UP | HOME

Table of Contents

1. Problem

We're using git subtree with xo-umbrella2, to maintain independent satellite repos (such as xo-indentlog).

Encountered unexpected error from git subtree split: 

$ cd ~/proj/xo-umbrella2
$ git subtree split --rejoin --prefix=xo-indentlog -b _dmeux/xo-indentlog
fatal: cache for 624178f1932508a687d85ddea56d03998193207a already exists!

$ which git
/home/roland/nixroot/nix/store/jwv5hg4kdb322qi0y9ss6xjx94bgxh8l-git-2.50.1/bin/git

Initially puzzling, because the 624178 commit doesn't actually intersect the xo-indentlog prefix

2. Diagnosis

Make a copy of the subtree implementation, so we can modify it. 

$ mkdir -p /tmp/git-exec-real
$ cp -r $(git --exec-path) /tmp/git-exec-real
$ chmod -R u+w /tmp/git-exec-real

Verify that it runs 

$ GIT_EXEC_PATH=/tmp/git-exec-real git subtree split --prefix=xo-indentlog -b _demux/xo-indentlog

Find the error message in /tmp/git-exec-real/git-subtree (about line 337). 

# Usage: cache_set OLDREV NEWREV
cache_set () {
    assert test $# = 2
    oldrev="$1"
    newrev="$2"
    if test "$oldrev" != "latest_old" &&
        test "$oldrev" != "latest_new" &&
        test -e "$cachedir/$oldrev"
    then
        die "fatal: cache for $oldrev already exists!"
    fi
    echo "$newrev" >"$cachedir/$oldrev"
}

add some diagnostics output 

# Usage: cache_set OLDREV NEWREV
cache_set () {
    assert test $# = 2
    oldrev="$1"
    newrev="$2"
    if test "$oldrev" != "latest_old" &&
            test "$oldrev" != "latest_new" &&
            test -e "$cachedir/$oldrev"
    then
        echo "DEBUG: dup cache entry for $oldrev" >&2
        echo "       newrev=$newrev" >&2
        echo "       existing=$(cat $cachedir/$oldrev)" >&2
        die "fatal: cache for $oldrev already exists!"
    fi
    echo "$newrev" >"$cachedir/$oldrev"
}


$ GIT_EXEC_PATH=/tmp/git-exec-real git subtree split --prefix=xo-indentlog -b _demux/xo-indentlog
DEBUG: dup cache entry for 624178f1932508a687d85ddea56d03998193207a
       newrev=624178f1932508a687d85ddea56d03998193207a
       existing=624178f1932508a687d85ddea56d03998193207a

Well, that's provocative! All 3 commits are the same!

That's consistent with commit not having any changes that intersect with the xo-indentlog/ prefix.

3. Treatment

Problem is that we have multiple merge commits in scope. In particular: merge commits created because: before running split for xo-indentlog, had run it for sibling xo-cmake.

In any case obvious correction is to suppress the error when git subtree split arrives at the same commit on multiple paths. Edit /tmp/git-exec-real/git-subtree as shown: 

# Usage: cache_set OLDREV NEWREV
cache_set () {
    assert test $# = 2
    oldrev="$1"
    newrev="$2"
    if test "$oldrev" != "latest_old" &&
        test "$oldrev" != "latest_new" &&
        test -e "$cachedir/$oldrev"
    then
        existing=$(cat "$cachedir/$oldrev")
        if test "$existing" = "$newrev"
        then
            # same mapping
            #  -> commit doesn't touch prefix
            #  -> must not cache
            return
        else
            die "fatal: cache for $oldrev already exists!"
        fi
    fi
    echo "$newrev" >"$cachedir/$oldrev"
}

..and try out the patch: 

$ GIT_EXEC_PATH=/tmp/git-exec-real git subtree split --prefix=xo-indentlog -b _demux/xo-indentlog
...
$

4. Followup

4.1. xo-umbrella2 nix shells

To maintain this patch locally, add to nix shell setup for xo-umbrella2.

