add --full mode; via this, the two phases can work together. faster namely.

1 files changed, 42 insertions, 16 deletions

diff --git a/create-git.sh b/create-git.sh
index 667fed0..4fa47e5 100755
--- a/create-git.sh
+++ b/create-git.sh
@@ -8,32 +8,58 @@ rm -rf git/* git/.git
 set -f
 mkdir -p git
 cd git
+git_root="$(pwd)"
 git init --bare
 git config core.logAllRefUpdates false
 git config prune.expire now
 mkdir -p objects/info
-targets=( $(find ../final/ -maxdepth 1 -mindepth 1 -printf '../final/%P/\n' | \
-  xargs -n1 readlink -f | \
-    while read l; do
-      [ -e "$l/cvs2svn-tmp/git-dump.dat" ] || continue;
-      echo "$l/git/objects" >> objects/info/alternates
-      echo "$l"
-    done
-  )
-)
 
-echo "loading all commits, linearizing, and rewriting history..."
-time (
-  "${root}/rewrite-commit-dump.py" "${targets[@]}" | \
-    tee ../export-stream-rewritten | \
-    git fast-import
-) 2>&1 | tee git-creation.log
+update_alternates() {
+  local alternates="$(readlink -f objects/info)/alternates"
+  cd "${root}"
+  while read l; do
+    l=$(readlink -f "$l")
+    [ -e "$l/cvs2svn-tmp/git-dump.dat" ] || { echo "ignoring nonexistant alternates source $l" >&2; continue; }
+    echo "$l/git/objects" >> "${alternates}"
+    echo "$l"
+  done
+  echo "starting history linearizing/rewriting" >&2
+}
+
+standalone_mode() {
+  echo "loading all commits" >&2
+  find ../final/ -maxdepth 1 -mindepth 1 -printf '../final/%P/\n' | \
+    xargs -n1 readlink -f | update_alternates
+}
+
+if [ "$1" == --fast ]; then
+  command=update_alternates
+else
+  command=standalone_mode
+  echo "loading all commits in parallel to their generation..." >&2
+fi
+
+# Roughly; since alternates are updated as we go- and since rewrite-commit-dump
+# doesn't actually output anything till it's linearized the history, we have
+# to delay fast-import's startup until we know we have data (meaning linearize
+# has finished- thus the alternates are all in place).
+# Bit tricky, but the gains have been worth it.
+time {
+  ${command} | \
+  "${root}/rewrite-commit-dump.py" | \
+  ( read line; { echo "$line"; cat; } | \
+      tee ../export-stream-rewritten |\
+      git fast-import
+  )
+} 2>&1 > >(tee git-creation.log)
+ret=$?
+[ $ret -eq 0 ] || { echo "none zero exit... the hell? $ret"; exit 1; }
 
 echo "recomposed; repacking and breaking alternate linkage..."
 # Localize the content we actual use out of the alternates...
 time git repack -Adf --window=100 --depth=100
 # Wipe the alternates.
-rm objects/info/alternates
+rm objects/info/alternates || { echo "no alternates means no sources..."; exit 2; }
 echo "doing cleanup..."
 time git prune
 echo "doing basic sanity check"