From time to time I find myself using git-svn with projects using subversion, this is all nice and fine but it takes quite some time to do the ''initial import'' if you are doing it over a remote transport such as ssh+svn.

One trick to speed things up is to use rsync to copy the svn repo locally and then ''git-svn'' it from your harddisk. The idea is originally documented at the allegro wiki

For example a typical sourceforge project will be done like this:

DESTDIR=/your/dest
PROJECTNAME=vde
PROJECTURL=https://$PROJECTNAME.svn.sourceforge.net/svnroot/$PROJECTNAME
SVN_COPY=$(mktemp -d)
rsync -c -av $PROJECTNAME.svn.sourceforge.net::svn/$PROJECTNAME/'*' $SVN_COPY
mkdir -p $DESTDIR
cd $DESTDIR
git svn init file://$SVN_COPY --rewrite-root=$PROJECTURL --stdlayout
git svn fetch
# ... git svn imports the repo ...
sed -i "s@url =.*@url = $PROJECTURL@" .git/config
rm -rf $SVN_COPY

The important bit is to remember to update the url key after updating otherwise subsequent commits will go to your rsync-ed repo and not the original one!

I have wrote already about this, now for an improved version some years later.

Nothing new under the sun: you have your /etc under git (I recommend doing that with etckeeper) and you want to be reminded about uncommited changes if you leave the root shell. This is handy for example when co-maintaining a server and everybody is required to leave a trace after changing a file.

So you put something like this in your favourite root shell initialization file (e.g. ~root/.bashrc) and make sure the shell is invoked as interactive/login (e.g. alias su='su -l')

git_functions="/usr/local/sbin/git-etc-common"

# export GIT_* variables
if [ -r "$git_functions" ]; then
  . "$git_functions"
  git_export_env
fi

case $- in
  *i*)    # interactive shell
    check_uncommitted(){
      ttyuser=$(ttyuser)
      logoutfile="$HOME/.logout-$ttyuser"
      # ttyuser is empty if the shell is not connected to a terminal
      # IOW, avoid forkbombing if the terminal has been closed
      if ! git_etc_status && [ ! -e "$logoutfile" ] && [ ! -z "$ttyuser" ]; then
        echo "Uncommitted changes to /etc. touch $logoutfile to force logout."
        # TODO show a list of files to commit?
        $SHELL -$-
      fi
      rm -f $logoutfile
    }

    trap check_uncommitted EXIT
  ;;
esac

And git-etc-common looks like this:

# source this file

ttyuser() {
    cur_tty=$(tty) && echo $(stat -c "%U" $cur_tty)
}

# detect who's currently at the console
git_export_env() {
  # TODO ask for name if GIT_AUTHOR_{NAME,EMAIL} is not set?

  ttyuser=$(ttyuser || echo "")
  if [ -n "$ttyuser" ]; then
    ttyuserhome=$(getent passwd "$ttyuser" | cut -d: -f6)
    conf=$ttyuserhome/.gitconfig
  fi

  if [ -z "$GIT_AUTHOR_NAME" ] && [ -z "$GIT_AUTHOR_EMAIL" ]; then
    if [ ! -z "$GIT_CONFIG_LOCAL" ] || [ ! -z "$GIT_CONFIG" ]; then
      export GIT_AUTHOR_NAME=$(git config --get user.name)
      export GIT_AUTHOR_EMAIL=$(git config --get user.email)
    elif [ -n "$conf" ] && [ -r "$conf" ]; then
      export GIT_AUTHOR_NAME=$(git config --file $conf --get user.name)
      export GIT_AUTHOR_EMAIL=$(git config --file $conf --get user.email)
    fi
  fi
}

git_etc_status() {
  changed=$(cd /etc && git ls-files --modified --deleted --others \
                                    --exclude-per-directory=.gitignore \
                                    --exclude-from=.git/info/exclude)
  ret=$?
  [ -z "$changed" ] && [ "$ret" -eq 0 ]
  return $?
}

Exercise: generalise the functions above so you can have warnings about uncommited changes in generic repositories.

