MahiroOS-jhalfs/extras/farce

#!/bin/bash

# $Id$
# Acknowledgment:
#  The following code is a no-modified version (except for this comment,
#  the `Id' comment, and the inline_doc fragment) of an original work written by
#  Ken Moffat and is included here with his permission.
#

# FARCE: Farce Assists Rebuild Comparison Evaluation ;)
#
# to answer the question "can it rebuild itself?"
#
#  We expect four arguments - first directory path, filelist
# containing the files in this directory which we wish to compare,
# second directory path, filelist for second directory.
#
# Yes, we could just compare everything in each tree, but the
# filelist script knows about files it can reasonably ignore,
# and this also allows us to build a sytem, boot it and get a
# list of files, build a full desktop environment, and only then
# build and boot the "can it build itself" test system and get
# _its_ filelist.
#
# What this script aims to do:
# ____________________________
#
# First, report files not in both builds.
#
# Then, confirm symlinks point to same targets.
#
# After that, compare individual files -
#  if different, run the file name through 'expected'
# to pick out files that are unlikely to match (logs,
# pids, fstab [assumes '/' is a different device each time],
# count these as 'expected'.
#
# For whatever is left, check the file type - ar archives
# have their members extraced and compared (every member has
# a timestamp), gzipped files are compared beyond their
# timestamp, binaries, at least those using shared libs or
# which are shared objects, are copied and subjected to
# --strip-debug.  If files match at this stage, count them as
# 'accepted'.
#
# As a last step for any file that doesn't match, copy it
# through some perl regexps to "process" it (convert any
# date, time, kernel-version information from standard formats
# into tokens, then see if the tokensi match.
#
# For details of the regexps, see the tokenize function.
# Those files that match after this are also counted as
# 'accepted'.  Note that I don't always start from the kernel
# version that I'm going to build, so this copes with e.g. perl
# files that hardcode the kernel version.
#
# We now have files that don't match.  A few of these seem to be
# common to all builds - some (members of) c++ libraries or ar
# archives, a few programs which perhaps use some sort of c++ code).
# The file name # is passed to the 'failure' function - these
# recognized filenames are labelled as 'predictable FAIL:',
# anything else is labelled as 'unexpected FAIL:'.
#
# output:
# stderr - files only in one of the builds, failure messages,
#  and totals.
#
# farce-results - more details, including which files were treated
# as expected differences, files where neither copy could be read,
# files treated as accepted, with the reason (and member for ar
# archives).  This data is typically up to 100 characters wide -
# sometimes it's a bit more, but it doesn't wrap too badly in a
# 100 character xterm.
#
# farce-extras - diffs for the files, or members, that didn't
# match.  This file is to establish new regexps for picking up
# date/time/kernel-version formats.
#
# farce-identical - the names of the files which are identical
#
# farce-substitutions - whenever using tokenizeanddiff results in a
# difference being accepted, for both versions diff the before and
# after versions to show what got changed.  If the file is a binary,
# the output may still be hard to read.  Note that I _know_ glibc
# version strings pass one of the regexps looking for a kernel version
# - since I expect you to use the same version of glibc for each
# build, this is not a problem.
#
# farce-differ - the names of the files which could not be treated
# as matching (whether or not I regard the failure as predictable)
# for possible input to ICA processing.
#
# Copyright (C) 2005, 2006 Ken Moffat <ken@linuxfromscratch.org>
#
# All rights reserved.
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 2 of the License, or (at
# your option) any later version.
#
# This program is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
# NON INFRINGEMENT.  See the GNU General Public License for more
# details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software
# Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
#

: <<inline_doc
    desc:       do farce analisys and report
    usage:      farce -directory $FARCELOGDIR/dir path1 list1 path2 list2
    input vars: $1 farce log dir for this comparation
                $2 full path to previous iteration
                $3 full path to filelist for previos iteration
                $4 full path to current iteration
                $5 full path to filelist for current iteration
    externals:  --
    modifies:   --
    returns:    --
    on error:
    on success:
inline_doc


VERSION="002"

# variables for output files
RESULT=farce-results
EXTRAS=farce-extras
IDENTICAL=farce-identical
SUBS=farce-substitutions
DIFFER=farce-differ

