Re: libjudy
On 04/29/17 19:35, Ketil Malde wrote: > > > I'm still waiting for my big run to complete, but I have a smaller test > > Unfortunately, the new executable appears to fail for the same data that > the old one did. So if there is indeed a subtle bug hidden in libjudy, > it doesn't seem to be affected by these options. I'm not sure that I would jump directly to the bug being in Judy. Judy is a VERY mature library that has been widely deployed for over two decades. As you noted there were reports of compiler optimizations leading to some crashes but a root cause was never identified for those specific reports. Unfortunately, being a library means that there is always additional third party code involved which makes debugging quite challenging. The best suggestion that I can offer at this point is to have you post directly to the upstream Judy mailing list for additional assistance. https://lists.sourceforge.net/lists/listinfo/judy-devel Troy signature.asc Description: PGP signature -- Ubuntu-devel-discuss mailing list Ubuntu-devel-discuss@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss
Re: libjudy
> I just built a version with using the -fno-strict-aliasing and > -fno-aggressive-loop-optimizations compiler options: Great, thanks for the quick response! I did a dpkg -i *.deb of your files, rebuilt my executable (statically linked, but I think that would pick up the new version?), and started a new test run. If you don't hear from me by Tuesday, feel free to prod. Unfortunately, there's a chance IT will scr..eh, modify our disk setup over the weekend, and cause things to crash or otherwise get messed up. If so, I'll just have to restart, I guess. :-/ -k -- If I haven't seen further, it is by standing in the footprints of giants -- Ubuntu-devel-discuss mailing list Ubuntu-devel-discuss@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss
Re: libjudy
> Would you please download and test that version and see if you are > able to still duplicate the segfault? I'm still waiting for my big run to complete, but I have a smaller test case which has been particularly troublesome (and which leads to corrupt output). Unfortunately, using the new Judy version produced the same incorrect output. (Assuming my recompile really did include it, not sure how I can test that) -k -- If I haven't seen further, it is by standing in the footprints of giants -- Ubuntu-devel-discuss mailing list Ubuntu-devel-discuss@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss
Re: libjudy
> I'm still waiting for my big run to complete, but I have a smaller test Unfortunately, the new executable appears to fail for the same data that the old one did. So if there is indeed a subtle bug hidden in libjudy, it doesn't seem to be affected by these options. I also tried this on multiple servers, so I'm about to rule out hardware error. Thanks for the help, anyway. -k -- If I haven't seen further, it is by standing in the footprints of giants -- Ubuntu-devel-discuss mailing list Ubuntu-devel-discuss@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss
libjudy
Hi, I have written an application which relies on libjudy. Very rarely, I tend to get a segfault or sometimes corrupted data. Now, I realize this could be a bug of my own making, or a problem with the Haskell FFI (my application is written in Haskell), and there are also a suspicion it could be a hardware error (faulty memory, perhaps). But I see this happen on two different servers, and playing around with the code has so far gotten me nowhere. I then stumbled over this discussion, indicating that libjudy is vulnerable to aggressive optimization by certain GCC version. The description looks very much like what I experience, so I thought I'd ask. Is this something that is known and fixed for sure? Or would it be worthwhile to try to look deeper into this? Unfortunately, I don't think I can give a test case, the data involved are several hundred gigabytes. But I (or someone) can build versions of my application linked to versions of libjudy with different optimizations, I can offer to run my analysis, and see if we can avoid the faults. Here's the discussion: https://sourceforge.net/p/judy/mailman/message/31505234/ Thanks for any answer or suggestion! -k -- If I haven't seen further, it is by standing in the footprints of giants -- Ubuntu-devel-discuss mailing list Ubuntu-devel-discuss@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss