**** BEGIN LOGGING AT Sat Jan 16 03:00:36 2021 Jan 16 13:14:35 How let mru in? Jan 16 13:15:21 mru: it is listed as a backport Jan 16 13:15:43 an arm patch from an intel address Jan 16 13:15:47 arg boost Jan 16 13:16:59 it's so old the history is lost Jan 16 13:18:38 yeah Jan 16 13:18:41 wtf Jan 16 13:18:47 it can't be a backport Jan 16 13:18:55 because it doesn't exist anywhere else that I can find Jan 16 13:18:55 says backport, but from where, you would think upstream would have it Jan 16 13:19:10 upstream boost doesn't have that code and never has Jan 16 13:19:19 all drink Jan 16 13:19:34 and it breaks with some gcc versions Jan 16 13:21:09 do you understand what it is trying to do? Jan 16 13:21:25 should we talk about dropping it? or raising an issue with upstream? Jan 16 13:22:43 I guess at some point the __sync_foo builtins were broken in gcc, or someone thought they were Jan 16 13:22:48 for arm only Jan 16 13:22:56 I wonder what buildroot has :) Jan 16 13:23:05 so they added that bit of asm Jan 16 13:23:09 the asm is correct Jan 16 13:23:09 yeah Jan 16 13:23:19 maybe broke or missing on armv7a? Jan 16 13:23:45 but it's missing a return statement Jan 16 13:23:54 so dropping patch is likely the right thing to do? Jan 16 13:23:56 which could be what trips up gcc Jan 16 13:24:07 at least it emits a function with no return instruction Jan 16 13:24:16 so it falls through to whatever happens to follow Jan 16 13:24:18 that sounds, bad Jan 16 13:24:26 yeah, sigsegv Jan 16 13:24:29 oops Jan 16 13:24:32 hmmm Jan 16 13:24:40 dropping the patch made the program work Jan 16 13:25:01 I would propose droppping it Jan 16 13:25:08 definitely Jan 16 13:25:23 Can you send a patch? Jan 16 13:25:29 where to? Jan 16 13:26:47 openembedded-core@openembedded.org Jan 16 13:26:54 layers readme shoudl ahve it Jan 16 13:27:00 against oe-coire, not poky :) Jan 16 13:27:21 does that list require subscription before posting? Jan 16 13:27:41 I forget Jan 16 13:29:00 if it is too much trouble I can send it Jan 16 13:29:15 but I'd like you to get credit Jan 16 13:29:29 I'll do it Jan 16 13:30:05 thanks! Jan 16 13:30:14 afk breakfast and good to hear from you Jan 16 13:34:28 sent Jan 16 13:36:01 thanks, I'll check for it getting stuck Jan 16 13:36:09 it's in the archive Jan 16 13:36:21 I am having flashbacks to that era, I bet I've cargo culted that code somewhere Jan 16 13:38:08 quick grep for ldrex says onlu in that patch Jan 16 13:38:22 and non in meta-oe Jan 16 13:38:24 the asm fragment is fine Jan 16 13:38:36 it's the standard atomic add sequence Jan 16 13:38:47 yeah, jsut looking for that style of patch Jan 16 13:39:22 what breaks? Jan 16 13:39:31 gcc does something dumb in some cases? Jan 16 13:39:33 it probably breaks because that branch of the ifdef doesn't have a C return statment in a function declared to return a value Jan 16 13:39:37 that's not allowed Jan 16 13:40:14 and as I said, gcc emits a function with no return instruction Jan 16 13:40:16 missing retunr around 22? Jan 16 13:40:25 right Jan 16 13:41:08 um how did it ever work? Jan 16 13:41:13 dumb luck Jan 16 13:41:37 ugly Jan 16 13:42:12 I recall copying code like that in other palces atmics where used I think Jan 16 13:42:17 could be fake memory though Jan 16 13:42:54 old gcc might have not supported those __sync_foo things at all on arm Jan 16 13:43:06 or used the obsolete SWP instruction Jan 16 13:43:21 lol lots of ldrex's in oe-classic :) Jan 16 13:43:41 * Crofton|cloud gets nostalgic for the good old days Jan 16 13:43:50 the instruction was added in armv6k or something Jan 16 13:44:37 armv7 cores may implement it Jan 16 13:44:43 cortex-a8 does, others not iirc Jan 16 13:44:44 recipes/gcc/gcc-4.3.3/debian/armel-atomic-builtins.dpatch Jan 16 13:45:03 never trust patches from debian Jan 16 13:45:24 remember openssl Jan 16 13:46:13 https://github.com/openembedded/openembedded/blob/master/recipes/glib-2.0/glib-2.0-2.23.6/gatomic_armv6.patch Jan 16 13:46:23 this is from a million years ago :) Jan 16 13:47:45 it obviously took a little while for compilers to catch up with the new instruction set Jan 16 13:47:52 yeah Jan 16 13:47:55 so a few patches at the time is understandable Jan 16 13:49:25 yeah Jan 16 13:49:57 the boost patch comes from oe-classic :) Jan 16 13:51:15 ANyone have any suggestions for testing oe builds with buildbot Jan 16 13:51:24 where I am not trying to dupe teh autobuilder :) Jan 16 14:06:58 on the bright side, I got some billable time out of this Jan 16 15:12:27 billable time is always a good thing Jan 16 15:12:45 I just looked at the patch in oe-classic and it is missing the return Jan 16 15:14:43 Oh jesus Jan 16 15:14:45 https://github.com/openembedded/openembedded/commit/74e7c9a96d176e32aac4791c7e308a0c904f735b Jan 16 15:16:45 damn that patch goes back to before the packages to recipes switch Jan 16 15:19:30 mru: I added that patch in 2008 Jan 16 15:19:39 with the missinfgreturn Jan 16 15:27:21 oh, the one with the return is wrong too Jan 16 15:28:38 I wrote it in 2008 when I knew nothing :) This is my excuse! Jan 16 15:29:01 I bet I even talked to you abot it :) Jan 16 15:29:30 Well search for the guilty complete, on to punishing the innocent Jan 16 15:32:08 Anyway, I owe you a beer **** ENDING LOGGING AT Sun Jan 17 02:59:57 2021