| 29 Nov 2024 |
 raboof | a constant would violate the "must be different for each translation unit" requirement even more though | 17:22:15 |
| raboof | but it'd be useful to find out what the impact of not specifying it at all like you suggested would be | 17:22:54 |
| raboof | * but it'd be useful to find out what the impact of not specifying it at all would be like you suggested | 17:23:11 |
 p14 | In reply to @raboof:matrix.org
a constant would violate the "must be different for each translation unit" requirement even more though I haven’t seen anyone articulate why this is important though, I am wondering if it is bogus. I also wonder exactly how/whether it affects the build id as was suggested | 17:31:14 |
|  lassulus changed their profile picture. | 18:30:27 |
| 30 Nov 2024 |
| p14 | raboof: the most significant lead I have for the rsync nonrepro is that the text section is different; and that a handful of symbols are different sizes. If I diff one of those symbols I find that it has extra code towards the end in my local build. | 09:15:03 |
| p14 | It seems as-if I have a reproducible build locally though, so what is divergent is my local build vs what is in cache.nixos.org | 09:16:00 |
| p14 | So I have another x86 machine, and the divergence with cache.nixos.org is not present there. Hardware difference? One's an old intel chip, another is a recent amd chip. | 10:31:20 |
| p14 | Diffing the config log shows that one machine is getting #define INET6 1 in the config.h and the other is not, from checking whether to enable ipv6. | 10:39:47 |
| p14 | What does one do about this sort of non-reproducibility according to machine configuration (in this case ipv6 being disabled)? Is this a tractable problem? Will upstream care? Or is reproducibility only defined if we talk about machines having identical configuration? | 11:54:33 |
 Mindavi | I think it makes sense to ask upstream whether this is intentional | 11:58:20 |
| p14 | I'm writing an issue. | 11:58:29 |
| p14 | In reply to @rick:matrix.ciphernetics.nl
I think it makes sense to ask upstream whether this is intentional https://github.com/RsyncProject/rsync/issues/675 | 11:59:59 |
| p14 | Off topic: I enabled ipv6 to prove that this brings the builds into alignment (which it does), and immediately firefox is now hanging with 'firefox is not responding, force quit?', immediately making me remember why I had ipv6 disabled in the first place :( | 12:11:14 |
 Atemu | Uh, huh, check whether your network's IPv6 configuration is actually correct. I once had a rogue WAP announcing a prefix for instance which caused massive issues (though not FF hanging O.o) | 12:38:30 |
| Atemu | I wonder how a program in the sandbox would even know that IPv6 is disabled | 12:38:49 |
| Atemu | How did you disable that? | 12:38:52 |
| p14 | Kernel command line ipv6.disable=1 | 12:39:03 |
| raboof | In reply to @atemu12:matrix.org
I wonder how a program in the sandbox would even know that IPv6 is disabled configure.sh does if (socket(AF_INET6, SOCK_STREAM, 0) < 0) | 12:39:58 |
| p14 | I don't know for sure that firefox hanging relates to ipv6 being enabled but it is suspicious since I've not otherwise observed that, and it happened pretty quickly after enabling it. But that's also my experience of ipv6 being enabled: random stuff stops working properly in tricky to diagnose ways | 12:40:16 |
| Atemu | p14: Do you also get super slow page loading times? | 12:40:45 |
| raboof | definitely a good idea to ask upstream. for nixpkgs we definitely want to build with ipv6 support even if the build machine doesn't have it enabled, so we might want to add --enable-ipv6 in the mean time? | 12:41:01 |
| p14 | And I have a good technical ISP who provides good ipv6 support, but I have it disabled at the router. | 12:41:02 |
| p14 | In reply to @raboof:matrix.org
definitely a good idea to ask upstream. for nixpkgs we definitely want to build with ipv6 support even if the build machine doesn't have it enabled, so we might want to add --enable-ipv6 in the mean time? I've not dug into the build script but would it still do the config test? | 12:41:30 |
| raboof | https://github.com/RsyncProject/rsync/blob/master/configure.ac#L379-L407 I'm not too good with automake, not sure | 12:44:00 |
| p14 | In reply to @raboof:matrix.org
https://github.com/RsyncProject/rsync/blob/master/configure.ac#L379-L407 I'm not too good with automake, not sure Tested it here, it looks like enabling it does skip the test. | 12:47:31 |
| p14 | And confirmed that it produces the same executable as output. | 12:50:34 |
| p14 | Is enabling ipv6 in rsync a change to target staging, or master? | 12:51:05 |
| raboof | depends on the number of rebuilds, I'd target master and wait for ofborg to tell you | 12:52:56 |
| p14 | Fix: https://github.com/NixOS/nixpkgs/pull/360414
I'll leave it as draft until the checkers are happy. | 12:58:38 |