# documenting the variables
# C1, C2	temp files to hold disassembled code
# D1, D2	temp directories for extracting member of ar archive
# DIFF		temp file for diffing the filelists
# F1, F2	temp files for tokenizeanddiff
# FTYPE		obsolete, commented out
# M1, M2	temp files to hold members of an ar archive
# MEMBERS	temp file to list members of ar archive
# OP1, OP2	the original $PWD, needed when extracting members of ar
#		archives
# P1, P2	paths to first and second build
# S1, S2	temp files for shared objects

# functions
function dohelp() {
	echo "`basename $0`: compare trees of files from a build"
	echo "and lists of the files they contained"
	echo ""
	echo "`basename $0` [ -help | -version ] || path1 list1 path2 list2"
}

function emessage() {
	# write a string to both stderr and $RESULT
	echo "$@" >&2
	echo "$@" >&5
}

function expected() {
	# if we expect it to differ because of its name,
	# allow it and report, return true ; else return false
	case $1 in
	/boot/grub/menu.lst)
		# just in case somebody puts this into the main filesystem
		true;;
	/etc/aliases.db)
		# some sort of database for postfix, not parsable
		true;;
	/etc/blkid.tab)
		# includes dev name for rootfs
		true;;
	/etc/fstab)
		# fstab, e.g. ' / ' will differ
		true;;
	/etc/group*)
		true;;
	/etc/hosts)
		# with dhcp client, I add current ip address to this in a hook
		true;;
	/etc/ld.so.*)
		# .conf and .cache can vary,
		# particularly if one system has a full build when I run this
		true;;
	/etc/lilo.conf|/etc/yaboot.conf)
		# bootloader control, I assume grub will all be on a separate
		true;;
	/etc/mtab)
		# at a minimum, different '/'
		true;;
	/etc/ntp.drift)
		true;;
	/etc/passwd*)
		true;;
	/etc/shadow*)
		true;;
	/etc/ssh/*key|/etc/ssh/*pub)
		# openssh keys
		true;;
	/misc/*)
		# where I put buildscripts (which mostly won't change)
		# and stamps containing name/time/space which will differ in the times
		true;;
	/root/*)
		# expect .bash_history etc to differ - if we can read them
		true;;
	/usr/bin/lynx)
		# part of my inital builds, I guess this uses anonymous namespaces
		true;;
	/usr/include/c++/*/*/bits/stdc++.h.gch/*)
		# precompiled headers
		true;;
	/usr/lib*/libstdc++.a|/usr/lib*/libstdc++.so*|/usr/lib*/libsupc++.a)
		# probably, anonymous namespaces
		# libstdc++.a, libstdc++.so.n.n.n, libsupc++.a
		true;;
	/usr/share/info/dir)
		# if one system has had extra stuff built, this will likely be bigger
		true;;
	/usr/share/man/whatis)
		# if one system has had extra stuff built, this will likely be bigger
		true;;
	/var/lib/locate/locatedb)
		# if one system has had extra stuff built, this will likely be bigger
		true;;
	/var/lib/nfs/*)
		# allow nfs bookkeeping
		true;;
	/var/log/*)
		true;;
	/var/run/utmp)
		true;;
	/var/spool/fcron*)
		true;;
	/var/state/*)
		# allow dhcp leases
		true;;
	/var/tmp/random-seed)
		true;;
	# following start with wildcards
	*Image*|*.PPCBoot*|*vmlinuz*|*lfskernel*)
		# compressed kernels, sometimes just building at a different
		# date/time is enough to change the length of them, because the
		# long format date and time is part of the compressed data
		true;;
	*pid*)
		# pids, including e.g. /var/spool/postfix/pid/*
		true;;
	*)
		# nothing else is expected to be different
		false;;
	esac
	if [ $? -eq 0 ]; then
		message "expected difference in $1"
		let expected=$expected+1
		case $TYPE in
			AR)
				let EXPAR=$EXPAR+1
				;;
			ELF)
				let EXPELF=$EXPELF+1
				;;
			UNK)
				let EXPUNK=$EXPUNK+1
				;;
			# so far, no other valid types, so don't accumulate them
			*)
				emessage "internal error, expected difference for $1 of type $TYPE not allowed"
				exit 2
				;;
		esac
		true
	else
		false
	fi
}

