:21] Ooh, thank you deason for 3569! [15:50:24] Hmm, this is also puzzling. Are afs_xdr_array, …, afs_xdr_vector exported on amd64 but not i386? [16:41:07] rra: 338abe5 “Don't attempt stop if module unloading is disabled” adds a bashism to debian/openafs-client.init. You mean s/==/=/. [16:42:18] Actually, you probably means something else entirely, since $disabled is set but never used. [17:07:45] Thanks, fixed in Git. [17:09:09] OpenAFS needs to explicitly link aklog with -lcom_err because aklog calls error_message directly. It currently probably assumes that this is part of the exported Kerberos API, but that's not the case. [17:10:41] These days, with krb5_get_error_message being the preferred interface, it's legitimate to link with -lkrb5 without linking with -lcom_err, which on Ubuntu (where all binaries are linked with --as-needed) will mean that error_message is not available to programs. [17:13:52] > Thanks, fixed in Git. That still looks wrong, because ‘return 0’ is true. [17:15:31] Good point. [17:16:05] The com_err fix is tricky because we don't want to link with -lcom_err on AIX. [17:16:15] Also, the aklog error handling code is horrific. [17:16:21] Also, modules_disabled looks like a very new feature. [17:16:51] New in v2.6.31-rc1, to be precise. [17:17:58] Okay, I'll suppress errors from the cat to make life easier for backports. [17:19:38] $disabled was probably a better name than $loading, because 1 means disabled. [17:22:13] Yeah, true. [17:23:17] * Russ will roll 1.6.0pre1 packages once someone gets a chance to look over the NEWS patch and I can submit it for 1.6 so that I can easily cherry-pick it. Or until I just approve it, since it's over Christmas and people are mostly not watching. [17:26:26] Cool. I don’t know how much you’ve worked on that yet, but here’s what I’m using for the PPA. (We might as well share .orig tarballs, at least.) http://andersk.mit.edu/gitweb/openafs-ubuntu.git/log/refs/heads/experimental [17:28:28] Orig tarball made from debian/rules get-orig-source? [17:28:50] Yeah, I followed the README.source directions. [17:55:27] Okay, cool, I'll pull that down. [18:12:08] > afs_xdr_array, …, afs_xdr_vector exported on amd64 but not i386? xdr_foo are conditionally #define'd to afs_xdr_foo according to a rather odd conditional I don't entirely understand the reasoning of [18:12:27] which iirc includes amd64 specifically [18:13:10] We've only conditionally exported some symbols for a while, I think, yeah. [18:16:03] Huh. rx-xdr-rename-userland-symbols-also-on-amd64-linux-20030627 is the relevant delta. [18:27:44] there's something I submitted semi-recently to expand doing that for almost all cases to make things a little simpler, but I don't remember where right now [18:56:24] Hm, I can't reproduce the aklog link failure. [18:56:37] Oh, right, I bet I have to use binutils-gold. [18:57:25] Yup, there we go. [18:57:45] Well, that still didn't reproduce the error_message problem, but now I have a different problem. [18:57:56] /usr/bin/ld: aklog.o: in function krb5_encrypt_tkt_part:aklog.c:209: error: undefined reference to 'krb5_c_encrypt_length' /usr/bin/ld: aklog.o: in function krb5_encrypt_tkt_part:aklog.c:217: error: undefined reference to 'krb5_c_encrypt' [18:58:05] That's when linking with Heimdal. [18:58:19] Oh, wait, no, it isn't. [18:58:30] That's MIT. [18:58:35] I bet they're in libk5crypto. [18:58:44] Which we also aren't linking with explicitly. [18:59:13] Still can't reproduce the error_message problem. [18:59:14] Huh. [19:00:58] I wonder if this is somehow explicitly an amd64 problem. [19:01:16] Regardless, --enable-reduced-depends isn't correct for aklog right now because aklog uses things that aren't part of the libkrb5 API. [19:01:35] It looks like I can't easily just patch this, so I think your fix is the right one for the time being. [19:13:37] 1.6.0pre1 packages for Ubuntu 8.04 through 11.04 are built here: https://launchpad.net/~openafs/+archive/master [19:14:24] > Still can't reproduce the error_message problem. I’m on natty amd64, in case you haven’t tried that. [19:15:30] (I would hope that such problems wouldn’t be amd64-specific, but this wouldn’t be the first time today I’ve had such hopes dashed.) [19:17:24] I also don’t have binutils-gold installed; I know that sometimes makes a difference, though usually gold makes more things fail. [19:19:14] I suspect it's for some reason amd64-specific. [19:19:26] Either that or natty has something newer than squeeze that's causing this to break. [19:20:30] If the latter, I would guess binutils 2.21-1ubuntu4 vs. 2.20.1-15. [19:25:34] Maybe it’s this. https://lists.ubuntu.com/archives/ubuntu-devel/2010-November/031991.html [19:29:16] Yeah, that was my first thought, but I explicitly passed in --as-needed, and still couldn't reproduce it. [19:29:33] The --no-add-needed part, not the --as-needed part. [19:32:41] Oh, hm. [19:33:30] Hm, no, adding that too doesn't seem to change things. [19:35:12] Maybe it’s relevant whether the dependencies (e.g. krb5) were built with those flags? [19:36:08] As I understood it, it shouldn't be, but maybe I don't understand it. [20:16:08] --- deason has left [21:14:03] --- jaltman/FrogsLeap has become available [21:51:20] Oh, ha. Ha ha ha. src/aklog/Makefile ignores LDFLAGS. I can reproduce in a squeeze-amd64 chroot with CC='gcc -Wl,--as-needed -Wl,--no-add-needed' ./configure --enable-reduced-depends [21:54:36] Yeah, it seems to only happen on amd64 and not on i386. [23:00:00] Ow. Just ow.