(If someone would please donate a usage case (an example of how nested repositories are used). Doing so would help the one person every month that wants to know how to do this with darcs.)

Note: this page is for describing how darcs behaves now wrt to nested repositories. See ProposalForSubrepositories for possible future directions that darcs could take

Quick documentation of the behavior of darcs for nested repositories

Although its usefulness is currently very limited, it is possible to nest repositories in darcs. (Note that the current support for nesting is certainly not flexible enough for you to manage things similar to the way you would with hierarchical systems like CVS and SVN, nor is it intended to)

The basic idea is that a darcs repository will basically ignore any darcs repositories it contains, and any repositories that it is contained in. They don’t interact directly. Changes in the sub-repository will not be seen in the root repository, you have to go into the sub repository to notice them.

Example: If you have repositories A and B nested inside a repository C, moving a file from A to B would result in recording a delete file patch in one repository and an add file patch in the other, with the super-repository C not even noticing the difference.

If you have an existing repository, it is safe to create a repository at a higher level that contains it. However, if you have an existing repository, it is NOT safe to create a sub-repository containing files in the first repository.

Scenario A (creating a parent repository)

An example of this scenario is you have a number of small tools (with preexisting darcs repositories) that you want to group under a single parent repository.

The desired behavior is that you would be able to tell people to darcs get parentProject, which results in both subProj1 and subProj2 being retrieved.

I think that the easiest way to do this is to create config file in parentProject with subrepos and subdirs:

<repo URL for subProject1> <dirname for subProject1>
<repo URL for subProject2> <dirname for subProject2>

and use following script to get the sub-repositories:

set -eu;
if [ ! "$#" -gt 1 ]; then
    echo "Usage: $0 <config file> [<options>]";
    echo "Args: $@";
    exit 1;
exec 5<"$1";
while read -u 5 A B; do
    echo "$B";
    case "$1" in
        get )
            if [ -e "$B" ]; then
       "$B" './';
            darcs "$@" "$A" "$B";
        * )
            #for check, whatsnew, ... commands
            darcs "$@" --repodir="$B" || echo ko "$?";
exit 0;

Where will find new unique name for the sub-repository directory, so we will not lose current changes. We must also update boringfile or else darcs will browse sub-repositories when looking for changes with –look-for-adds option. See the manual for more info about customizing boringfile.

(credit Korusef for the script)

Scenario A’ (creating “sub-repo” nodes within a parent repository)

Again, assume you have a number preexisting darcs repositories that you want to group under a single parent repository.

The desired behavior in this situation is that you want darcs to manage “subrepo nodes”, which are simply empty darcs repositories which you can run “darcs pull -a” to retrieve. Performing pull or push in the parent repository only pushes/pulls the nodes, and does not affect any file or patch in the subrepo.

The trick in this situation is that you want the nodes to be valid darcs repositories as soon as they arrive wherever they are pulled or pushed. This functionality would be best if built-in to darcs, but can be hacked-in using scripts.

First, a usage scenario:

$ darcs get
$ cd projectgroup
$ ls
component1 component2 component3

$ cd component1
$ darcs pull -a
// all patches are retrieved for component1
// edit, record, and then
$ darcs push -a
// now you decide to add a new component
$ cd ..
$ mkdir component4
$ cd component4
$ darcs init
// edit, and record
$ cd ..
$ darcs addrepo component4
$ darcs record -a -m "added subrepo for component 4"
$ darcs push -a
$ cd component4
$ darcs push -a

Note that “darcs addrepo” is not a real command.

And now a script which adds the missing functionality…. I apologize in advance for the ugliness and brittleness of this hack, but with darcs 1.0.5 it works quite nicely. (There are other more reliable ways to pull this off, but aren’t quite as efficient. If you write one, feel free to replace this code)

  1. Install the utility “strace”
  2. Rename $prefix/bin/darcs to $prefix/bin/darcs_binary
  3. Save the following script as $prefix/bin/darcs


# Subroutine to run "darcs init" if needed, on a possible repo stub
InitIfNeeded() {
   if [[ ! -d "$1/_darcs" ]]; then
      if pushd "$1" >/dev/null; then
         darcs init \
           && cp $REPOMARKER _darcs/prefs/repos \
           && head -n 1 $REPOMARKER > _darcs/prefs/defaultrepo \
           && cp _darcs/prefs/defaultrepo _darcs/prefs/lastrepo \
           || echo "Failed to init darcs and make pref files in $PWD" >&2
         popd >/dev/null
         echo "cant change to $1 to run darcs init" >&2