function failure() {
	# first parm is filename or token
	# second parm is the error message
	# update the appropriate total
	# and write to both stderr and the results
	# by using emessage

	let different=$different+1
	case $TYPE in
		AR)
			let DIFAR=$DIFAR+1
			;;
		ELF)
			let DIFELF=$DIFELF+1
			;;
		GZ)
			let DIFGZ=$DIFGZ+1
			;;
		SYM)
			let DIFSYM=$DIFSYM+1
			;;
		UNK)
			let DIFUNK=$DIFUNK+1
			;;
		*)
			emessage "internal error in failure() for TYPE $TYPE"
			exit 2
			;;
		esac
		test -f ${P1}$1 && echo $1 >&9
		emessage "FAIL: $2"
}

function fatal() {
	# unrecoverable error
	echo $*
	exit 1
}

function filetype() {
	TYPE=`file ${P1}${FILE}`
	case $TYPE in
	*'current ar archive'*)
		let TOTAR=$TOTAR+1
		TYPE=AR
		;;
	*' ELF '*)
		let TOTELF=$TOTELF+1
		TYPE=ELF
		;;
	*'gzip compressed data'*)
		let TOTGZ=$TOTGZ+1
		TYPE=GZ
		;;
	*)
		let TOTUNK=$TOTUNK+1
		TYPE=UNK
		;;
	esac
}

function message() {
	# write a string to $RESULT
	echo $* >&5
}

function onlyone() {
	#report files only in one build
	# text should go to both stderr and the results,
	# but blank lines only go to the results
	if [ $1 == '<' ]; then
		emessage "File(s) only in the first build"
	else
		emessage "File(s) only in the second build"
	fi
	message ""
	FILES=`cat $DIFF | grep "^$1" | cut -d ' ' -f 2`
	for F in $FILES; do
		emessage $F
		let only=$only+1
	done
	message ""
}

# 'test' functions are called with three arguments:
# the two pathes and the filename
# - we know the file is of this type, so see if we
# can get it to match by reasonalbe means.
# if not, treat it as different.
#
# NB if pathes are absolute, we need to prefix them
# with the original $PWD to access the .a files
#
function testar() {
	# ar archives include timestamps for the members,
	# but diff doesn't show file timestamps unless the data differs
	# put out a message to help locate which archive any messages
	# about the members refer to.

	# try just stripping them U1,2 undebuggable
	U1=`mktemp` || fatal "cannot create a temporary file"
	U2=`mktemp` || fatal "cannot create a temporary file"
	cp ${1}${3} $U1
	cp ${2}${3} $U2
	strip --strip-debug $U1
	strip --strip-debug $U2
	cmp -s $U1 $U2
	rm $U1 $U2
	if [ $? -eq 0 ]; then
		let accepted=$accepted+1
		let ACCAR=$ACCAR+1
		message "archive $3 matches after strip --strip-debug"
		return
	fi
	# rest of this function retained primarily for pathologically bad builds
	# put out a message in the log to help identify which archive has issues.
	message "examining ar archive $3"
	D1=`mktemp -d` || fatal "cannot create a temporary directory"
	D2=`mktemp -d` || fatal "cannot create a temporary directory"
	cd $D1
	ar -x ${OP1}${1}${3}
	cd $D2
	ar -x ${OP2}${2}${3}
	cd
	# diff the members - true means they match
	diff -Na $D1 $D2 >/dev/null
	if [ $? -eq 0 ]; then
		message "accept: $3 after diffing the members"
		let accepted=$accepted+1
		let ACCAR=$ACCAR+1
	else
		# process individual members to eliminate date/time/kernel-version
		# first, check the members are the same
		M1=`mktemp` || fatal "cannot create a temporary file"
		M2=`mktemp` || fatal "cannot create a temporary file"
		cd $D1
		MEMBERS=
		for F in *; do
			MEMBERS="$MEMBERS $F"
		done
		cd
		echo $MEMBERS | sort >$M1
		cd $D2
		MEMBERS=
		for F in *; do
			MEMBERS="$MEMBERS $F"
		done
		cd
		echo $MEMBERS | sort >$M2
		cmp -s $M1 $M2
		if [ $? -ne 0 ]; then
			# oh dear, different members
			echo "list of members differs for archive $3" >&6
			diff $M1 $M2 >&6
			failure $3 "$3 list of members differs"
		else
			# members (names) are same,
			# process each one
			STATUS=0
			for M in $MEMBERS; do
				#avoid firing up perl on matching members
				cmp -s $D1/$M $D2/$M
				if [ $? -ne 0 ]; then
					tokenizeanddiff $D1/$M $D2/$M $FILE:$M
					if [ $? -eq 0 ]; then
						message "member $M matches after processing"
					else
						message "member $M DIFFERS after processing"
						STATUS=1
					fi
				fi
			done
			if [ $STATUS -eq 0 ]; then
				let accepted=$accepted+1
				let ACCAR=$ACCAR+1
			else
				let different=$different+1
				let DIFAR=$DIFAR+1
				echo $3 >&9
				emessage "FAIL: in $3"
			fi
		fi
		rm $M1 $M2
	fi
	rm -rf $D1 $D2
}

