1
0
mirror of https://git.savannah.gnu.org/git/emacs.git synced 2024-11-24 07:20:37 +00:00
emacs/lib-src/rcs2log
Paul Eggert e852d58d04 (initialize_fullname): Add support for NIS+.
(hostname): Fully qualify the default hostname with the domainname
if the hostname lacks a `.'.
1996-01-15 01:17:56 +00:00

608 lines
15 KiB
Bash
Executable File

#! /bin/sh
# RCS to ChangeLog generator
# Generate a change log prefix from RCS files and the ChangeLog (if any).
# Output the new prefix to standard output.
# You can edit this prefix by hand, and then prepend it to ChangeLog.
# Ignore log entries that start with `#'.
# Clump together log entries that start with `{topic} ',
# where `topic' contains neither white space nor `}'.
# Author: Paul Eggert <eggert@twinsun.com>
# $Id: rcs2log,v 1.27 1996/01/15 01:17:03 eggert Exp $
# Copyright 1992, 1993, 1994, 1995, 1996 Free Software Foundation, Inc.
# 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, 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. 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; see the file COPYING. If not, write to
# the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
tab=' '
nl='
'
# Parse options.
# defaults
: ${AWK=awk}
: ${TMPDIR=/tmp}
hostname= # name of local host (if empty, will deduce it later)
indent=8 # indent of log line
length=79 # suggested max width of log line
logins= # login names for people we know fullnames and mailaddrs of
loginFullnameMailaddrs= # login<tab>fullname<tab>mailaddr triplets
recursive= # t if we want recursive rlog
rlog_options= # options to pass to rlog
tabwidth=8 # width of horizontal tab
while :
do
case $1 in
-i) indent=${2?}; shift;;
-h) hostname=${2?}; shift;;
-l) length=${2?}; shift;;
-[nu]) # -n is obsolescent; it is replaced by -u.
case $1 in
-n) case ${2?}${3?}${4?} in
*"$tab"* | *"$nl"*)
echo >&2 "$0: -n '$2' '$3' '$4': tabs, newlines not allowed"
exit 1
esac
loginFullnameMailaddrs=$loginFullnameMailaddrs$nl$2$tab$3$tab$4
shift; shift; shift;;
-u)
# If $2 is not tab-separated, use colon for separator.
case ${2?} in
*"$nl"*)
echo >&2 "$0: -u '$2': newlines not allowed"
exit 1;;
*"$tab"*)
t=$tab;;
*)
t=:
esac
case $2 in
*"$t"*"$t"*"$t"*)
echo >&2 "$0: -u '$2': too many fields"
exit 1;;
*"$t"*"$t"*)
;;
*)
echo >&2 "$0: -u '$2': not enough fields"
exit 1
esac
loginFullnameMailaddrs=$loginFullnameMailaddrs$nl$2
shift
esac
logins=$logins$nl$login
;;
-r) rlog_options=$rlog_options$nl${2?}; shift;;
-R) recursive=t;;
-t) tabwidth=${2?}; shift;;
-*) echo >&2 "$0: usage: $0 [options] [file ...]
Options:
[-h hostname] [-i indent] [-l length] [-R] [-r rlog_option]
[-t tabwidth] [-u 'login<TAB>fullname<TAB>mailaddr']..."
exit 1;;
*) break
esac
shift
done
month_data='
m[0]="Jan"; m[1]="Feb"; m[2]="Mar"
m[3]="Apr"; m[4]="May"; m[5]="Jun"
m[6]="Jul"; m[7]="Aug"; m[8]="Sep"
m[9]="Oct"; m[10]="Nov"; m[11]="Dec"
# days in non-leap year thus far, indexed by month (0-12)
mo[0]=0; mo[1]=31; mo[2]=59; mo[3]=90
mo[4]=120; mo[5]=151; mo[6]=181; mo[7]=212
mo[8]=243; mo[9]=273; mo[10]=304; mo[11]=334
mo[12]=365
'
