Contrib: update-copyright.sh general improvements

 - make lookup of last modification year more robust

 - add a --pedantic mode that greps the first year a file was modified
   from git log

 - restructure processing into a separate bash function and add parallel
   processing structure

 - run script on all files of the repository

contrib/utilities/update-copyright.sh

@@ -14,6 +14,7 @@
 ##
 ## ---------------------------------------------------------------------
 
+set -u
 
 # Purpose: Update the copyright year of every file based on the last
 #          modification recorded in the git logs
@@ -24,51 +25,144 @@ if test ! -d source -o ! -d include -o ! -d examples ; then
   exit
 fi
 
+processes=1
+accurate_first_year=false
+until [[ "$@" == "" ]]; do
+  case $1 in
+    --pedantic)
+      accurate_first_year=true
+      shift;;
+    -j)
+      shift
+      if [[ "$@" == "" ]]; then
+        echo "Error: »-j« must be followed by a number" > /dev/stderr
+        echo "Usage: update-copyright.sh [--pedantic] [-j N]" > /dev/stderr
+        exit 1
+      fi
+      processes="${1}"
+      shift;;
+    *)
+      echo "Error: invalid option »$1«" > /dev/stderr
+      echo "Usage: update-copyright.sh [--pedantic] [-j N]" > /dev/stderr
+      exit 1;;
+  esac
+done
+
+#
+# A shell function that updates the copyright string for a given file $1:
+#
+
+update_copyright()
+{
+  file="${1}"
+
+  if ! [ -f ${file} ]; then
+    echo "Skipping ${file}: not a file"
+    return
+  fi
+
+  if ! head -13 ${file} | grep -q "^.. This file is part of the deal.II library.$" ; then
+    echo "Skipping ${file}: no deal.II copyright header"
+    return
+  fi
+
+  #
+  # Get the last year this file was modified from the git log. We don't
+  # want to see patches that just updated the copyright year, thus find the
+  # first commit that
+  #  - does not mention both the words "update" and "copyright", as well as
+  #  - "Update license headers".
+  #
+
+  last_year=`git log -n 3 --date=short --format="format:%cd %s" ${file} | \
+    egrep -i -v "update.*copyright|copyright.*update|Update license headers" | \
+    head -n 1 | \
+    perl -p -e 's/^(\d\d\d\d)-.*/\1/g;'`
+
+  #
+  # It should not happen, that the grep removes all 3 most recent commits
+  # simultaneously but if it does then run the git log command again with
+  # full history:
+  #
+
+  [ -z "$last_year" ] && last_year=`git log --date=short --format="format:%cd %s" ${file} | \
+      egrep -i -v "update.*copyright|copyright.*update|Update license headers" | \
+      head -n 1 | \
+      perl -p -e 's/^(\d\d\d\d)-.*/\1/g;'`
+
+  if [ -z "$last_year" ]; then
+    echo "Skipping ${file}: internal error: could not determine last copyright year"
+    return
+  fi
 
-files="
-  $(echo contrib/*/*.{py,sh} \
-         contrib/python-bindings/CMakeLists.txt \
-         contrib/python-bindings/*/*.{h,cc,py} \
-         contrib/utilities/{update-copyright,indent} \
-         doc/doxygen/*/*.{h,h.in} \
-         doc/doxygen/scripts/*.pl \
-         doc/screen.css)
-  $(find include/ source/ examples/ | egrep '\.(h|in|cc|cu)$')
-  $(find cmake/ | egrep '\.(cmake|in|cc)$')
-  $(find . -name CMakeLists.txt)
-  $(find tests/ | egrep '\.(h|cc)$')
-  $(find doc/ | egrep '\.html$')
-"
-
-
-for i in $files ; do
-  # get the last year this file was modified from the git log.
-  # we don't want to see patches that just updated the copyright
-  # year, so output the dates and log messages of the last 3
-  # commits, throw away all that mention both the words
-  # "update" and "copyright", and take the year of the first
-  # message that remains
   #
-  # (it should be enough to look at the last 2 messages since
-  # ideally no two successive commits should have updated the
-  # copyright year. let's err on the safe side and take the last
-  # 3 commits.)
-  last_year=`git log -n 3 --date=short --format="format:%cd %s" $i | \
-             egrep -i -v "update.*copyright|copyright.*update" | \
-             head -n 1 | \
-             perl -p -e 's/^(\d\d\d\d)-.*/\1/g;'`
-
-  # get the first year this file was modified from the actual
-  # file. this may predate the git log if the file was copied
-  # from elsewhere
-  first_year=`cat $i | egrep 'Copyright \(C\) [0-9]{4}' | \
+  # Get the first year this file was modified from the actual file. This is
+  # fast but might be inaccurate.
+  #
+
+  first_year=`egrep 'Copyright \(C\) [0-9]{4}' ${file} | \
               perl -p -e "s/.*Copyright \(C\) (\d{4}).*/\1/g;"`
 