function testgzip() {
	# bytes 4,5,6,7 are the timestamp, so ignore these
	cmp -s -i 8 ${1}${3} ${2}${3}
	if [ $? -eq 0 ]; then
		message "accept: $3 after ignoring gzip timestamp"
		let accepted=$accepted+1
		let ACCGZ=$ACCGZ+1
	else
		failure $3 " $3 even after ignoring gzip timestamp"
	fi
}

function testso() {
	# shared object - first try stripping it
	# in fact, this now handles ALL ELF files
	S1=`mktemp` || fatal "cannot create a temporary file"
	S2=`mktemp` || fatal "cannot create a temporary file"
	cp ${1}${3} $S1
	strip --strip-debug $S1
	cp ${2}${3} $S2
	strip --strip-debug $S2
	cmp -s $S1 $S2
	if [ $? -eq 0 ]; then
		message "accept: $3 after --strip-debug"
		let accepted=$accepted+1
		let ACCELF=$ACCELF+1
	else
		tokenizeanddiff $S1 $S2 $3
		if [ $? -ne 0 ]; then
			failure $3 " $3 differs after stripping and processing"
		else
			message "accept: $3 after --strip-debug and processing"
			let accepted=$accepted+1
			let ACCELF=$ACCELF+1
		fi
	fi
	rm $S1 $S2
}

function tokenize() {
	# use regexes to replace date/time/kernel-version text
	# with tokens which may allow files to match even though
	# they have hardcoded date/time/kernel-version.
	# arguments are file to process, and where to put it.
	# these regexes are somewhat long, and the order they
	# are applied in is important (to stop short ones being
	# used when a longer version would match).
	# KV00 linux version date (e.g. as in the kernel itself)
	# allow 2 or 3 groups of three alphas here - optional smp,  with day, mon
	# KV01 kernel version, including possible cpu details (that is for cdda2wav)
	# KV02 just the version, in quotes e.g. "2.6.12.6" or '2.6.13', for perl stuff
	# except that "|' gives me grif, so try a boundary
	# also, it might need local version on the end, I really want
	# quote2.\d+.\d+.{0,32}quote - it is the quotes that don't work.
	# DT00 Day Mon .d+ hh:mm:ss TZN CCYY variations include non-caps and 'mon d'
	# DT01 Mon .d+ CCYY hh:mm:ss
	# DT02 hh:mm:ss Mon .d CCYY
	# DT03 Mon .d CCYY
	# DT04 Day Mon { ,d}d hh:mm:ss CCYY - for groff example postscript files
	# (somewhat similar to DT00, but '  d' or ' dd' for day of month and no TZN )
	# DT05 hh:mm:ss
	# DT06 ISO date using space as separator
	# DT07 ISO date using dash as separator
	# DT08 ISO date using slash as separator
	# DT09 fullmonth (capitalised), day number, comma, 4-digit year (groff 1.18.1 ps)
	# DT10 dd, fullmonth (capitalised), 4-digit year (groff 1.18.1 manpages)
	# DT11 '(xample comma space digit(s) backslash ) in groff memef.ps which is
	# quite clearly the day of the month when it was compiled, preceded by 'example'
	# with something weird between the e and the x.