# Put rlog output into $rlogout.
# If no rlog options are given,
# log the revisions checked in since the first ChangeLog entry.
case $rlog_options in
'')
date=1970
if test -s ChangeLog
then
# Add 1 to seconds to avoid duplicating most recent log.
e='
/^... ... [ 0-9][0-9] [ 0-9][0-9]:[0-9][0-9]:[0-9][0-9] [0-9]+ /{
'"$month_data"'
year = $5
for (i=0; i<=11; i++) if (m[i] == $2) break
dd = $3
hh = substr($0,12,2)
mm = substr($0,15,2)
ss = substr($0,18,2)
ss++
if (ss == 60) {
ss = 0
mm++
if (mm == 60) {
mm = 0
hh++
if (hh == 24) {
hh = 0
dd++
monthdays = mo[i+1] - mo[i]
if (i == 1 && year%4 == 0 && (year%100 != 0 || year%400 == 0)) monthdays++
if (dd == monthdays + 1) {
dd = 1
i++
if (i == 12) {
i = 0
year++
}
}
}
}
}
# Output comma instead of space to avoid CVS 1.5 bug.
printf "%d/%02d/%02d,%02d:%02d:%02d\n", year,i+1,dd,hh,mm,ss
exit
}
'
d=`$AWK "$e" <ChangeLog` || exit
case $d in
?*) date=$d
esac
fi
datearg="-d>$date"
esac
# If CVS is in use, examine its repository, not the normal RCS files.
if test ! -f CVS/Repository
then
rlog=rlog
repository=
else
rlog='cvs log'
repository=`sed 1q <CVS/Repository` || exit
test ! -f CVS/Root || CVSROOT=`cat <CVS/Root` || exit
case $CVSROOT in
*:/*)
# remote repository
;;
*)
# local repository
case $repository in
/*) ;;
*) repository=${CVSROOT?}/$repository
esac
if test ! -d "$repository"
then
echo >&2 "$0: $repository: bad repository (see CVS/Repository)"
exit 1
fi
esac
fi
# With no arguments, examine all files under the RCS directory.
case $# in
0)
case $repository in
'')
oldIFS=$IFS
IFS=$nl
case $recursive in
t)
RCSdirs=`find . -name RCS -type d -print`
filesFromRCSfiles='s|,v$||; s|/RCS/|/|; s|^\./||'
files=`
{
case $RCSdirs in
?*) find $RCSdirs -type f -print
esac
find . -name '*,v' -print
} |
sort -u |
sed "$filesFromRCSfiles"
`;;
*)
files=
for file in RCS/.* RCS/* .*,v *,v
do
case $file in
RCS/. | RCS/..) continue;;
RCS/.\* | RCS/\* | .\*,v | \*,v) test -f "$file" || continue
esac
files=$files$nl$file
done
case $files in
'') exit 0
esac
esac
set x $files
shift
IFS=$oldIFS
esac
esac
llogout=$TMPDIR/rcs2log$$l
rlogout=$TMPDIR/rcs2log$$r
trap exit 1 2 13 15
trap "rm -f $llogout $rlogout; exit 1" 0
case $rlog_options in
?*) $rlog $rlog_options ${1+"$@"} >$rlogout;;
'') $rlog "$datearg" ${1+"$@"} >$rlogout
esac || exit
# Get the full name of each author the logs mention, and set initialize_fullname
# to awk code that initializes the `fullname' awk associative array.
# Warning: foreign authors (i.e. not known in the passwd file) are mishandled;
# you have to fix the resulting output by hand.
initialize_fullname=
initialize_mailaddr=
case $loginFullnameMailaddrs in
?*)
case $loginFullnameMailaddrs in
*\"* | *\\*)
sed 's/["\\]/\\&/g' >$llogout <<EOF || exit
$loginFullnameMailaddrs
EOF
loginFullnameMailaddrs=`cat $llogout`
esac
oldIFS=$IFS
IFS=$nl
for loginFullnameMailaddr in $loginFullnameMailaddrs
do
case $loginFullnameMailaddr in
*"$tab"*) IFS=$tab;;
*) IFS=:
esac
set x $loginFullnameMailaddr
login=$2
fullname=$3
mailaddr=$4
initialize_fullname="$initialize_fullname
fullname[\"$login\"] = \"$fullname\""
initialize_mailaddr="$initialize_mailaddr
mailaddr[\"$login\"] = \"$mailaddr\""
done
IFS=$oldIFS
esac
case $llogout in
?*) sort -u -o $llogout <<EOF || exit
$logins
EOF
esac
output_authors='/^date: / {
if ($2 ~ /^[0-9]*[-\/][0-9][0-9][-\/][0-9][0-9]$/ && $3 ~ /^[0-9][0-9]:[0-9][0-9]:[0-9][0-9][-+0-9:]*;$/ && $4 == "author:" && $5 ~ /^[^;]*;$/) {
print substr($5, 1, length($5)-1)
}
}'
authors=`
$AWK "$output_authors" <$rlogout |
case $llogout in
'') sort -u;;
?*) sort -u | comm -23 - $llogout
esac
`
case $authors in
?*)
cat >$llogout <<EOF || exit
$authors
EOF
initialize_author_script='s/["\\]/\\&/g; s/.*/author[\"&\"] = 1/'
initialize_author=`sed -e "$initialize_author_script" <$llogout`
awkscript='
BEGIN {
alphabet = "abcdefghijklmnopqrstuvwxyz"
ALPHABET = "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
'"$initialize_author"'