# Decide what to do based on what darcs command is given
case "$1" in
   # each of push/pull/apply needs to search for repo stubs.  I "creatively" do this
   # by collecting a list of all the directories where darcs writes a file named
   # $REPOMARKER and then running the subroutine on them.  In a fit of pure insanity,
   # I also detect the "repo root" by looking for where darcs creates the lock file.
   # It would be tricky to detect this ourselves, though, since there's chances of a
   # misplaced "_darcs".  A proper search would need to use the same hieuristic that
   # darcs itself uses to determine whether a directory named "_darcs" is actually
   # the repo we're in.
   # "darcs query reporoot" would be a nice command to have built-in.
   NEWFILE_TRACE=`mktemp /tmp/darcstrace.XXXXXX`
   strace -e trace=open -o $NEWFILE_TRACE $REAL_DARCS "$@" \
     && sed -ne '/open("\(...*\)\/'$REPOMARKER'.*O_CREAT.*/ s//\1/p' \
            -e '/open("\(...*\)\/_darcs\/darcs_lock.*O_CREAT.*/ s//\1/p' <$NEWFILE_TRACE \
   | while read fname; do
      if [[ -z "$basepath" ]]; then
         InitIfNeeded "$basepath/$fname"
   # This adds the director(y|ies) specified, and sets up the $REPOMARKER file,
   # which contains the default URLs for push/pull.  If a repo URL list has
   # not been created for the subrepo yet, prompt the user for the default repo.
   while [[ -n "$2" ]]; do
      if [[ -d "$dir" && -d "$dir/_darcs" ]]; then
         if [[ -f "$dir/$REPOMARKER" ]]; then
            echo "Subrepo has already been marked" >?
            if [[ -f "$dir/_darcs/prefs/repos" ]]; then
               cp "$dir/_darcs/prefs/repos" "$dir/$REPOMARKER";
               read -p "Enter the default repo path: " apath
               echo "$apath" > "$dir/$REPOMARKER"
            $REAL_DARCS add "$dir/$REPOMARKER";
         echo "$dir is not a darcs repo";
   # all other ocmmands pass through to the real darcs
   $REAL_DARCS "$@";

Note that you must use this hack on each instance of darcs you intend to use: all your client machines and the server as well. If this script is not in place, the “darcs init” will not happen on the repo stubs, and you will not be able to push into them remotely.

Scenario B (creating a sub-repository)

Once again, it is NOT safe to create a sub-repository containing files in the first repository.

If you still want to try this, one way to do this might be to remove the files from the first repository before initializing the sub-repository. Unfortunately, this means that you loose the history of files in the inferior repository, which means that you won’t be able to apply any patches from clones of the complete repository.

Another solution is to duplicate the repository, remove spurious files in both clones, move files in the inferior repository, and move it in place. This will allow applying some patches from clones of the compound repository to either repository (those that don’t span the two repositories).

Both solutions cause major bloat, unless you tag, checkpoint and get –partial after the split.

(credit Juliusz for the above explanation)

Its completely safe to do something like this:

cd parentProject
mkdir subProject
cd subProject
darcs init

Any changed/adds/removes you do in the newProject dir, or its subdirs will be isolated from the myProject repository.

Known issues:

darcs record --look-for-adds

will result in files being added to the wrong repository. The subrepo could end up being added as a normal dir in the root-repo

darcs whatsnew --look-for-adds

will give you (perhaps dangerously) misleading feedback if you have new files in a nested repository.

(credit David Roundy for above warnings)

What exactly are nested repositories?

I thought of this:

mkdir a
mkdir b
cd a/
darcs init
mkdir a
touch a/1 a/2 a/3
darcs add -r a/
darcs record -a -m "a1"
cd ../b/
darcs init
mkdir b
touch b/1 b/2 b/3
darcs add -r b/
darcs record -a -m "b1"


cd ../
mkdir c
cd c/
darcs init
darcs pull ../a/
darcs pull ../b/


cd ../
mkdir d
cd d/
darcs init
darcs pull ../c

a and b never conflict. a and b nest inside c.

Cross repository patches spanning a and b or patches against c should not be recorded. If you need to be able to record patches against c – there is ad hoc approach which is not recommended – a script could read a patch and divide it’s pieces into two piles based on which repository they modify. (This breaks moves which would have to be replaced with a delete and an add. I don’t see a way of getting around this without forgetting about this entire idea and just merging the repositories together which in many ways is a lot simpler. I know the benefits of this are the ability to work on parts individually without having a copy of everything locally, but I believe this benefit is weak.) With this script the example behavior described above at the beginning (the one in the third paragraph ignoring parenthetical thoughts) could be implemented without as much trouble as the prior two “scenarios”.

In this implementation, you don’t nest repositories (as in nesting directories containing a _darcs subdirectory). You create a repository which contains all the patches from one or more other repositories; in other words this repository is the union of several repositories. by giving each repositories files a unique prefix and putting them within a directory with that name results in patches which cannot conflict between repositories.

I don’t remember if I mentioned this elsewhere but to allow repositories without unique prefixes an option like darcs pull --patch-prefix=a would make all of the patches pulled act as if they had been recorded from within a/ in other words probably more clearly and by example 1 would act like a/1


If someone is interested: I wrote a python script which splits the latest patch of a repo into independent patches for any number of given sub directories. If you have any use for it you my contact me on #darcs at (nickname: asmanian) or via email: hhasemann at web dot de. My script tries to handle can handle moves even between subdirectories and binary files, too. But note that its not tested much so make backups of any valuable data before using it. – HenningHasemann