	if [ $# -ne 2 ]; then
		fatal "tokenizing called with $# arguments : $*"
	fi

	cat $1 | perl -p \
	 -e 's/(L|l)inux.*\d\.\d\.\d+.* \#\d+( [A-Za-z][a-z]{2}){2,3} \d+ \d\d:\d\d:\d\d [A-Za-z]{3} \d{4}\b/%KV00%/g;' \
	 -e 's/(L|l)inux( (\w|_)+)?(-| |_)\d\.\d(\.\d+){1,2}((-|_)?(\w|_)+)?( |\000)*/%KV01%/g;' \
	 -e 's/\W2(\.\d+){2,3}(-|_)?((\w|_)+)?\s*\W/%KV02%/g;' \
	 -e 's/\b([A-Za-z][a-z]{2} ){2}( |\d)?\d \d\d:\d\d:\d\d [A-Za-z]{3} \d{4}\b/%DT00%/g;' \
	 -e 's/\b[A-Z][a-z]{2} ( |\d)\d \d{4} \d\d:\d\d:\d\d\b/%DT01%/g;' \
	 -e 's/\b\d\d:\d\d:\d\d [A-Z][a-z]{2} ( |\d)\d \d{4}\b/%DT02%/g;' \
	 -e 's/\b[A-Z][a-z]{2} ( |\d)\d \d{4}\b/%DT03%/g;' \
	 -e 's/\b([A-Z][a-z]{2} ){2}( |\d)\d \d\d:\d\d:\d\d \d{4}/%DT04%/g;' \
	 -e 's/\b\d\d:\d\d:\d\d\b/%DT05%/g;' \
	 -e 's/\b\d{4} \d\d \d\d\b/%DT06%/g;' \
	 -e 's/\b\d{4}-\d\d-\d\d\b/%DT07%/g;' \
	 -e 's/\b\d{4}\/\d\d\/\d\d\b/%DT08%/g;' \
	 -e 's/\b[A-Z][a-z]{2,} \d{1,2}, \d{4}/%DT09%/g;' \
	 -e 's/\b\d\d [A-Z][a-z]{2,} \d{4}/%DT10%/g;' \
	 -e 's/\(xample, \d{1,2}\\\)/%DT11%/g;' \
	>$2
}

function tokenizeanddiff() {
	# Call tokenize for the inputs, then compare the results
	# Input arguments are path/filename for old and new versions
	# third parm is readable name (filename, or archivename:member)
	# to help understand what is in the extras output.
	#  - sometimes called for files, but other times called for
	# members of ar archives extracted into temporary directories
	#message tokenizeanddiff called for $1 $2 $3
	F1=`mktemp` || fatal "cannot create a temporary file"
	F2=`mktemp` || fatal "cannot create a temporary file"
	tokenize $1 $F1
	tokenize $2 $F2

	# actually, cmp is probably more efficient
	# but for picking up the pieces it will be better to
	# use diff to see what got through.
	cmp -s $F1 $F2
	TOKENRESULT=$?
	if [ $TOKENRESULT -ne 0 ]; then
		echo "failure in $3..." >&6
		diff -a $F1 $F2 >&6
		rm $F1 $F2
		false
	else
		# show what we did
		echo "substitutions for $3" >&8
		echo "build one" >&8
		diff -a $1 $F1 >&8
		echo "build two" >&8
		diff -a $2 $F2 >&8
		rm $F1 $F2
		true
	fi
}

function validateargs() {
# validate the arguments
BAD=0
if ! [ -d $1 ]; then
	echo "Error: first argument is not a directory" >&2
	let BAD=$BAD+1
fi
NAME=`basename ${2%%-*}`
if [ $NAME != filelist ]; then
	echo "Error: second argument is not a recognized filelist" >&2
	let BAD=$BAD+1
fi
if ! [ -d $3 ]; then
	echo "Error: third argument is not a directory" >&2
	let BAD=$BAD+1
fi
NAME=`basename ${4%%-*}`
if [ $NAME != filelist ]; then
	echo "Error: fourth argument is not a recognized filelist" >&2
	let BAD=$BAD+1
fi
for I in $1 $2 $3 $4; do
	if ! [ -r $I ]; then
		echo "Error: cannot read $I" >&2
		let BAD=$BAD+1
	fi
done
if [ $1 == $3 ]; then
	echo "Error: directory pathes are identical" >&2
	let BAD=$BAD+1
fi
if [ $2 == $4 ]; then
	echo "Error: filelist names are identical" >&2
	let BAD=$BAD+1
fi
if [ $BAD -eq 0 ]; then
	ARGS=valid
fi
}

