IRC channel logs

2025-03-19.log

back to list of logs

<mightysands>Oh cool ! janneke, I read your blogpost about getting guix/hurd working on a Thinkpad X60 !
<mightysands>Mondo cool
<Pellescours>youpi: I don’t know if debian is impacted but I opened a PR to RHash : https://github.com/rhash/RHash/pull/280 This will fix the shared library extension name and the --target value for hurd
<youpi>Pellescours: thanks !
<janneke>mightysands: ty!
<janneke>ACTION hasn't really looked into using audio/video/ using a guix hurd as a daily driver or desktop
<janneke>there's too many things that need to be fixed addressed firest
<gnu_srs1>FYI: I have patches for libdrm and mesa to build most drivers. Current problem is a bus error when running glxdemo/glxgears with llvmpipe. :(
<gnu_srs1>libdrm patches will soon be submitted upstream. However, the mesa patches will not be sent upstream until the llvmpipe problem is solved. Help needed :D
<camm>youpi: you here?
<youpi>camm: I'm now here
<youpi>note: I have updated the preinstalled hurd-amd64 image
<Pellescours>I did a ps aux and I found that compared to Linux, the memory reported are quite high. Do you think it’s expected/normal?
<Pellescours>everything reported in my ps have SZ field with at least 100M
<Pellescours>in Linux I rarely see something reaching the megabyte
<Pellescours>(for "normal" background programs)
<camm>youpi: Thanks! GCL Hurd is looking good. Some questions:
<youpi>Pellescours: SZ is the virtual size. The default heap pre-allocation is 128M, so no wonder. Rather read RSS
<camm>1) gdb hangs if the process forks a child, e.g. popen or posix_spawn
<camm>2) Can hurd-amd64 access heap memory over 0x80000000?
<youpi>1): investigation welcome
<youpi>2): I don't remember the exact figure, but the addressing space for user is very large
<camm>3) Prompted by your earlier comments, I've abandoned mremap in favor of plain mmap. Apparently it can run right over the 386 heap map at 0x20000000 without unmap. Any problems here?
<youpi>3): The lower 3G is fully adressable by userland
<camm>I was under some impression that mmap would fail if there was already an entry in the maps table at that address.
<youpi>the vm_map mach call would, yes, but mmap works around that
<camm>4) My hurd-amd64 kvm cannot be upgraded, and locks up randomly, but basically shows things working (20241115). Any way to stabilize further?
<youpi>use the newer image
<camm>There is a newer image?
<youpi>that's my first message when I came back here
<camm>Great!
<camm>Regarding 1), you are unaware of any existing issue where children work as expected outside of gdb, but under hang at read in open, and waitpid after posix_spawn? Seems like the children are never called. If this is unfamiliar, no worries, will just grab the latest and see what is current.
<youpi>I don't know
<youpi>it's just not possible for people to throw issues at me and how that I would magically have ideas about them or time to work on them
<youpi>sorry for being grumpy, but I've had my 10h workday, I have *not* been able to process my mail for that whole day and now I'm receiving a flurry on irc as well
<camm>Of course, I am just polling you general knowledge -- no need to waste more time at this point, will retest. Thanks for the new image!
<youpi>s/how/hope/
<Pellescours>ok, the RSS is much more reasonable (even if higher than Linux). I see that after a fresh boot, ext2fs has RSS around 6M. But after some compilation and installation of files, I got it being 60M. I don’t know if it seems reasonable, or if it can show a memory leak
<youpi>it contains the page cache
<Pellescours>so normal then
<youpi>if ext2fs had memory leaks, buildds wouldn't be able to build packages for days long :)
<Pellescours>youpi: does m4 build on x86_64? I try to build it but I get error ""Insufficient heuristics for detecting a stack overflow. Either define CFG_STACKVMA and HAVE_STACKVMA correctly, or define SIGSEGV_FAULT_STACKPOINTER correctly, or undefine HAVE_STACK_OVERFLOW_RECOVERY!"
<Pellescours>and for what I see https://sources.debian.org/src/m4/1.4.19-7/lib/sigsegv.c/#L341 it’s effectivelly not defined for x86_64
<azert>gnu_srs1: cool, do you keep the libdrm patches in some public repo? I’d be curious to see your approach to the ioctls
<Pellescours>Ah I see that guix is doing a sed https://git.savannah.gnu.org/cgit/guix.git/tree/gnu/packages/m4.scm#n78
<youpi>Pellescours: see the source package in unreleased which has a patch