-  # print a status message. we really only have to update
-  # the copyright year if the first and last year are
-  # different
-  echo "Processing $i: ${first_year} - ${last_year}"
-  if test ! "${first_year}" = "${last_year}" ; then
-    perl -pi -e "s/(Copyright \(C\) \d{4})( - \d{4})?(, \d{4}( - \d{4})?)*/\1 - ${last_year}/g;" $i
+  if $accurate_first_year; then
+    #
+    # Get the first (plausible) year this file was modified. While each file
+    # (ideally) already contains a start year, experience suggests that this
+    # information is typically wildly incorrect because files (and copyright
+    # headers) get copied all the time. We thus grab this information from
+    # git history (which is a rather expensive operation - but so is running
+    # this script in the first place).
+    #
+    #  * In order to make the git log lookup as  fast as possible we use
+    #    `--diff-filter=A` to skip all but the first commit in which the file
+    #    was created.
+    #
+    # We try to find simple renames with the `--follow` toggle. Note that
+    # we enforce a 100% similarity when trying to find a potential rename
+    # candiate with `-M100%`
+    #
+    # In corner cases when new, almost empty files are created then "git
+    # log --follow" might be too aggressive. As a sanity check let's
+    # restrict the admissible date range to $first_year:
+    #
+    git_first_year=`git log --since="$first_year" --follow -M100% --diff-filter=A --date=short --format="format:%cd %s" ${file} | \
+      tail -n 1 | \
+      perl -p -e 's/^(\d\d\d\d)-.*/\1/g;'`
+
+    # make sure that ${first_year} is set to a valid variable
+    first_year="${git_first_year:-${first_year}}"
   fi
-done
+
+  if [ -z "$first_year" ]; then
+    echo "Skipping ${file}: internal error: could not determine first copyright year"
+    return
+  fi
+
+  #
+  # Print a status message and update copyright line:
+  #
+
+  if [ "${first_year}" = "${last_year}" ]; then
+    echo "Processing ${file}: ${last_year}"
+    perl -pi -e "s/(^.. Copyright \(C\)) \d{4}( - \d{4})?(, \d{4}( - \d{4})?)*/\1 ${last_year}/g if 1..13;" ${file}
+  else
+    echo "Processing ${file}: ${first_year} - ${last_year}"
+    perl -pi -e "s/(^.. Copyright \(C\)) \d{4}( - \d{4})?(, \d{4}( - \d{4})?)*/\1 ${first_year} - ${last_year}/g if 1..13;" ${file}
+  fi
+}
+
+#
+# Run copyright update in parallel:
+#
+
+process()
+{
+  i=0
+  find ${1} -type f -regextype egrep -regex "${2}" | while read file; do
+    (( i=i%processes )); (( i++==0 )) && wait
+    update_copyright "${file}" &
+  done
+}
+
+process "."  "CMakeLists.txt|CTestConfig.cmake" update_copyright
+process "cmake contrib doc examples include source tests" ".*" update_copyright
+