Add xo-umbrella2/patches/git-subtree-cache-fix.patch: 

--- a/git-subtree
+++ b/git-subtree
@@ -341,7 +341,13 @@ cache_set () {
        if test "$oldrev" != "latest_old" &&
                test "$oldrev" != "latest_new" &&
                test -e "$cachedir/$oldrev"
        then
-               die "fatal: cache for $oldrev already exists!"
+               existing=$(cat "$cachedir/$oldrev")
+               if test "$existing" = "$newrev"
+               then
+                       return
+               else
+                       die "fatal: cache for $oldrev already exists!"
+               fi
        fi
        echo "$newrev" >"$cachedir/$oldrev"
}

Then modify xo-umbrella2/default.nix so it creates git with the patch applied. We don't add this as an overlay, since we don't need/want working nix packages to change. 

let
  pkgs = import nixpkgs-patch {
    # overlays..
  };

  # bona fide bux in git 2.50.1
  patched-git = pkgs.git.overrideAttrs (old: {
      postInstall = (old.postInstall or "") + ''
        patch $out/libexec/git-core/git-subtree ${./patches/git-subtree-cache-fix.patch}
      '';
    });
in
let
  devutils = [
    patched-git # instead of pkgs.git
    pkgs.catch2
    # ..etc..
  ]

4.2. upstream git repo

Clone git repo, and make a branch 

$ cd ~/proj
$ git clone https://githjub.com/git/git.git

$ cd git
$ git switch -c subtree-fix

Script we need to modify is contrib/subtree/git-subtree.sh. Make parallel change similar to original patch + commit.

Write a detailed commit message, since that's expected 

$ git log -1
  commit 9117eabc9c55187fd6c0204063734b72979f2082 (HEAD -> subtree-fix)
  Author: Roland Conybeare <rconybeare@gmail.com>
  Date:   Sun May 24 16:29:58 2026 -0400

      subtree: fix cache_set failure on commit reachable by multiple paths

      When splitting a subtree, committs that do not intersect prefix
      receive identity mapping (oldrev -> oldrev). If such commit
      is reachable by multiple paths in the revision DAG, the cache_set()
      function may be called twice for the same (oldrev -> newrev) pair.

      This triggers fatal error "cache for <hash> already exists"

      Bugfix is to make cache_set() idempotent when the same
      (oldrev -> newrev) pair appears multiple times.

      Signed-off-by: Roland Conybeare <rconybeare@gmail.com>

Follow git's contribution process. Setup in ~/.gitconfig 

[sendemail]
    smtpserver = smtp.gmail.com
    smtpserverport = 587
    smtpencryption = tls
    smtpuser = rconybeare@gmail.com

Get git to generate patch for us. 

$ cd ~/proj/git
$ mkdir outbound
$ git format-patch --cover-letter -o outbound/ --base=auto @{u}
..edit the cover-letter patch...

outbound/0000-cover-letter.patch: 

From 9117eabc9c55187fd6c0204063734b72979f2082 Mon Sep 17 00:00:00 2001
From: Roland Conybeare <rconybeare@gmail.com>
Date: Sun, 24 May 2026 16:50:26 -0400
Subject: [PATCH 0/1] bugfix git subtree split

I have a project that combines multiple independent repos
into an unmbrella repo, relying on git subtree.
Encountered a unrecoverable fatal error
from 'git subtree split' with error

    fatal: cache for <hash> already exists!

Problem arises because history to be split contains merge commits
that cause DAG traversal to consider the same umbrella commit on
multiple paths. The fatal triggers when 'git subtree split' tries
to cache the same commit twice; enclosed patch prunes these duplicate
paths.

Roland Conybeare (1):
  subtree: fix cache_set failure on commit reachable by multiple paths

 contrib/subtree/git-subtree.sh | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)


base-commit: 6a4418c36d6bad69a599044b3cf49dcbd049cb45
--
2.50.1

Finally, send it: 

$ git send-email --to=git@vger.kernel.org outbound/*.patch
...[y] to prompts...

Author: Roland Conybeare

Created: 2026-05-24 Sun 18:48

Validate