#!/bin/bash 
set -eou pipefail

# Example:
#   ./workspace-pack-hg-8.sh https://hg.openjdk.java.net/jdk8u/jdk8u jdk8u-jdk8u

# Enable aggressive repository format for Mercurial. Cloning the repository into
# this format would take a while. This is mostly convenient for the CI jobs that
# repackage the workspaces, but not for user clones
HG_OPTS=" --config=format.generaldelta=1 --config=format.aggressivemergedeltas=1"

URL=$1
NAME=$2
UPSTREAM_NAME=$NAME-upstream

# Clone the repository, if it does not exist yet. The particular trouble with
# forest repository is that we need to update the root repository to get the
# forest management script, which can then be used to clone the subrepos.
if [ ! -d $UPSTREAM_NAME ]; then
  hg clone $URL $UPSTREAM_NAME
  cd $UPSTREAM_NAME
  hg update
  HGFOREST_GLOBALOPTS="${HG_OPTS}" sh common/bin/hgforest.sh clone
  cd ..
fi

# Perform the very first pull of the repository. This would already merge some
# metadata using general+aggressive deltas. Make sure repository does not have
# working copy checked up, only the metadata itself. Again, we need to get root
# repository in order before dealing with forests.
cd $UPSTREAM_NAME
hg pull
hg update
HGFOREST_GLOBALOPTS="${HG_OPTS}" sh common/bin/hgforest.sh pull
sh common/bin/hgforest.sh update null
hg update null
cd ..

# Perform the *second* clone of the repository. This apparently repackages the
# manifest and metadata to a much more compact form. Make sure working copy is
# not materialized, an that we use the non-local clone that actually rewires
# metadata. This clone would record the "local" repo as parent, overwrite it
# with the upstream repo URL.
rm -rf $NAME
hg $HG_OPTS clone --pull --uncompressed -U $UPSTREAM_NAME $NAME
echo -e "[paths]\ndefault = $URL" > $NAME/.hg/hgrc
for R in corba jaxp jaxws jdk hotspot langtools nashorn; do
  if [ -d $UPSTREAM_NAME/$R ]; then
    hg $HG_OPTS clone --pull --uncompressed -U $UPSTREAM_NAME/$R $NAME/$R &
  fi
done
wait
for R in corba jaxp jaxws jdk hotspot langtools nashorn; do
  if [ -d $NAME/$R ]; then
    echo -e "[paths]\ndefault = $URL/$R" > $NAME/$R/.hg/hgrc
  fi
done

# Package the repository. In the presence of bulk file moves, the stream gets
# much more compressible if we stack similar files together. The default order
# comes from the filesystem walk, and might not be ideal for this. The good proxy
# for similarity is the file name: it captures lots of moves of the same file,
# and clusters similar files together. Create the sorted list of paths,
# and use that order to create the tarball.
find $NAME/ -type f | \
  awk -F '/' '{ k = $(NF); sub(/ /, "", k); print k " " $L; }' | \
  sort | \
  cut -d" " -f 2- > list.txt
tar -c -f $NAME.tar -T list.txt

# Finally, compress the whole thing with the max compression available. This 
# probably clashes with Mercurial's own storage compression, and it would be
# interesting to somehow request the metadata compressor to unpack the repository
# internal metadata for this compressor to work on raw stream. That is what 
# "hg bundle" apparently does? We are working with the repository copy itself,
# so we have to be content with whatever compression we can get.
rm -f $NAME.tar.xz
xz -9 $NAME.tar

# Verification: unpack the resulting tarball and check repository is still fine
rm -rf temp
mkdir temp
cd temp
tar xJf ../$NAME.tar.xz
cd $NAME
hg update
hg verify
sh common/bin/hgforest.sh update
sh common/bin/hgforest.sh verify
cd ../../
rm -rf temp

# Sample jdk8u/jdk8u tarball sizes:
#   - 326M uncompressed, 209M compressed originally
#   - 322M uncompressed, 208M compressed after file reordering