path: root/hurd/running/nix.mdwn

[[!meta copyright="Copyright © 2012, 2013, 2014, 2016 Free Software Foundation,
Inc."]]

[[!meta license="""[[!toggle id="license" text="GFDL 1.2+"]][[!toggleable
id="license" text="Permission is granted to copy, distribute and/or modify this
document under the terms of the GNU Free Documentation License, Version 1.2 or
any later version published by the Free Software Foundation; with no Invariant
Sections, no Front-Cover Texts, and no Back-Cover Texts.  A copy of the license
is included in the section entitled [[GNU Free Documentation
License|/fdl]]."]]"""]]

[[!meta title="Nix-based GNU/Hurd System, Guix"]]

[[!toc levels=2]]


# Nix

<http://www.nixos.org/>

  * <http://hydra.nixos.org/jobset/gnu/hurd-master>

      * <http://hydra.nixos.org/job/gnu/hurd-master/qemu_image/latest/download>

  * <http://hydra.nixos.org/job/gnu/hurd-master/qemu_test>

This QEMU image is not (yet) comparable to NixOS, because the latter provides
extra features, such as whole-system configuration (including services, etc.),
and whole-system transactional update and rollback.  It is is cross-built using
Nix, and because of that, it uses per-package installation directories under
`/nix/store`.


## IRC, freenode, #hurd, 2013-02-04

    <braunr> is it possible to use nix ?
    <braunr> or nixos
    <civodul> you mean the Nix-based Hurd image?
    <braunr> yes
    <civodul> it's currently broken:
      http://hydra.nixos.org/jobset/gnu/hurd-master
    <braunr> aw, nixos uses systemd :(
    <civodul> yeah, but the Hurd image is a different thing
    <civodul> is uses runsystem.sh™
    <civodul> (my favorite)
    <civodul> i've been willing to unbreak it, but now i rather invest time in
      Guix


# <a name="guix">Guix</a>

## <http://www.gnu.org/software/guix/>

## IRC, freenode, #hurd, 2014-02-13

    <phant0mas> is debian hurd the only way to use hurd?
    <braunr> maybe
    <braunr> there is arch hurd but i haven't heard from them in some time
    <braunr> building from source is difficult
    <phant0mas> what is the problem with building from source?
    <braunr> no automated build system, except for debian
    <braunr> each project has its own build system
    <braunr> but there is no tool to take care of the global procedure
      (i.e. build in the correct order, with the correct options, paths and
      patches, etc...)
    <youpi> well, there is, it's called Debian :)
    <braunr> 00:17 < braunr> no automated build system, except for debian
    <phant0mas> how far away is it  from building a gnu system with let's say
      guix as a package manager?
    <phant0mas> and hurd as the kernel?
    <braunr> i don't know
    <youpi> phant0mas: there are already proofs of concepts
    <phant0mas> youpi: any more info about the proofs of concepts?
    <youpi> phant0mas: ask civodul
    <youpi> apparently he's not here atm, though
    <phant0mas> I will ask him at guix channel