While working on freej we'd need to update spidermonkey/mozjs to the latest version and while doing that integrate it better with the rest of autoconf/automake. AFAICT spidermonkey/mozjs is distributed as part of XULRunner nowadays.

The requirements are to make it easy to put a new version of spidermonkey into the tree and redistribute it with make dist and also to have a successful make distcheck.

The first step is a trick explained in automake manual about third-party Makefiles so we're using a GNUMakefile.in (git) to "proxy" the targets down to the real Makefile or to ignore some targets called by autotools, e.g:

# otherwise linking static libmozjs.a with dynamic libfreej.so won't work
CFLAGS += -fPIC

# the makefile to proxy targets to
js_makefile = $(builddir)/Makefile

# proxy these targets to the real makefile
all export js-config clean libs tools:
  $(MAKE) -f $(js_makefile) $(AM_MAKEFLAGS) $@

# targets required by autotools but which we don't need at all
.PHONY: dvi pdf ps info html installcheck check install uninstall
dvi pdf ps info html installcheck check install uninstall:

This more or less fixes the make part, now for the autoconf part there's a similar tecnique in the autoconf manual involving running arbitrary configuration commands. I first tried to treat spidermonkey as a separate project via AC_CONFIG_SUBDIRS as expained in Configuring Other Packages in Subdirectories but it didn't work and I forgot exactly why.

The configure.ac (git) snippet is something like this: (sorry no highlighting, highlight doesn't support m4 yet)

if test x$have_mozjs = xno || test x$static_mozjs = xyes; then
  dnl run lib/javascript/configure after freej's configure, building it static
  AC_CONFIG_COMMANDS([lib/javascript/.xulrunner-subdir],
    (cd lib/javascript &&
      CXXFLAGS="$GLOBAL_CFLAGS $CXXFLAGS -fPIC" \
      CFLAGS="$GLOBAL_CFLAGS $CFLAGS -fPIC" \
      $ac_srcdir/configure --enable-static \
        --enable-js-static-build --disable-jit) || exit $?
  ])
fi

This will run spidermonkey configure after the main configure, at this point config.status will also expand GNUmakefile.in into GNUmakefile with the correct variables. The final step after all this work is of course to add the spidermonkey directory to SUBDIRS (e.g. SUBDIRS += lib/javascript), after that make will hopefully recurse into lib/javascript and use our GNUmakefile.

While this is nice it doesn't cater for the make dist family of autotools targets which build the real distribution tarball.

To accomplish this I've used an explicit list of spidermonkey files that will end up in the tarball and made make distdir to copy only those files plus GNUmakefile.in and the list itself (.js-distfiles). Which is the second part of the GNUmakefile.in above, i.e. distdir/distfiles targets plus distclean to clean them up, see comments in the file as well.

I hope this is somewhat useful to people who are trying to properly (autotools-wise) ship spidermonkey in their project, comments welcome as usual

Today I was dealing with an instance of GNU mailman without the web interface and there were some messages held for approval.

How to approve them from the command line? Turns out it is possible thanks to the withlist tool shipped with mailman and a bit of fiddling. Code below.

# installation:
# copy this file in the same dir as withlist and name it approvefile.py

# usage:
# bin/withlist -l -r approvemessage <list> <cookie>

# cookie is the digest in the Subject: after approve

import sys
from Mailman import mm_cfg
from Mailman import Pending

# ripped out from MailList.py ProcessConfirmation
def approvemessage(ml, cookie=None):
  if cookie is None:
    print >> sys.stderr, "!!!! No cookie specified"
    return

  rec = ml.pend_confirm(cookie)
  if not rec:
    print >> sys.stderr, "!!!! Cookie not found"
    return

  try:
    op = rec[0]
    data = rec[1:]
  except ValueError:
    print >> sys.stderr, "!!!! No op found for cookie"
    return

  if op != Pending.HELD_MESSAGE:
    print >> sys.stdout, "!!!! Message isn't held"
    return

  ml.HandleRequest(data[0], mm_cfg.APPROVE)
  ml.Save()
  print >> sys.stdout, "!!!! Message approved"

