Re: [gentoo-alt] Interix issues

[Michael Haubenwallner]

Hi Ben,

you really want to look at prefix-launcher, specifically svn-trunk[1]
and the wiki[2].
[1] http://prefix-launcher.svn.sourceforge.net/svnroot/prefix-launcher/prefix-launcher/trunk/
[2] http://prefix-launcher.wiki.sourceforge.net/

The released prefix-launcher-1.0 does not work for Interix, but with
svn-trunk, you should get more far. Within prefix-launcher, I already
have working coreutils, findutils, python, bash, sed and others,
inclusive prefix-portage-2.1.23.6381 used to bootstrap an eprefix.

For findutils, currently I completely ignore the mountlist on Interix,
it is not required to bootstrap an eprefix. But once, we could have that
patch in the tree eventually.

Using this portage, I was already able to emerge these packages:

sys-apps/sed-4.1.5
sys-libs/ncurses-5.5-r03.2    (with attached patch)
app-shells/bash-3.2_p17
sys-devel/m4-1.4.9
sys-apps/baselayout-prefix-1.12.5-r4
sys-apps/portage-2.1.27.6600

Currently I'm creating an external toolchain-prefix-wrapper[1] package,
because the source-files are too large (by definition) to be put into
binutils-config/files/ in the tree directly, and recent GNU-ld does not
work, although /opt/gcc.3.3/bin/ld is some patched "GNU ld version
2.13.90 20021111".

Some notes on the wiki for usage of eprefix-bootstrap on not-yet-working
platforms:
*) Use the argument '--onestage', eventually in combination with
'--stage=...' to have some more control over things going on.
*) Check out the portage-tree instead of using the snapshot, manually 
after --stage=tree, and put it in usr/portage

And yes (although not tried myself yet), there's an interix-patch
available for gcc-4.2.0 (c,c++, others unknown):
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15212

