Import patchsync version 2

[utils/utils.git] / patchsync

#!/bin/bash
# patchsync: Synchronizes a trunk, a branch, and a patch containing the
# differences between them.
# Version 2
# -- Matt McCutchen
#
# usage: patchsync [--dry-run] <staging> [branch | patch]
#
# Patchsync is invoked on a "staging directory", which holds some configuration
# (including the locations of the trunk, patch, and branch it is to synchronize)
# and some synchronization state.  It determines whether each of the trunk,
# patch, and branch has changed since the last successful synchronization and
# updates the patch or branch as appropriate:
# 
# Changed since last sync   Patchsync's behavior
# -------------------------------------------------
# Nothing                   Do nothing
# Trunk only                Update branch
# Patch but not branch      Update branch
# Branch but not patch      Update patch
# Branch and patch          Complain about conflict
#
# <staging>: path to the staging directory
#
# --dry-run: show what would happen without actually modifying the trunk, patch,
#   branch, or synchronization state
#
# {branch | patch}: force patchsync to update the specified thing from the
#   others instead of deciding automatically; you can use this argument to
#   revert or to resolve a conflict
#
# CAVEAT: Patchsync might make a mess if the trunk, patch, or branch is
# modified in a way not hidden by the filters while patchsync is running!
#
# CAVEAT: Patchsync only notices creations, deletions, and modifications of
# regular files in the trunk and branch, not other changes like empty directory
# creations.  If you make a change like that to the trunk, you can force
# patchsync to update the branch.
#
# Staging directory format: A staging directory contains the following items:
#   "trunk", trunk directory or symlink to it
#   "patch", patch regular file or symlink to it
#   "branch", branch directory or symlink to it
#     [Why symlinks?  Expose as much as possible to tools like symlinks(8).]
#   "settings", shell script defining the following shell functions:
#     - do_diff <trunk> <branch> <write-patch>: diff the specified trunk and
#         branch and write the patch to the specified file; define it to use
#         your favorite diff format
#       - example: exitoneok diff -urN $1 $2 \
#                    | sed -re 's/^(\+\+\+|---) ([^\t]+).*$/\1 \2/' \
#                    | exitoneok grep -v '^diff' >$3
#     - do_patch <patch> <convert-trunk-to-branch>: apply the patch to the
#         specified trunk; define it to understand your favorite diff format
#       - example: patch --no-backup-if-mismatch -d $2/ -p1 <$1
#     - Note: patchsync runs these functions under "pipefail", but the
#         "set -e" it uses does not propagate into the functions.  Patchsync
#         provides an "exitoneok" function you can use to treat an exit code of
#         1 as 0.  You might want to && successive commands together.
#     - There are several possible ways to handle failed hunks.  The simplest
#         and safest is to make do_patch fail, but that's inconvenient for the
#         user, who must investigate the *.rej files in the staging directory
#         and either fix the patch or fix the branch and force updating the
#         patch.  One could make do_patch succeed, but if the user then modifies
#         the branch, the failed hunks will merely be dropped from the patch,
#         which is probably unacceptable.  The clever way is to let do_patch
#         succeed but make do_diff fail if any *.rej files exist in the branch.
#   "filters" (optional): rsync filters to use when accessing the trunk and
#     branch; hide filters apply to reading, protect filters to writing;
#     hint: you probably want to hide and protect build outputs
#
# Other usage: patchsync --new <trunk> <patch> <branch> <staging>
# Mostly sets up a new staging directory for the given trunk, branch, and patch
# at the given location.  You still have to provide settings, and filters if
# you want them.
# - If one of the patch or branch exists, the other will be calculated when
#   you first synchronize.
# - If both exist, you will get a conflict when you first synchronize and you
#   will need to specify which to update.
# - If neither exists, you get an empty patch and a branch identical to the trunk.