# Mainline
ARGS=unproven
OUTDIR=
if [ $# -eq 1 ]; then
	case $1 in
		-version|--version)
			echo "`basename $0` version $VERSION"
			exit 0
			;;
		-help|--help)
			dohelp
			exit 0
			;;
	esac
fi
if [ $1 = "--directory" ]; then
	OUTDIR=$2
	shift 2
	grep '/$' $OUTDIR >/dev/null 2>&1 || OUTDIR=`echo $OUTDIR | sed 's%$%/%'`
	echo "creating directory $OUTDIR"
	mkdir -p $OUTDIR
	if [ $? -ne 0 ]; then
		echo "cannot mkdir $OUTDIR"
		exit 1
	fi
fi
if [ $# -eq 4 ]; then
	validateargs $*
fi
if ! [ $ARGS == valid ]; then
	dohelp
	fatal "`basename $0`: error in arguments"
fi

# ok, we're happy, lets hit these files
exec 5>${OUTDIR}$RESULT
exec 6>${OUTDIR}$EXTRAS
exec 7>${OUTDIR}$IDENTICAL
exec 8>${OUTDIR}$SUBS
exec 9>${OUTDIR}$DIFFER

>${OUTDIR}$RESULT
if [ $? -ne 0 ]; then
	fatal "cannot write to ${OUTDIR}$RESULT"
fi

emessage "will compare:"
emessage " first build at $1 with files listed in $2"
emessage "second build at $3 with files listed in $4"

let accepted=0
let different=0
let expected=0
let matched=0
let only=0
let predictable=0
let unreadable=0
let total=0

# break down the accepted
let ACCAR=0
let ACCELF=0
let ACCGZ=0
let ACCUNK=0

# break down definitely different
let DIFAR=0
let DIFELF=0
let DIFGZ=0
let DIFSYM=0
let DIFUNK=0

# break down the expected differences
let EXPAR=0
let EXPELF=0
let EXPGZ=0
let EXPUNK=0

# break down the identical files
let MATAR=0
let MATELF=0
let MATGZ=0
let MATSYM=0
let MATUNK=0

# break down how many of each type
let TOTAR=0
let TOTELF=0
let TOTGZ=0
let TOTSYM=0
let TOTUNK=0

# now identify differences between the two trees
DIFF=`mktemp` || fatal "cannot create a temporary file"
diff $2 $4 >$DIFF

for RUN in '<' '>' ; do
	grep -q "$RUN" $DIFF && onlyone "$RUN"
done

rm $DIFF

# and compare them
message "Results of file comparison:"
message ""

# Strip any trailing slash from the path for tidyness,
# because the filenames all start with a slash.  Unfortunately,
# unfortunately, '/' becomes empty, which breaks subroutines,
# so special case it.
# also, to process ar archives we need to extract them in temp
# directories - that means that after cd'ing we've broken any
# relative path, so save original pwd as necessary.
P1=`echo $1 | sed 's%/$%%'`
echo $1 | grep '^/' >/dev/null
if [ $? -ne 0 ]; then
	# relative path
	OP1=${PWD}/
	#echo "setting OP1 to $OP1"
else
	OP1=
	#echo "$1 is an absolute path"
fi
test -z "$P1" && P1='/'
P2=`echo $3 | sed 's%/$%%'`
echo $3 | grep '^/' >/dev/null
if [ $? -ne 0 ]; then
	# relative path
	OP2=${PWD}/
	#echo "setting OP2 to $OP2"
else
	OP2=
	#echo "$3 is an absolute path"
fi
test -z "$P2" && P2='/'