## IRC, freenode, #hurd, 2014-02-14

    <phant0mas> can I ask a question about configuring gnu mach from source?
    <taylanub> phant0mas: IRC etiquette: don't ask to ask, just ask, it saves
      time.  People often leave IRC open for long durations but don't always
      check it, if you just leave your question in the channel someone might
      come back and see it at any time (even hours later).
    <phant0mas> when I try to configure gnumach with 
    <phant0mas> CPP='gcc -m32 -E -x c -undef -ansi' CC='gcc -m32' LD='ld
      -melf_i386' ./configure --prefix= --host=i686-unknown-linux-gnu
    <phant0mas> on a 64 bit system
    <phant0mas> I get the error 
    <phant0mas> checking for i686-unknown-linux-gnu-gcc... gcc -m32
    <phant0mas> checking whether the C compiler works... no
    <phant0mas> configure: error: in `/home/manolis/Downloads/gnumach-1.4':
    <phant0mas> configure: error: C compiler cannot create executables
    <phant0mas> but if I remove the -m32 from CC='gcc -m32' the error
      disappears
    <phant0mas> what is the problem?
    <braunr> what do you think there is a problem ?
    <braunr> i don't think -m32 should be part of CPP/CC, but rather CPPFLAGS
    <braunr> also, i don't think you need it since the target is well defined
    <phant0mas> I am following exaclty the instruction from here
      http://www.gnu.org/software/hurd/microkernel/mach/gnumach/building.html
    <braunr> hm that's weird
    <braunr> phant0mas: what does gcc -v says and what's your host system ?
    <phant0mas> Using built-in specs.
    <phant0mas> COLLECT_GCC=gcc
    <phant0mas>
      COLLECT_LTO_WRAPPER=/usr/lib/gcc/x86_64-unknown-linux-gnu/4.8.2/lto-wrapper
    <phant0mas> Target: x86_64-unknown-linux-gnu
    <phant0mas> Configured with: /build/gcc/src/gcc-4.8-20131219/configure
      --prefix=/usr --libdir=/usr/lib --libexecdir=/usr/lib
      --mandir=/usr/share/man --infodir=/usr/share/info
      --with-bugurl=https://bugs.archlinux.org/
      --enable-languages=c,c++,ada,fortran,go,lto,objc,obj-c++ --enable-shared
      --enable-threads=posix --with-system-zlib --enable-__cxa_atexit
      --disable-libunwind-exceptions --enable-clocale=gnu
      --disable-libstdcxx-pch --disable-libssp --enable-gnu-unique-
    <phant0mas> object --enable-linker-build-id --enable-cloog-backend=isl
      --disable-cloog-version-check --enable-lto --enable-plugin
      --enable-install-libiberty --with-linker-hash-style=gnu
      --disable-multilib --disable-werror --enable-checking=release
    <phant0mas> Thread model: posix
    <phant0mas> gcc version 4.8.2 20131219 (prerelease) (GCC) 
    <braunr> check config.log for the actual error message then
    <phant0mas> http://pastebin.com/raw.php?i=eQ75qafX 
    <phant0mas> if you want to have a look as well
    <braunr> install gcc-multilib maybe
    <braunr> or lib32gcc1
    <sjbalaji> Installing gcc-multilib solves it.
    <phant0mas> braunr: it works like a charm now
    <braunr> phant0mas: good


## IRC, freenode, #hurd, 2014-02-18

    <phant0mas> why mig has no make target, and the executable gets generated
      from the ./configure?
    <phant0mas> I mean shouldn't make be the one building mig?


## IRC, freenode, #hurd, 2014-02-19

    <phant0mas> mig binary shouldn't be built by make? why is it built at the
      end of the configure command?
    <braunr> no idea
    <phant0mas> http://pastebin.com/raw.php?i=2HVni53Y
    <braunr> "creating mig" you mean ?
    <phant0mas> loot at the end of the config output
    <phant0mas> config.status: creating mig <--
    <phant0mas> the binary
    <phant0mas> and then when you call make
    <phant0mas> it says no target
    <braunr> weird
    <phant0mas> normally binaries are built from make
    <braunr> what system are you building on ?
    <phant0mas> on Arch Linux x86_64 with gcc version 4.8.2
    <phant0mas> I am using the flag i686-pc-gnu to crossbuild it for 32 bit
    <phant0mas> I am using the tar file from here
      http://ftp.gnu.org.ua/gnu/mig/
    <phant0mas> version 1.4
    <braunr> tar file ?
    <braunr> ok, i guess it's fine
    <phant0mas> so source from the tar file builds mig through configure?
    <braunr> again, i don't know
    <braunr> i never build mig myself
    <braunr> but it does look weird
    <braunr> look at the debian package maybe
    <braunr> who knows, maybe you'll find a patch with some explanation
    <phant0mas> okay then,going over that way right away
    <phant0mas> thnx braunr
    <teythoon> phant0mas: mig is a shell script wrapper
    <phant0mas> so it's not a binary....
    <teythoon> no