# Disable branch/.patchsync support because it's a bad idea in general, and the
# cyclic symlink confuses Eclipse in particular. -- Matt 2006.11.30

set -e
trap "echo 'Patchsync encountered an unexpected error!  ABORTING!' 1>&2; exit 2;" ERR
set -o errtrace
set -o pipefail

# Make sure we have rsync.
type rsync >/dev/null 2>&1 || \
	{ echo "Patchsync requires rsync, but there's no rsync on your path!" 1>&2; exit 1; }
# If a cp2 is available, use it; otherwise define our own.
type cp2 >/dev/null 2>&1 || function cp2 { rsync -rltE --chmod=ugo=rwx "$@"; }

function exitoneok {
	"$@" || [ $? == 1 ]
}

# wdpp_from <B> ==> the shortest relative prefix-path from directory B to the current directory
# (prefix-path means it ends in a slash unless it's `' which means '.')
# "patchsync" uses this to link-dest when copying the branch out.
# "patchsync --new" uses it to reverse the staging dir path when creating symlinks.
function wdpp_from {
	AtoB="$1"
	# Start with symlink-followed absolute prefix-paths without the initial slash.
	# NOT bash builtin pwd; it tells us how we got here, not where we are
	pA="$(/bin/pwd)/"
	pA="${pA#/}"
	pB="$( (cd "$AtoB" && /bin/pwd) )/"
	pB="${pB#/}"
	# Lop off the longest common prefix of components that we can.
	# While first components are equal...
	# (Empty correctly doesn't equal remaining)
	while { [ -n "$pA" ] || [ -n "$pB" ]; } && [ "${pA%%/*}" == "${pB%%/*}" ]; do
		# Remove them.
		pA="${pA#*/}"
		pB="${pB#*/}"
	done
	ans="$pA"
	# Translate remaining components of $pB to ../s
	while [ -n "$pB" ]; do
		ans="$ans../"
		pB="${pB#*/}"
	done
	# Double check; add dot to the end to enforce ending in a slash and handle empty ans
	(cd "$AtoB" && [ "$ans." -ef /proc/self/fd/3 ]) 3<.
	# Yay
	echo "$ans"
}

function hash_file {
	# Lop off the filename and binary indicator
	sha1sum -b "$1" | sed -re 's/^([^ ]*).*$/\1/'
}

function patchsync_sync {

if [ "$1" == --dry-run ]; then
	echo "Dry run mode."
	dryrun=1
	shift
fi

staging="$1"
if [ -r "$staging/settings" ]; then
	echo "Using staging dir $staging"
else
	echo "Specify a staging directory containing a settings file!" 1>&2
	exit 1
fi
cd "$staging" || { echo "Failed to enter staging dir!" 1>&2; exit 1; }
shift

. settings
type do_diff >/dev/null 2>&1 || { echo "do_diff is not defined!" 1>&2; exit 1; }
type do_patch >/dev/null 2>&1 || { echo "do_patch is not defined!" 1>&2; exit 1; }

whichtoupdate="$1"
# patchsync --new doesn't need this any more except for identical-branch
#if [ -z "$whichtoupdate" ] && [ -s whichtoupdate ]; then
#	# Hook for patchsync --new
#	whichtoupdate="$(< whichtoupdate)"
#	echo "Updating $whichtoupdate according to staging dir."
#el
if [ -n "$whichtoupdate" ]; then
	echo "Updating $whichtoupdate according to command line argument."
else
	echo "Synchronizing."
fi

filteropts=()
! [ -e filters ] || filteropts=("${filteropts[@]}" --filter='. filters')
# 'R *' or 'S *' disables filtering on the staging dir side.

COPYIN=(cp2 --del --filter='R *' "${filteropts[@]}")
COPYOUT=(cp2 --del --filter='S *' "${filteropts[@]}" --no-t --checksum) # be nice to mtimes

# hash_dir foo/ ==> a hash code covering all of the shown files in foo/
function hash_dir {
	# Itemize the dir, extract filenames, hash the files, and hash the list of
	# hashes.
	"${COPYIN[@]}" -i -n $1 nonexistent/ \
		| sed -n -e '/^>f/{ s/^[^ ]* //; p }' \
		| (cd $1 && xargs --no-run-if-empty --delimiter='\n' sha1sum -b) \
		| hash_file /dev/stdin
}

echo "Checking for changes..."
hash_dir trunk/ >trunk-new-hash
cmp trunk-{save,new}-hash &>/dev/null || { trunkch=1; echo "Trunk has changed"; }
hash_file patch >patch-new-hash
cmp patch-{save,new}-hash &>/dev/null || { patchch=1; echo "Patch has changed"; }
hash_dir branch/ >branch-new-hash
cmp branch-{save,new}-hash &>/dev/null || { branchch=1; echo "Branch has changed"; }