}
{
if (author[$1]) {
fullname = $5
if (fullname ~ /[0-9]+-[^(]*\([0-9]+\)$/) {
# Remove the junk from fullnames like "0000-Admin(0000)".
fullname = substr(fullname, index(fullname, "-") + 1)
fullname = substr(fullname, 1, index(fullname, "(") - 1)
}
if (fullname ~ /,[^ ]/) {
# Some sites put comma-separated junk after the fullname.
# Remove it, but leave "Bill Gates, Jr" alone.
fullname = substr(fullname, 1, index(fullname, ",") - 1)
}
abbr = index(fullname, "&")
if (abbr) {
a = substr($1, 1, 1)
A = a
i = index(alphabet, a)
if (i) A = substr(ALPHABET, i, 1)
fullname = substr(fullname, 1, abbr-1) A substr($1, 2) substr(fullname, abbr+1)
}
# Quote quotes and backslashes properly in full names.
# Do not use gsub; traditional awk lacks it.
quoted = ""
rest = fullname
for (;;) {
p = index(rest, "\\")
q = index(rest, "\"")
if (p) {
if (q && q<p) p = q
} else {
if (!q) break
p = q
}
quoted = quoted substr(rest, 1, p-1) "\\" substr(rest, p, 1)
rest = substr(rest, p+1)
}
printf "fullname[\"%s\"] = \"%s%s\"\n", $1, quoted, rest
author[$1] = 0
}
}
'
initialize_fullname=`
(
cat /etc/passwd
for author in $authors
do nismatch $author passwd.org_dir
done
ypmatch $authors passwd
) 2>/dev/null |
$AWK -F: "$awkscript"
`$initialize_fullname
esac
# Function to print a single log line.
# We don't use awk functions, to stay compatible with old awk versions.
# `Log' is the log message (with \n replaced by \r).
# `files' contains the affected files.
printlogline='{
# Following the GNU coding standards, rewrite
# * file: (function): comment
# to
# * file (function): comment
if (Log ~ /^\([^)]*\): /) {
i = index(Log, ")")
files = files " " substr(Log, 1, i)
Log = substr(Log, i+3)
}
# If "label: comment" is too long, break the line after the ":".
sep = " "
if ('"$length"' <= '"$indent"' + 1 + length(files) + index(Log, CR)) sep = "\n" indent_string
# Print the label.
printf "%s*%s:", indent_string, files
# Print each line of the log, transliterating \r to \n.
while ((i = index(Log, CR)) != 0) {
logline = substr(Log, 1, i-1)
if (logline ~ /[^'"$tab"' ]/) {
printf "%s%s\n", sep, logline
} else {
print ""
}
sep = indent_string
Log = substr(Log, i+1)
}
}'
case $hostname in
'')
hostname=`(
hostname || uname -n || uuname -l || cat /etc/whoami
) 2>/dev/null` || {
echo >&2 "$0: cannot deduce hostname"
exit 1
}
case $hostname in
*.*) ;;
*)
domainname=`(domainname) 2>/dev/null` &&
case $domainname in
*.*) hostname=$hostname.$domainname
esac
esac
esac
# Process the rlog output, generating ChangeLog style entries.
# First, reformat the rlog output so that each line contains one log entry.
# Transliterate \n to \r so that multiline entries fit on a single line.
# Discard irrelevant rlog output.
$AWK <$rlogout '
BEGIN { repository = "'"$repository"'" }
/^RCS file:/ {
if (repository != "") {
filename = $3
if (substr(filename, 1, length(repository) + 1) == repository "/") {
filename = substr(filename, length(repository) + 2)
}
if (filename ~ /,v$/) {
filename = substr(filename, 1, length(filename) - 2)
}
}
}
/^Working file:/ { if (repository == "") filename = $3 }
/^date: /, /^(-----------*|===========*)$/ {
if ($0 ~ /^branches: /) { next }
if ($0 ~ /^date: [0-9][- +\/0-9:]*;/) {
date = $2
if (date ~ /-/) {
# An ISO format date. Replace all "-"s with "/"s.
newdate = ""
while ((i = index(date, "-")) != 0) {
newdate = newdate substr(date, 1, i-1) "/"
date = substr(date, i+1)
}
date = newdate date
}
# Ignore any time zone; ChangeLog has no room for it.
time = substr($3, 1, 8)
author = substr($5, 1, length($5)-1)