## IRC, freenode, #hurd, 2014-02-21

    <phant0mas> do I need some minimal set of drivers to build gnumach?
    <phant0mas> because I get this error
      ../linux/src/drivers/scsi/BusLogic.c:53:24: fatal error: FlashPoint.c: No
      such file or directory
    <phant0mas> when running make
    <teythoon> i thought we fixed that
    <teythoon> are your sources up to date ?
    <phant0mas> I am using the tar ball
    <phant0mas> cause I am trying to package it for guix
    <teythoon> what tarball ?
    <braunr> 1.4
    <phant0mas> yes
    <braunr> phant0mas: just don't build scsi drivers
    <phant0mas> 1.4
    <phant0mas> worked
    <teythoon> why do we keep the driver if it doesn't even build ?
    <gg0> on debian it builds with --disable-net-group --disable-pcmcia-group
      --disable-wireless-group


## IRC, freenode, #hurd, 2014-02-23

    <phant0mas> why when I configure gnumach like this CPP='gcc -m32 -E -x c
      -undef -ansi' CC='gcc -m32' LD='ld -melf_i386' ./configure --prefix=
      --host=i686-unknown-linux-gnu it builds just fine but when I try to
      configure it with ./configure --prefix= --host=i686-unknown-linux-gnu
      CFLAGS='-m32' CPPFLAGS='-m32 -E -x c -undef -ansi' LDFLAGS='-melf_i386'  
    <phant0mas> I am building it on a 64 bit machine
    <phant0mas> when setting env vars before configuring everythings works like
      a charm
    <phant0mas> but if I pass them as flags to the configure ,it won't


## IRC, freenode, #hurd, 2014-02-25

    <phant0mas> what version of mae do I need in order to compile glibc for
      hurd?
    <phant0mas> make*
    <azeem> phant0mas: did you have issues with a particular version?
    <azeem> I believe GNU make is required, though
    <phant0mas> I am using gnu make 4.0 and I get the error 
    <phant0mas> These critical programs are missing or too old: make
    <phant0mas> checking version of gmake... 4.0, bad
    <azeem> phant0mas: that sounds bogus
    <azeem> can you pastebin the relevant part of the config.log or so?
    <phant0mas> of course one sec
    <phant0mas> http://pastebin.com/raw.php?i=4CHZJi4W
    <phant0mas> azeem_: any news about the problem I have with glibc?
    <azeem_> phant0mas: sorry - got distracted - I suggest you post this to
      bug-hurd or so
    <azeem_> though it could well be Hurd-independent, then checking glibc
      master and possibly filing a report there might be better
    <azeem_> phant0mas: or in case you speak autoconf, check the conf check for
      the make version


## IRC, freenode, #hurd, 2014-02-26

    <tschwinge> ph4nt0mas: Which glibc sources, and how are you
      configuring/building?
    <phant0mas> tschwinge: I am trying to crossbuild it from a linux 64 bit
      machine with gnu make 4.0 and gcc 4.8.2 
    <phant0mas> ../configure --prefix=/home/manolis/gnu/glibc/
      --target=i686-pc-gnu
    <phant0mas> wrong config
    <phant0mas> ../configure --prefix=/home/manolis/gnu/glibc/
      --with-headers=/home/manolis/gnu/include/ --host=i686-pc-gnu
    <phant0mas> I am using the last one
    <phant0mas> it says gnu make is too old
    <phant0mas> this time I tried with glibc from the hurd git repo
    <phant0mas> baseline branch
    <phant0mas> should I build an older toolchain just for it?
    <tschwinge> phant0mas: If you'd like to experiment with this, then please
      use the tschwinge/Roger_Whittaker branch.  However, that one might still
      be missing the glibc upstream change to allow GNU Make 4.0 and greater,
      commit 28d708c44bc47b56f6551ff285f78edcf61c208a, so cherry-pick that one
      (assuming there are no additional patches needed for GNU Make 4.0).
    <phant0mas> okay going to  do that right away
    <phant0mas> thnx :-)
    <tschwinge> phant0mas: You have seen
      http://www.gnu.org/software/hurd/toolchain/cross-gnu.html -- but beware
      this may be out of date somewhat.
    <phant0mas> tschwinge: I worked around that problem as you told me
    <phant0mas> but it seems I forgot to build hurd 
    <phant0mas> so I got the tarball
    <phant0mas> but I am getting this error
    <phant0mas> No rule to make target 'lowlevellock.h', needed by 'timefmt.o'.
      Stop.
    <phant0mas> after I manage to make all of this work I will write an up to
      date guide on how to build the hurd  system
    <phant0mas> for future reference