# interactive version:
# bin/withlist -l admin
#Loading list admin (locked)
#The variable `m' is the admin MailList instance
#>>> from Mailman import mm_cfg
#>>> res = m.pend_confirm('862db35189d896667888114e67db8f41fa986c9f')
#>>> m.HandleRequest(res[1], mm_cfg.APPROVE)
#>>> m.Save()

blurb: public domain, no warranty whatsoever, use at your risk. Comments welcome though

This was long due, I switched from pyblosxom to the marvelous ikiwiki and gave a general revamp to my website by tweaking ikiwiki's default templates. note: the template still needs a general cleanup, don't look at it as it is not pretty (nor are the CSSes)

With some luck and mod_rewrite karma nobody's feed readers nor planets will be flooded.

Comments are powered by disqus.

It seems that automake 1.11 made it into unstable and, besides other features, that means silent rules like linux does since quite some time. For the impatient, in configure.ac:

AM_INIT_AUTOMAKE
m4_ifdef([[AM_SILENT_RULES([yes](AM_SILENT_RULES],))])

Then after a round of automake/configure the make invocation will be like

make[3]: Entering directory `foo'
  CXX    source1.lo
  CXX    source2.lo
  CCLD   lib1.la
[...]

Credit note: the autotools mythbuster guide

I'm an avid user of mozilla ubiquity but I couldn't find any commands related to debian to replace those of yubnub. So, prodded by zack I've produced some like bts or pts.

The command feed is located at ubiquity-commands (don't be scared by the "this-code-can-be-evil" subscription page) with the actual code maintained under ubiquity-commands.git.

Note that there's much room for improvement so it would be nice to have a common set of commands for the various debian web resources.

Trying to associate with 00:1c:58:10:1d:30 (SSID='ALMAWIFI' freq=2417 MHz)
                                     ^^^^^
Associated with 00:1c:58:10:15:b0
                            ^^^^^ WTF?

In other words, my broadcom card won't let me choose the access point to connect to, instead the driver picks one for you (my guess: the one with the best signal) except that autentication on that AP doesn't work.

I'm fine with machines trying by default to be smarter than me, I'm less fine when there's no way to convince said machine to do something supposedly not smart.

As explained before by Zack, it might happen that you co-administer a machine and have /etc under some VCS (which you should anyway by using etckeeper).

There are a few problems like detecting the identity of the committer and don't leave /etc with uncommitted changes so I recently came up with a slighly different solution than the one we thought two years ago (besides switching from bzr to git).

Take committer name/email from ~/.gitconfig of the user owning the tty or GIT_CONFIG or GIT_CONFIG_LOCAL if they are set.

Trap a function on shell's EXIT rather than loop on .bash_logout which applies only to login shell. The resulting code to be put in /root/.bashrc:

git_functions="/path/to/git-etc-common"

# export GIT_* variables
if [ -f "$git_functions" ]; then
  . "$git_functions"
  git_export_env
fi

case $- in
*i*)    # interactive shell
  check_uncommitted(){
    if [ -f "$git_functions" ]; then
      . "$git_functions"
      if ! git_etc_status; then
        echo "Uncommitted changes to /etc found, please commit them"
        bash -$-
      fi
    fi
  }
  trap check_uncommitted EXIT
;;
esac

Where git-etc-common is to be found here.

The two RSS feeds previously at http://people.debian.org/~filippo/NEW_rss/ have been moved to a more proper location on ftp-master and the script (tools/queue_rss.py) merged into dak thanks to ganneff.

JFTR the script scans a directory with .changes files (which happens to be NEW in this case) and generates the said feeds, it is not tied to ftp-master and requires python-pyrss2gen and python-debian.