printf "%s %s %s %s %c", filename, date, time, author, 13
next
}
if ($0 ~ /^(-----------*|===========*)$/) { print ""; next }
printf "%s%c", $0, 13
}
' |
# Now each line is of the form
# FILENAME YYYY/MM/DD HH:MM:SS AUTHOR \rLOG
# where \r stands for a carriage return,
# and each line of the log is terminated by \r instead of \n.
# Sort the log entries, first by date+time (in reverse order),
# then by author, then by log entry, and finally by file name (just in case).
sort +1 -3r +3 +0 |
# Finally, reformat the sorted log entries.
$AWK '
BEGIN {
# Some awk variants do not understand "\r" or "\013", so we have to
# put a carriage return directly in the file.
CR="
" # <-- There is a single CR between the " chars here.
# Initialize the fullname and mailaddr associative arrays.
'"$initialize_fullname"'
'"$initialize_mailaddr"'
# Initialize indent string.
indent_string = ""
i = '"$indent"'
if (0 < '"$tabwidth"')
for (; '"$tabwidth"' <= i; i -= '"$tabwidth"')
indent_string = indent_string "\t"
while (1 <= i--)
indent_string = indent_string " "
# Set up date conversion tables.
# RCS uses a nice, clean, sortable format,
# but ChangeLog wants the traditional, ugly ctime format.
# January 1, 0 AD (Gregorian) was Saturday = 6
EPOCH_WEEKDAY = 6
# Of course, there was no 0 AD, but the algorithm works anyway.
w[0]="Sun"; w[1]="Mon"; w[2]="Tue"; w[3]="Wed"
w[4]="Thu"; w[5]="Fri"; w[6]="Sat"
'"$month_data"'
}
{
newlog = substr($0, 1 + index($0, CR))
# Ignore log entries prefixed by "#".
if (newlog ~ /^#/) { next }
if (Log != newlog || date != $2 || author != $4) {
# The previous log and this log differ.
# Print the old log.
if (date != "") '"$printlogline"'
# Logs that begin with "{clumpname} " should be grouped together,
# and the clumpname should be removed.
# Extract the new clumpname from the log header,
# and use it to decide whether to output a blank line.
newclumpname = ""
sep = "\n"
if (date == "") sep = ""
if (newlog ~ /^\{[^'"$tab"' }]*}['"$tab"' ]/) {
i = index(newlog, "}")
newclumpname = substr(newlog, 1, i)
while (substr(newlog, i+1) ~ /^['"$tab"' ]/) i++
newlog = substr(newlog, i+1)
if (clumpname == newclumpname) sep = ""
}
printf sep
clumpname = newclumpname
# Get ready for the next log.
Log = newlog
if (files != "")
for (i in filesknown)
filesknown[i] = 0
files = ""
}
if (date != $2 || author != $4) {
# The previous date+author and this date+author differ.
# Print the new one.
date = $2
author = $4
# Convert nice RCS date like "1992/01/03 00:03:44"
# into ugly ctime date like "Fri Jan 3 00:03:44 1992".
# Calculate day of week from Gregorian calendar.
i = index($2, "/")
year = substr($2, 1, i-1) + 0
monthday = substr($2, i+1)
i = index(monthday, "/")
month = substr(monthday, 1, i-1) + 0
day = substr(monthday, i+1) + 0
leap = 0
if (2 < month && year%4 == 0 && (year%100 != 0 || year%400 == 0)) leap = 1
days_since_Sunday_before_epoch = EPOCH_WEEKDAY + year * 365 + int((year + 3) / 4) - int((year + 99) / 100) + int((year + 399) / 400) + mo[month-1] + leap + day - 1
# Print "date fullname (email address)".
# Get fullname and email address from associative arrays;
# default to author and author@hostname if not in arrays.
if (fullname[author])
auth = fullname[author]
else
auth = author
printf "%s %s %2d %s %d %s ", w[days_since_Sunday_before_epoch%7], m[month-1], day, $3, year, auth
if (mailaddr[author])
printf "<%s>\n\n", mailaddr[author]
else
printf "<%s@%s>\n\n", author, "'"$hostname"'"
}
if (! filesknown[$1]) {
filesknown[$1] = 1
if (files == "") files = " " $1
else files = files ", " $1
}
}
END {
# Print the last log.
if (date != "") {
'"$printlogline"'
printf "\n"
}
}
' &&
# Exit successfully.
exec rm -f $llogout $rlogout