## IRC, freenode, #hurd, 2014-02-27

    <ph4n70m4s> when trying to build gnumach microkernelin a 32 bit enviroment
      it builds just fine
    <ph4n70m4s> but when i try to crossbuild it from a 64 bit machine, even
      though i am using --target=i686-gnu --host=i686-gnu  to crossbuild it I
      am just getting the error i386/i386/i386asm.symc:116:1: warning: asm
      operand 0 probably doesn't match constraints [enabled by default]
    <ph4n70m4s>  __asm ("\n\
    <ph4n70m4s>  ^
    <ph4n70m4s> i386/i386/i386asm.symc:116:1: error: impossible constraint in
      'asm'
    <ph4n70m4s> http://pastebin.com/raw.php?i=emag63N4 
    <ph4n70m4s> the config.log
    <ph4n70m4s> and the build output if anyone want to have a look
    <ph4n70m4s> http://pastebin.com/raw.php?i=jAZPnybB
    <ph4n70m4s> wants*
    <ph4n70m4s> I am building through guix so you may see some strange paths 
    <tschwinge> ph4n70m4s: Nice, Guix!  :-)
    <tschwinge> ph4n70m4s: Does that help?  CC=gcc\ -m32
    <tschwinge> Because:
    <tschwinge> configure:3574: gcc -v >&5
    <tschwinge> Using built-in specs.
    <tschwinge> COLLECT_GCC=gcc
    <tschwinge>
      COLLECT_LTO_WRAPPER=/nix/store/7awbhk5hdkd4lqj4wsj6lm6h84630jhm-gcc-4.8.2/libexec/gcc/x86_64-unknown-linux-gnu/4.8.2/lto-wrapper
    <tschwinge> Target: x86_64-unknown-linux-gnu
    <teythoon> guix looks nice
    <tschwinge> ph4n70m4s: Also you don't need --target with GNU Mach -- it
      doesn'T target any architecture, that is, doesn'T create code or similar.
    <tschwinge> Alternative to -m32, as you're specying --host=i686-gnu, you
      could have (that is, you'd need) a i686-gnu-gcc (that internally defaults
      to -m32 then).
    <tschwinge> ph4n70m4s: Are you generally working on GNU Hurd support for
      Guix, or just trying to build GNU Mach?
    <ph4n70m4s> I am working on porting gnu guix to gnu hurd
    <ph4n70m4s> so I need to port hurd to it
    <ph4n70m4s> bootstrap it
    <ph4n70m4s> and the make guix work on hurd
    <ph4n70m4s> :-)
    <teythoon> very cool :)
    <tschwinge> \o/
    <ph4n70m4s> then*
    <ph4n70m4s> and If I manage to do all that we will be able to make a qemu
      image 
    <ph4n70m4s> with hurd -guix
    <tschwinge> Yep, and then I'll be happy.  :-)
    <ph4n70m4s> :-)
    <tschwinge> For then we'll be easily able to change some detail in, say,
      glibc, rebuild the whole system, and see whether it still works.
    <ph4n70m4s> exaclty
    <tschwinge> Of course, that doesn't strictly need Guix, but it's one way to
      achieve this goal.
    <tschwinge> For the initial cross compiler toolchain, I suggest you look at
      my cross-gnu script (link provided yesterady) -- the gist of it should
      all still be valid.
    <ph4n70m4s> I do 
    <tschwinge> For example, you don't actually need to build GNU Mach
      initially, but just need to install the headers/defs files
      (install-data).
    <tschwinge> Perfect.
    <ph4n70m4s> I have already done that, installing gnu mach headers was the
      easy part
    <ph4n70m4s> it's already on guix
    <tschwinge> \o/
    <tschwinge> Should really clone Git repository.  And subscribe to the
      mailing list, and all that.
    <ph4n70m4s> I have already done that
    <ph4n70m4s> I am actually following hurd from last year
    <tschwinge> Yeah, I was talking about me.  ;-)
    <ph4n70m4s> aaaa
    <ph4n70m4s> :P
    <tschwinge> ph4n70m4s: Are you doing this as a Google Summer of Code
      project?
    <ph4n70m4s> I suggested the idea for the page
    <ph4n70m4s> to ludovic
    <ph4n70m4s> and he agreed
    <tschwinge> :-)
    <tschwinge> Good guy.
    <tschwinge> Both of you.  ;-)
    <ph4n70m4s> trying my best to help :-)
    <ph4n70m4s> tschwinge: so to build glibc I only need mach and hurd headers 
    <ph4n70m4s> right?
    <braunr> yes


