#!/bin/bash #set -x # uncomment for bash script debugging ### ============================================================================ ### Licensed under the Apache License, Version 2.0 (the "License"); ### you may not use this file except in compliance with the License. ### You may obtain a copy of the License at ### ### http://www.apache.org/licenses/LICENSE-2.0 ### ### Unless required by applicable law or agreed to in writing, software ### distributed under the License is distributed on an "AS IS" BASIS, ### WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. ### See the License for the specific language governing permissions and ### limitations under the License. ### ============LICENSE_END===================================================== ### ### c2m ### ### AUTHOR(S): ### Thomas Kulik, Deutsche Telekom AG, 2020 ### ### DESCRIPTION: ### c2m automates additional tasks required in case you want to export and ### convert a set of wiki pages. the export and first conversion to markdown is ### done by confluence2md, provided by viaboxx. ### c2m processes a list of (to be exported) wiki pages, creates corresponding ### export directories, exports and converts pages (in various formats if ### required), opens an editor and cleans up afterwards. ### c2m checks also for problematic content in the export and creates a warning ### in case of detection. ### ### ISSUES: ### - markdown (md) output of confluence2md contains sometimes tags that are ### somehow "merged" with the topic headline; manual edit is required here ### ### OPEN: ### - confluence2md does not support all of the currently used confluence page ### types (structured-macros) - result for unsupported pages is ### "not satisfying"; enhancements (java) are required ### - opt: toc creation in root document in case you export a tree of documents ### to separate files ### - opt: remove wiki credentials from script ### ### REQUIRED: ### - pandoc, retext, confluence2md, java (older version for confluence2md), ### login for the confluence wiki ### ### SEE ALSO: ### - https://www.viaboxx.de/code/confluence2md/ ### - https://github.com/viaboxxsystems/confluence2md ### ### ### CHANGELOG (LATEST ON TOP) ### ### 1.1.0 (2020-03-10) added support for http/https proxy and anonymous wiki ### access. thx to eric, nicolas and sylvain (orange, france) ### confluence2md jar file now has to be in the same path as ### c2m. ### 1.0.0 (2020-03-09) initial release ### ### ### c2m example pagelist ### ### example pagelist (field descriptions below); it uses the delimiter "|" for ### the four fields per line. ### copy/paste page id and title from wiki; to get the wiki page_id you have to ### login to the wiki, open the page and choose e.g. the history. ### depth: use depth to follow down the child-pages hierarchy if required: ### -1=infinte, 0=no children, #=number of child-pages to follow. ### every hierarchy "0" entry will lead into the creation of a dedicated working ### directory where the page and child-pages are stored. ### for better readability you can add spaces to the list, but use "|" as a ### delimiter. lines starting with a # are filtered by c2m. ### ### hierarchy | page_id | page_title | depth ### ### 0 | 1018748 | ONAP Portal | 0 ### 1.1 | 1018759 | ONAP Portal for users | 0 ### 1.2 | 1018762 | ONAP Portal for administrators | 0 ### 1.2.1 | 1018764 | Admins | 0 ### 1.2.2 | 1018811 | Users | 0 ### 1.2.3 | 1018821 | Portal Admins | 0 ### 1.2.4 | 1018826 | Application Onboarding | 0 ### 1.2.5 | 1018832 | Widget Onboarding | 0 ### 1.2.6 | 1018835 | Edit Functional Menu | 0 ### 1.2.7 | 16004953 | Portal Microservices Onboarding | 0 ### ### in case you want to export to only one single output page (that contains all ### child-pages of the above example) use: ### ### 0 | 1018748 | ONAP Portal | -1 ### ### ### some initial variables ### script_version="1.1.0 (2020-03-10)" user="*****"; # replace ***** with your wiki login name passwd="*****"; # replace ***** with your wiki password credentials="${user}":"${passwd}"; server="https://wiki.onap.org"; rst_editor="retext --preview"; # remove credentials for those using anonymous access test "${credentials}" = "*****:*****" && credentials="" # explicit script dir to locate jar file basedir="$(cd "$(dirname "$0")"; pwd)" ### ### some inital tasks after script has been started ### ### ### print script version, date and time ### echo "INFO ***************************************************************************" echo "INFO c2m Version ${script_version}, started $(date)"; ### ### simple script argument handling ### page_list=$1; # check if there is an argument at all if [[ "$page_list" == "" ]] ; then echo 'Usage: c2m [PAGELIST]' exit 1 fi # check if argument is a file if [ ! -f $page_list ] ; then echo "Error: can't find pagelist \"$page_list\"" exit 1 fi ### ### declare the functions of this script ### ### ### function: create working directory; save (only the last) existing one; remove older versions; do some error handling ### function create_working_dir { # compose name for working directory #working_dir="${page_id}-${page_title}"; #working_dir="${page_title}-id${page_id}"; working_dir="${page_title}"; echo "INFO ***************************************************************************" echo "INFO working directory \"$working_dir\" will be created" # check if current working directory is already in the list if [[ " ${existing_working_dirs[@]} " =~ " ${working_dir} " ]]; then echo "ERRR ***************************************************************************" echo "ERRR working directory \"${working_dir}\" already exists - check entries in page_list for duplicates" echo "ERRR exiting ..." exit -1 else # store working_dir name for error handling existing_working_dirs+=(${working_dir}) fi # sample code #if [[ ! " ${array[@]} " =~ " ${value} " ]]; then # # whatever you want to do when arr doesn't contain value #fi # check existence of working directory if [ -d "$working_dir" ]; then # check existence of old saved working directory if [ -d "${working_dir}.old" ]; then # remove the old saved working directory rm -r "${working_dir}.old"; fi # save (only) the latest working directory mv $working_dir "$working_dir.old"; fi # finally create the working directory and cd into it mkdir $working_dir; cd $working_dir; } ### ### function: pull pages from wiki - currently we are testing some export variations ### function pull_pages_from_wiki { # define outfile name #out_file="${page_title}-id${page_id}"; out_file="${page_title}"; # set proxy for those who need test -n "${http_proxy}" && proxy="$(echo $http_proxy |sed -e 's,http://,-Dhttp.proxyHost=,' -e 's/:/ -Dhttp.proxyPort=/' -e 's:/$::')" test -n "${https_proxy}" && proxy="$proxy $(echo $https_proxy |sed -e 's,http://,-Dhttps.proxyHost=,' -e 's/:/ -Dhttps.proxyPort=/' -e 's:/$::')" # pull pages from wiki and convert to markdown (as a source for conversion by pandoc) java $proxy -jar "${basedir}"/confluence2md-2.1-fat.jar +H true +T false +RootPageTitle false +FootNotes true -maxHeaderDepth 7 -depth $depth -v true -o ${out_file}.md -u "${credentials}" -server $server $page_id } ### ### function: simple search and (red colored) warning if special terms are detected in the md output file ### function detect_unwanted_content_in_md_outfile { for search_term in "ecomp" "wiki.onap.com" "10.53.199.7" "at&t" do if grep $search_term ${out_file}.md; then echo -e "\e[31mWARN ***************************************************************************\e[39m"; echo -e "\e[31mWARN term \"${search_term}\" detected in ${out_file}.md\e[39m"; fi done } ### ### function: pandoc conversion from md (variants) to rst - currenty testing some conversion formats ### function convert_md_outfile_to_rst { #depending on the given source format (--from) the results may vary #pandoc -s --toc --toc-depth=5 --from markdown_mmd --to rst "${out_file}.md" -o "${out_file}-markdown_mmd.rst" #pandoc -s --toc --toc-depth=5 --from markdown_strict --to rst "${out_file}.md" -o "${out_file}-markdown_strict.rst" #pandoc -s --toc --toc-depth=5 --from markdown_phpextra --to rst "${out_file}.md" -o "${out_file}-markdown_phpextra.rst" #pandoc -s --toc-depth=5 --from markdown_phpextra --to rst "${out_file}.md" -o "${out_file}-markdown_phpextra.rst" pandoc -s --toc-depth=5 --from markdown_phpextra --to rst "${out_file}.md" -o "${out_file}.rst" } ### ### function: check results in rst editor ### function open_rst_editor { #echo "DBUG ***************************************************************************" #echo "DBUG open \"${out_file}\*.rst\" with rst editor" $rst_editor ${out_file}*.rst & } ### ### function: clean up export directories from files no longer needed ### function clean_up { rm *.md 2>/dev/null rm attachments/*.json 2>/dev/null rm attachments/.*.json 2>/dev/null } ### ### main: let's start the work ... ### # read in pagelist file, filter lines starting with a comment and create an array that contains all (uncommented) lines of the file # sample code # IFS=',' read -r -a page_array <<< "$page_list" # in case $page_list was defined as a varable in this script; use "," as the delimiter #readarray -t page_array < $page_list; # old version readarray -t page_array < <(grep -v "^#" $page_list); # new version which skips line with comments # INFO: show list of pages by printing every line of the array echo "INFO ***************************************************************************" for line in "${page_array[@]}" do echo "INFO $line" done # the main loop reads the page_array line by line and processes the content for line in "${page_array[@]}" do # cut out values from the current line (delimiter is now the "|") and assign them to the correct variables hierarchy=$(echo $line | cut -f1 -d\|) page_id=$(echo $line | cut -f2 -d\|) page_title=$(echo $line | cut -f3 -d\|) depth=$(echo $line | cut -f4 -d\|) # remove leading and trailing spaces from variables hierarchy="$(echo -e "${hierarchy}" | sed -e 's/^[[:space:]]*//' -e 's/[[:space:]]*$//')"; page_id="$(echo -e "${page_id}" | sed -e 's/^[[:space:]]*//' -e 's/[[:space:]]*$//')"; page_title="$(echo -e "${page_title}" | sed -e 's/^[[:space:]]*//' -e 's/[[:space:]]*$//')"; depth="$(echo -e "${depth}" | sed -e 's/^[[:space:]]*//' -e 's/[[:space:]]*$//')"; # substitude all blanks in page_title with a minus sign page_title=$(echo -e ${page_title} | tr '[:blank:]' '-'); echo "DBUG page_title=\"$page_title\"" # convert page_title to lowercase page_title=$(echo -e ${page_title} | tr '[:upper:]' '[:lower:]'); #echo "DBUG page_title=\"$page_title\"" # remove all characters from page_title which may cause problems in the shell ... or are reserved by conventions of this script #page_title="$(echo -e "${page_title}" | sed -e 's/[^A-Za-z0-9._-]//g')"; # a less strict version page_title="$(echo -e "${page_title}" | sed -e 's/[^A-Za-z0-9-]//g')"; echo "DBUG page_title=\"$page_title\"" # INFO: print variables to check content echo "INFO ***************************************************************************" echo "INFO hierarchy = \"$hierarchy\"" echo "INFO page_id = \"$page_id\"" echo "INFO page_title = \"$page_title\"" echo "INFO depth = \"$depth\"" # create working directory - done for every! "hierarchy 0" entry of page_list if [ "$hierarchy" == "0" ] then create_working_dir fi # call functions to process page pull_pages_from_wiki detect_unwanted_content_in_md_outfile convert_md_outfile_to_rst open_rst_editor clean_up # main loop end done # bye! echo "INFO ***************************************************************************" echo "INFO c2m Version ${script_version}, ended $(date)" echo "" exit 0