# If we're in synchronization mode, decide what to update.
if [ -z "$whichtoupdate" ] && [[ -n $trunkch || -n $branchch || -n $patchch ]]; then
	if [ -e identical-branch-flag ] && ! [ $patchch ] && ! [ $branchch ]; then
		# We still want to create an identical branch.
		whichtoupdate=identical-branch
	elif ! [ $branchch ]; then
		# Trunk, patch, or both changed.  Update branch.
		whichtoupdate=branch
	elif ! [ $patchch ]; then
		# Branch changed, and trunk may have also changed.  Update patch.
		whichtoupdate=patch
	else
		# Branch and patch both changed.  A message appears later.
		whichtoupdate=conflict
	fi
	#echo "Synchronization will update $whichtoupdate."
fi

if [ -n "$whichtoupdate" ]; then

# Always show what would happen if patch-new and branch-new were copied out.
# (If there was a problem creating one of them, patchsync would have just
# deleted it.)  But only actually copy them out and update synchronization
# state if no error.
error=

# Don't let stuff from an old run confuse us.
rm -rf patch-new branch-new

function prepare_branch {
	echo "Preparing updated branch..."
	# No link-dest because we will modify and then link-dest when copying out
	"${COPYIN[@]}" trunk/ branch-new/
	do_patch patch branch-new || \
		{ error=1; echo "Failed to prepare updated branch!" 1>&2; rm -rf branch-new; }
}

function prepare_patch {
	echo "Preparing updated patch..."
	# Link-dest is fine because these are temporary read-only copies
	"${COPYIN[@]}" --link-dest=../trunk/ trunk/ trunk-tmp/
	"${COPYIN[@]}" --link-dest=../branch/ branch/ branch-tmp/
	do_diff trunk-tmp branch-tmp patch-new || \
		{ error=1; echo "Failed to prepare updated patch!" 1>&2; rm -rf patch-new; }
	rm -rf trunk-tmp branch-tmp
}

case $whichtoupdate in
(identical-branch)
	echo "Creating identical branch..."
	# No link-dest because we will link-dest when copying out
	"${COPYIN[@]}" trunk/ branch-new/
	echo "Creating empty patch..."
	do_diff branch-new branch-new patch-new || \
		{ error=1; echo "Failed to create empty patch!" 1>&2; rm -rf patch-new; }
	;;
(branch)
	prepare_branch
	;;
(patch)
	prepare_patch
	;;
(conflict)
	error=1
	cat <<EOF 1>&2
CONFLICT: both branch and patch changed!
Run patchsync <staging> {branch | patch} to
update the specified thing from the others.
I'll leave updated copies of both branch
and patch in the staging directory to help
you decide which way you want to update.
EOF
	prepare_branch
	prepare_patch
	;;
(*)
	echo "Internal error, whichtoupdate should not be $whichtoupdate!" 1>&2
	exit 1
	;;
esac

if ! [ $error ] && ! [ $dryrun ]; then
	# Disable locking for now...
	# ! [ -e lock ] || { echo "Staging dir is locked!  Delete the file \`lock' if the other instance of patchsync is gone." 1>&2; exit 1; }
	# echo "patchsync lock file pid $$ date $(date)" >lock
	
	echo "Copying out..."
	! [ -e branch-new ] || {
		hash_dir branch-new/ >branch-new-hash
		"${COPYOUT[@]}" -i --link-dest="$(wdpp_from branch/)branch-new/" branch-new/ branch/
		rm -rf branch-new
	}
	! [ -e patch-new ] || cmp -s patch-work patch || {
		hash_file patch-new >patch-new-hash
		# Don't use rsync because we might have to write through a symlink.
		echo "> patch"
		cp --preserve=timestamps patch-new patch
		rm -f patch-new
	}
	
	echo "Remembering synchronized state for next time..."
	for i in trunk patch branch; do
		mv $i-new-hash $i-save-hash
	done
	