## IRC, freenode, #hurd, 2014-02-28

    <phant0mas> tschwinge: In your cross-gnu script while configuring hurd you
      pass--disable-profile 
    <phant0mas>  --without-parted . Why do we need to pass these flags to it?
    <teythoon> well, --disable-profile turns off profiling afaiui
    <teythoon> you should keep it
    <teythoon> --without-parted is not needed if you have libparted
    <phant0mas> ok


## IRC, freenode, #hurd, 2014-03-01

    <phant0mas> The hurd tarball has as a dependency autoconf
    <phant0mas> normally it shouldn't, right?
    <azeem> phant0mas: you mean there's no configure script?
    <phant0mas> no no, for some reason the configure script states as
      dependency autoconf which it shouldn't as it is a tarball
    <azeem> phant0mas: how does it state that?
    <phant0mas> you already have the configure script in it
    <azeem> you get an error running it saying autoconf is required?
    <phant0mas> yes
    <azeem> hrm
    <phant0mas> actually it gives an error for autoconf and git
    <phant0mas> but if you have autoconf ,it forgets about git


## IRC, freenode, #hurd, 2014-03-02

    <phant0mas> youpi: civodul told me you are the one to ask about libpthread 
    <phant0mas> how should I handle Hurd's libpthread in order to build it,with
      the rest of the hurd system?
    <phant0mas> anything I should be extra carefull?
    <phant0mas> I am building it from the tarball
    <youpi> phant0mas: nothing I can think of


## IRC, freenode, #hurd, 2014-03-03

    <phant0mas> youpi: what does libpthread do when we do "make
      install-data-local-headers" ?
    <phant0mas> Does it patch the mach headers in the prefix folder?
    <phant0mas> tschwinge: why do we need this flag passed to configure at
      libpthread "ac_cv_lib_ihash_hurd_ihash_create"?  
    <youpi> phant0mas: I don't remember such detajils of the build system
    <phant0mas> I am studying the cross-gnu script from tschwinge
    <phant0mas> and if I try to call configure without that flag I am getting
      the error  
    <phant0mas> configure: error: need libihash
    <tschwinge> Well, there's this comment: # `$TARGET-gcc' doesn't work yet
      (to satisfy the Autoconf checks), but [...]
    <tschwinge> At this point we're only intested in the header files, so that
      was the "path of least resistance".
    <phant0mas> so we are only interested in the headers of libpthread?
    <phant0mas> do I need to pass as --prefix the folder which the include
      folder with the mach headers are installed?
    <tschwinge> I wonder whether it'd be better for you to directly include
      libpthread in glibc, as Debian is doing -- probably yes.

[[open_issues/packaging_libpthread]].

    <phant0mas> maybe it would be much simpler that way
    <phant0mas> let me check the debian package
    <braunr> i'd consider it mandatory, since libpthread can't work correctly
      now if it's not part of glibc
    <tschwinge> Ack.