echo "about to read $2"
while read FILE ; do
#echo "process $FILE"
#echo "test existence of ${P2}${FILE}"
	# confirm it exists in second build
	# we have already reported files only in one build
	if [ -f ${P2}"${FILE}" ]; then
		let total=$total+1
		# check we can read both of them
		# or count as unreadable - I used to separate only-one-unreadable,
		# but if you compre '/' and a _copy_  of /mnt/lfs that assumption
		# breaks, so be less picky.
		if ! [ -r "${P1}${FILE}" ] || ! [ -r "${P2}${FILE}" ]; then
			message "cannot read one or both versions of $FILE"
			let unreadable=$unreadable+1
			continue
		fi
		if [ -h "${P1}${FILE}" ]; then
			# for symlink, look at what it points to
			# exceptionally, do not call filetype
			TYPE=SYM
			let TOTSYM=$TOTSYM+1
			SL1=`ls -l "${P1}${FILE}" | awk '{ print $11 }'`
			SL2=`ls -l "${P2}${FILE}" | awk '{ print $11 }'`
			if [ "$SL1" = "$SL2" ]; then
				echo "symlink $FILE matches for $SL1" >&5
				let matched=$matched+1
				let MATSYM=$MATSYM+1
			else
				failure TARGET " symlink $FILE points to $SL1 and $SL2"
				echo $FILE >&9
			fi
		else
			# regular file, start by typing it for accounting,
			# then compare it
			filetype ${P1}${FILE}
			cmp -s "${P1}${FILE}" "${P2}${FILE}"
			if [ $? -eq 0 ]; then
				let matched=$matched+1
				case $TYPE in
					AR)
						let MATAR=$MATAR+1
						;;
					ELF)
						let MATELF=$MATELF+1
						;;
					GZ)
						let MATGZ=$MATGZ+1
						;;
					UNK)
						let MATUNK=$MATUNK+1
						;;
					*)
						echo "unexpected TYPE of $TYPE for $FILE" >&2
						exit 2
						;;
				esac
				echo ${FILE} >&7
			else
				# seems different, can we do better ?
				# test if we expect it to differ
				expected $FILE
				if [ $? -ne 0 ]; then
					case $TYPE in
					GZ)
						testgzip $P1 $P2 $FILE ;;
					AR)
						testar $P1 $P2 $FILE ;;
					ELF)
						testso $P1 $P2 $FILE ;;
					*)
						# long-stop - strip dates from text files
						tokenizeanddiff "${P1}${FILE}" "${P2}${FILE}" "$FILE"
						if [ $? -eq 0 ]; then
							message "accepted $FILE after processing"
							let accepted=$accepted+1
							let ACCUNK=$ACCUNK+1
						else
							failure "$FILE" " $FILE is different"
						fi
						;;
					esac
				fi
			fi
		fi
	fi
done < $2

message ""
# write totals to stderr as well as the results file
emessage "$only files in only one of the builds"
emessage "$total files compared, of which"
emessage "$unreadable files could not be read, skipped"
emessage "$matched files are identical"
emessage "$expected files differed as expected"
emessage "$accepted files had allowable differences"
#emessage "$predictable files differed as they normally do"
emessage "$different files differed"

# totals of different file types
emessage ""
emessage "$TOTAR ar archives"
emessage " of which $MATAR are identical"
emessage " of which $ACCAR are accepted after strip-debug or extracting, diffing, tokenizing"
emessage " of which $EXPAR differed as expected"
emessage "    of which $DIFAR differed"
emessage "$TOTELF ELF executables or shared libraries"
emessage " of which $MATELF are identical"
emessage " of which $ACCELF are accepted after stripping and tokenizing"
emessage " of which $EXPELF differed as expected"
emessage "    of which $DIFELF differed"
emessage "$TOTGZ gzipped files"
emessage " of which $MATGZ are identical"
emessage " of which $ACCGZ are accepted after comparing beyond timestamp"
emessage " of which $DIFGZ are different"
emessage "$TOTSYM symbolic links"
emessage " of which $MATSYM are identical"
emessage "    of which $DIFSYM have different targets"
emessage "$TOTUNK other files"
emessage " of which $MATUNK are identical"
emessage " of which $ACCUNK are accepted after tokenizing"
emessage " of which $EXPUNK differed as expected"
emessage "    of which $DIFUNK differed"