	# rm lock
else
	echo "Would copy out as follows:"
	! [ -e branch-new ] || {
		hash_dir branch-new/ >branch-new-hash
		"${COPYOUT[@]}" -n -i --link-dest="$(wdpp_from branch/)branch-new/" branch-new/ branch/
		#rm -rf branch-new
	}
	! [ -e patch-new ] || cmp -s patch-work patch || {
		hash_file patch-new >patch-new-hash
		# Don't use rsync because we might have to write through a symlink.
		echo "> patch"
		#cp --preserve=timestamps patch-new patch
		#rm -f patch-new
	}
	echo "Would remember synchronized state for next time."
	echo "I'm leaving \"new\" files in the staging dir so you can inspect them."
fi

else # whichtoupdate
	# Easy case
	echo "Nothing changed."
	rm -f {trunk,patch,branch}-new-hash
fi

if [ $error ]; then
	echo "Synchronization failed." 1>&2
	exit 1
else
	echo "Synchronization finished."
	if [ -e identical-branch-flag ]; then
		if ! [ $dryrun ]; then
			rm identical-branch-flag
			echo "Removed identical-branch-flag."
		else
			echo "Would remove identical-branch-flag."
		fi
	fi
	# Yay!  Done patchsync_sync!
fi
}

function patchsync_new {
	if [ $# != 4 ]; then
		echo "Expected 4 arguments after --new, got $#." 1>&2
		echo "usage: patchsync --new <trunk> <patch> <branch> <staging>" 1>&2
		exit 1
	fi
	
	# Set up arguments.  Open templates because we will change directories.
	trunk="$1"
	patch="$2"
	branch="$3"
	staging="$4"
	
	# What exists?  Whichtochange first?
	! [ -e "$staging" ] || { echo "Staging dir already exists!" 1>&2; exit 1; }
	[ -d "$trunk" ] || { echo "Trunk does not exist!" 1>&2; exit 1; }
	
	# Create staging dir.
	mkdir "$staging"
	wdpp="$(wdpp_from "$staging")"
	cd "$staging"
	echo "Created staging dir at $staging."
	
	# Adjust paths appropriately.
	trunk="$wdpp$trunk"
	patch="$wdpp$patch"
	branch="$wdpp$branch"
	
	# Create links to areas
	ln -s "$trunk" trunk
	ln -s "$patch" patch
	ln -s "$branch" branch
	echo "Created links to areas."
	
	# This approach is better than setting whichtochange because we'll notice
	# if the user puts something into one of the areas we created before first
	# sync.
	function create_patch {
		touch "$patch"
		hash_file patch >patch-save-hash
		echo "Created empty patch."
	}
	function create_branch {
		mkdir "$branch"
		# Can't do hash_dir because ${COPYIN[@]} hasn't been set <== no filters
		hash_file /dev/null >branch-save-hash
		echo "Created empty branch."
	}
	
	if [ -e "$patch" ] && ! [ -e "$branch" ]; then
		create_branch
		echo "Patch exists; branch will be calculated when you first synchronize."
	elif [ -e "$branch" ] && ! [ -e "$patch" ]; then
		create_patch
		echo "Branch exists; patch will be calculated when you first synchronize."
	elif ! [ -e "$patch" ] && ! [ -e "$branch" ]; then
		create_patch
		create_branch
		echo "Neither branch nor patch exists;"
		echo "a branch identical to the trunk will be created when you first synchronize."
		echo flag >identical-branch-flag
		echo "Created identical-branch-flag to tell first run of patchsync about this."
	else
		echo "Both patch and branch exist."
		echo "You will need to specify whether to overwrite the"
		echo "patch or the branch when you first synchronize!"
	fi
	
	# Write settings file.
	cat >settings <<END
# Define do_diff and do_patch here!
END
	echo "Wrote settings file placeholder."
	
	echo ""
	echo "Patchsync initialized."
	echo "Now add your definitions of do_diff and do_patch to the settings file,"
	echo "add a filter file if you wish, and perform the first sync."
}

function patchsync_help {
	cat <<EOF
Patchsync version 2 by Matt McCutchen
usage: patchsync [--dry-run] <staging> [branch | patch]
       patchsync --new <trunk> <patch> <branch> <staging>
Please read the top of the script for complete documentation.
EOF
}

case "$1" in
(--help|--version)
	patchsync_help ;;
(--dry-run)
	patchsync_sync "$@" ;;
(--new)
	shift
	patchsync_new "$@" ;;
(''|--*)
	patchsync_help 1>&2
	exit 1 ;;
(*)
	patchsync_sync "$@" ;;
esac