## IRC, freenode, #hurd, 2014-03-05

    <teythoon> braunr: did i understand it correctly that our libpthread needs
      libihash ?
    <teythoon> if so, won't that be a problem for phant0mas bootstrapping
      efforts ?
    <braunr> apparently, they're used only for thread-specific data
    <braunr> which could now be implemented on top of TLS
    <braunr> although i'm not sure
    <braunr> but to answer, yes, it depends on libihash
    <braunr> why would it be a problem ?
    <teythoon> b/c it then forms a dependency loop
    <braunr> which one ?
    <teythoon> hurd->libc (which includes libpthread, thus)->hurd
    <braunr> isn't that already the case ?
    <teythoon> yes
    <braunr> what's the problem then ?
    <braunr> actually, it's not a dependency loop
    <teythoon> y not ?
    <braunr> hurd and libc depend on each other headers
    <teythoon> ah, right
    <braunr> actually, hurd-libs depend on libc headers
    <braunr> hurd executables do depend on libc
    <braunr> it's a bit tedious to bootstrap but not much more than the
      three-step build required for linux already
    <phant0mas> actually for that libihash if I pass the flag
      "ac_cv_lib_ihash_hurd_ihash_create=yes" at configure it stops asking for
      it

    <phant0mas> when glibc's configure says "configure: running configure
      fragment for add-on libpthread"
    <phant0mas> does it mean it runs configure inside the folder of libpthread?
    <youpi> I guess so
    <phant0mas> yeh because I am getting the error configure: error: cannot
      find sources (include/pthread/pthread.h) in .. or ..
    <phant0mas> which I should not as I am passing
    <phant0mas> the path to that folder
    <phant0mas> it's an error from libpthread configure


## IRC, freenode, #glibc, 2014-03-05

    <phant0mas> when we pass to the configure script of glibc the flag
      --enable-add-ons=something what it the process followed for that addon?
      How does it build it?
    <phant0mas> what is*


## IRC, freenode, #hurd, 2014-03-05

    <phant0mas> ../configure --without-cvs --build=i686-pc-gnu
      --host=i686-pc-gnu --target=i686-pc-gnu --prefix="/home/manolis/gnu/"
      --with-headers="/home/manolis/gnu/include/" --disable-profile
      --disable-multi-arch --enable-add-ons=libpthread
    <phant0mas> trying to configuring glibc with libpthreads as addon I get
      this
    <phant0mas> configure: running configure fragment for add-on libpthread
    <phant0mas> configure: WARNING: you should use --build, --host, --target
    <phant0mas> configure: WARNING: you should use --build, --host, --target
    <phant0mas> configure: error: cannot find sources
      (include/pthread/pthread.h) in .. or ..
    <phant0mas> without libpthread it moves on
    <phant0mas> in the include folder it should find al the headers it needs
    <phant0mas> maybe it's a problem that I installed the headers from the
      tarballs
    <phant0mas> while glibc is from the git repo
    <phant0mas> if I remove libpthread
    <phant0mas> I get the error configure: error: Hurd headers not installed or
      too old
    <phant0mas> I really think I should reinstall the headers using the source
      from the repo the headers 
    <phant0mas> the headers from the tar ball are okay
    <phant0mas> when I pass this flag to "--enable-add-ons=libpthread" to glibc
      configure how can I pass flags for the libpthread configure that it's
      called while configuring? 
    <teythoon> phant0mas: you could look at what the debian package does
    <phant0mas> ok
    <braunr> phant0mas: check debian glibc for all the patches


## IRC, freenode, #hurd, 2014-03-10

    <ph4n70m4s> tschwinge: While crosscompiling glibc I get the error "Error:
      incorrect register `%rax' used with `l' suffix"
    <ph4n70m4s> http://pastebin.com/raw.php?i=ZJKHrm4s
    <ph4n70m4s> Any idea why is this happening?
    <braunr> ph4n70m4s: something is trying to compile as an x86-64 object,
      while the hurd is i386 only


## FOSDEM 2016

  * {{$community/meetings/fosdem_2016#ragkousis_guix_hurd}}