/haubi/


On Tue, 2007-06-05 at 01:08 +1000, Ben Peddell wrote:
> I've been trying to build a gentoo installation on Interix.
> I've been following the instructions at
> http://www.gentoo.org/proj/en/gentoo-alt/prefix/bootstrap-macos.xml
>
> First issue I ran into was the bug that I mailed to grobian.
> ===================================================================
> --- bootstrap-prefix.sh.orig    Mon Jun  4 00:53:50 2007
> +++ bootstrap-prefix.sh    Mon Jun  4 01:38:15 2007
> @@ -523,6 +523,7 @@
>                         exit 1
>                      ;;
>                  esac
> +                MAKE=make
>                  ;;
>              *)
>                  eerror "Nothing known about platform `uname -s`."
> ===================================================================
>
> Next issue was with findutils. It couldn't determine how to get the list
> of mounted filesystems.
> I found a patch for gnulib (included in findutils) at
> <http://www.mail-archive.com/bug-coreutils@×××.org/msg05403.html>.
> I don't know whether or not it is GPL clean, as it has not been included
> in gnulib CVS.
> Even so, I have made a correction to it, and added a check in
> "gnulib/m4/ls-mntd-fs.m4".
> See "gnulib-interix.patch".
> I reckon that it would be good if one could enumerate all of the mounts,
> rather than having only one directory to scan.
>
> I am now up to trying to `emerge --oneshot sed`, but am getting "masked
> by: corruption". This could be due to the lacking md5sum, sha1sum, etc.
> On trying to build coreutils to attempt to fix this, I came upon the
> same problem as with findutils.
>
> I'll probably ask bug-gnulib@×××.org what happened to the patch for
> Interix - whether it is clean or not.
>
> I'll also get back on the progress of the install. I'll most likely have
> to bring Interix back to a clean slate, as I had attempted to install a
> non-prefix gentoo on it.
>
> Keep up the good work.
> plain text document attachment (gnulib-interix.patch)
> --- findutils-4.2.27/findutils-4.2.27/gnulib/m4/ls-mntd-fs.m4~	Mon Jun  4 21:05:21 2007
> +++ findutils-4.2.27/findutils-4.2.27/gnulib/m4/ls-mntd-fs.m4	Mon Jun  4 22:35:09 2007
> @@ -278,6 +278,32 @@
>  fi
>
>  if test -z "$ac_list_mounted_fs"; then
> +  # INTERIX
> +  AC_CHECK_HEADERS(dirent.h)
> +  AC_CHECK_HEADERS(sys/statvfs.h)
> +  AC_MSG_CHECKING([whether the system is Interix])
> +  AC_CACHE_VAL(fu_cv_sys_mounted_interix,
> +    [
> +      if test $ac_cv_header_dirent_h = yes \
> +          && test $ac_cv_header_sys_statvfs_h = yes; then
> +        AC_TRY_CPP([#include <interix/interix.h>],
> +                    fu_cv_sys_mounted_interix=yes,
> +                    fu_cv_sys_mounted_interix=no)
> +      else
> +        fu_cv_sys_mounted_interix=no
> +      fi
> +    ])
> +  AC_MSG_RESULT($fu_cv_sys_mounted_interix)
> +  if test $fu_cv_sys_mounted_interix = yes; then
> +    ac_list_mounted_fs=found
> +    AC_DEFINE(MOUNTED_INTERIX, 1,
> +              [Define if the system is Interix. This system has all of
> +               the mounted filesystems, other than /, /net and /proc, under
> +               /dev/fs. (Interix) ])
> +  fi
> +fi
> +
> +if test -z "$ac_list_mounted_fs"; then
>    # SVR2
>    AC_MSG_CHECKING([whether it is possible to resort to fread on /etc/mnttab])
>    AC_CACHE_VAL(fu_cv_sys_mounted_fread,
> --- findutils-4.2.27/findutils-4.2.27/gnulib/lib/mountlist.c~	Mon Jun  4 21:05:35 2007
> +++ findutils-4.2.27/findutils-4.2.27/gnulib/lib/mountlist.c	Mon Jun  4 22:34:58 2007
> @@ -35,7 +35,9 @@
>
>  #include <errno.h>
>
> -#include <fcntl.h>
> +#ifdef HAVE_FCNTL_H
> +# include <fcntl.h>
> +#endif
>
>  #ifdef HAVE_UNISTD_H
>  # include <unistd.h>
> @@ -93,6 +95,11 @@
>  # include <dirent.h>
>  #endif
>
> +#ifdef MOUNTED_INTERIX		/* Interix */
> +# include <dirent.h>
> +# include <sys/statvfs.h>
> +#endif
> +
>  #ifdef MOUNTED_FREAD		/* SVR2.  */
>  # include <mnttab.h>
>  #endif
> @@ -418,6 +425,95 @@
>        goto free_then_fail;
>    }
>  #endif /* MOUNTED_GETMNT. */
> +
> +#if defined MOUNTED_INTERIX
> +  /* Windows with Interix / Microsoft Services For UNIX */
> +  {
> +    /* Unlike Cygwin which attempts to implement as many of the UNIX
> +       API's as possible Interix implements a great deal and then depends
> +       on developers to port the rest. On Interix, the POSIX implementation
> +       prefers to deny that UNIX style file systems exist and therefore
> +       does not implement the mount tables for the system.
> +
> +       This implementation simply scans the mounted file system directory
> +       and then reads the statvfs data for each entry to construct the
> +       mount list.
> +
> +       The implementation also chooses to use the reentrant implementation
> +       of readdir_r in order to favor a threading friendly system. */
> +
> +    int statvfs_errno = 0;
> +    int fail;
> +
> +    DIR *dirp = opendir ("/dev/fs");
> +    if (dirp == NULL)
> +      {
> +       fail = 1;
> +      }
> +    else
> +      {
> +       char file_name[9 + NAME_MAX]; /* 8 for /dev/fs/ + 1 for NUL */
> +       int saved_errno;
> +
> +       while (1)
> +         {
> +           struct statvfs stat_buf;
> +           struct dirent entry;
> +           struct dirent *result;
> +
> +           fail = readdir_r (dirp, &entry, &result);
> +
> +           if (fail || result == NULL)
> +             break;
> +
> +           strcpy (file_name, "/dev/fs/");
> +           strcat (file_name, entry.d_name);
> +
> +           fail = statvfs (file_name, &stat_buf);
> +           if (fail == 0)
> +             {
> +               char const *Magic_mount_from_name
> +                 = "/Device/LanmanRedirector/";
> +               me = xmalloc (sizeof *me);
> +               me->me_devname = xstrdup (stat_buf.f_mntfromname);
> +               me->me_mountdir = xstrdup (stat_buf.f_mntonname);
> +               me->me_type = xstrdup (stat_buf.f_fstypename);
> +               me->me_type_malloced = 1;
> +               me->me_dummy = 0;
> +               me->me_dev = stat_buf.f_fsid;
> +               me->me_remote
> +                 = strncmp (stat_buf.f_mntfromname, Magic_mount_from_name,
> +                            strlen (Magic_mount_from_name)) == 0;
> +
> +               /* Add to the linked list. */
> +               *mtail = me;
> +               mtail = &me->me_next;
> +             }
> +           else
> +             {
> +               statvfs_errno = errno;
> +             }
> +         }
> +
> +       saved_errno = errno;
> +       closedir (dirp);
> +       errno = saved_errno;
> +      }
> +
> +    if (!fail && statvfs_errno)
> +      {
> +       /* In the unlikely event that opendir and each readdir
> +          succeed, but all statvfs calls fail, ensure that we
> +          fail with a valid errno value.  */
> +       fail = 1;
> +       errno = statvfs_errno;
> +      }
> +
> +    if (fail)
> +      goto free_then_fail;
> +
> +  }
> +#endif
>
>  #if defined MOUNTED_FS_STAT_DEV /* BeOS */
>    {