diff options
Diffstat (limited to 'tools/warnstats.sh')
-rwxr-xr-x | tools/warnstats.sh | 318 |
1 files changed, 318 insertions, 0 deletions
diff --git a/tools/warnstats.sh b/tools/warnstats.sh new file mode 100755 index 000000000..96bbfb8bc --- /dev/null +++ b/tools/warnstats.sh @@ -0,0 +1,318 @@ +#!/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===================================================== + +### +### warnstats +### +### AUTHOR(S): +### Thomas Kulik, Deutsche Telekom AG, 2020 +### +### DESCRIPTION: +### warnstat helps to find the onap modules (projects) and rst-files which are +### responsible for the most warnings during the documentation build process +### it requires a tox build logfile, parses it line by line and prints out some +### statistics +### + +### +### CHANGELOG (LATEST ON TOP) +### +### 1.3.1 (2020-03-10) fixed minor typo in usage message +### 1.3.0 (2020-03-09) initial release +### + +script_version="1.3.1 (2020-03-10)" + +echo " "; +echo "warnstats - Version ${script_version}"; +echo " "; + +declare -A module_array +declare -A message_short_array +declare -A message_long_array +declare -A rstfile_array + +### +### simple script argument handling +### + +logfile=$1; + +# check if there is an argument at all +if [[ "$logfile" == "" ]] ; then + echo 'Usage: warnstats [tox-logfile]' + exit 1 +fi + +# check if argument is a file +if [ ! -f $logfile ] ; then + echo "Error: can't find tox-logfile \"$logfile\"" + exit 1 +fi + +# read in the tox build logfile - use only lines which contain a warning +readarray -t logfile_array < <(grep ": WARNING:" $logfile); + +# process filtered logfile line by line +for line in "${logfile_array[@]}" +do + # count warning lines + (( counter++ )); + echo -n -e "lines processed: $counter\r"; + + # extract module name from line + module=$(echo "$line" | sed -r 's:^/.+/doc/docs/(submodules|guides)/::' | cut -f1 -d\/); + + # in case the extraction has no valid name fill the missing field + if [[ "$module" == "" ]] ; then + module="<missing_module_name>"; + fi + + # extract rst file name from line and do some formatting to use it later as an array name + #echo "DBUG line: $line"; + rstfile=$(echo "$line" | grep -oP "[\w -]*\.rst"); + rstfile=$(echo -e ${rstfile} | tr '[:blank:]' '_'); + #echo "DBUG rst-file: $rstfile"; + + # count the number of warnings for the module/rstfile combination + (( rstfile_array[$module | $rstfile]++ )); + + # count the number of warnings for the single module + #echo "DBUG $module | $rstfile | $message"; + (( module_array[$module]++ )); + + # extract the warning message and do some formatting + #message=$(echo "$line" | sed -r 's:^/.+WARNING\:\ ::'); + message=$(echo "$line" | sed -r 's:^.+WARNING\:\ ::'); + message=$(echo -e ${message} | tr '[:blank:]' '_'); + message=$(echo -e ${message} | tr '/' '_'); + message=$(echo -e ${message} | tr '.' '_'); + + # remove all characters from message which may cause problems in the shell + message="$(echo -e "${message}" | sed -e 's/[^A-Za-z0-9_-]//g')"; + #echo "DBUG message=\"$message\"" + + # count the number of warnings for the single message (long version) + message_long="$(echo -e "${message}")"; + (( message_long_array[$message_long]++ )) + + # reduce length of message to group them more easily and then ... + # count the number of warnings for the single message (short version) + message_short="$(echo -e "${message}" | cut -c -20)"; + (( message_short_array[$message_short]++ )) + +done + +#format counter to have always x digits +counter=$(printf "%05d" $counter); +echo " "; +echo " $counter LINES WITH WARNING IN FILE '$logfile'"; + +echo " "; +echo "################################################################################"; +echo "~~~ MESSAGES LONG ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"; +echo "################################################################################"; +echo " "; + +#print array content and append to temporary outfile +for i in "${!message_long_array[@]}" +do + m=$i; + n=${message_long_array[$i]}; + ((nc += n)) + #format counter to have always x digits + n=$(printf "%05d" $n); + echo " $n | $m" >>tempoutfile; +done + +#format counter to have always x digits +nc=$(printf "%05d" $nc); +echo " $nc WARNINGS IN TOTAL WITH ${#message_long_array[@]} UNIQUE MESSAGES" >>tempoutfile; + +#print a sorted version of the temporary outfile +sort -br tempoutfile + +# clean up +rm tempoutfile +nc=0 + +echo " "; +echo "################################################################################"; +echo "~~~ MESSAGES SHORTENED (FOR BETTER GROUPING) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"; +echo "################################################################################"; +echo " "; + +#print array content and append to temporary outfile +for i in "${!message_short_array[@]}" +do + m=$i; + n=${message_short_array[$i]}; + ((nc += n)) + #format counter to have always x digits + n=$(printf "%05d" $n); + echo " $n | $m" >>tempoutfile; +done + +#format counter to have always x digits +nc=$(printf "%05d" $nc); +echo " $nc WARNINGS IN TOTAL WITH ${#message_short_array[@]} UNIQUE MESSAGES" >>tempoutfile; + +#print a sorted version of the temporary outfile +sort -br tempoutfile + +# clean up +rm tempoutfile +nc=0 + +echo " "; +echo "################################################################################"; +echo "~~~ MODULES ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"; +echo "################################################################################"; +echo " "; + +#create temporary outfile +for i in "${!module_array[@]}" +do + m=$i; + n=${module_array[$i]}; + ((nc += n)) + n=$(printf "%05d" $n); + echo " $n | $m" >>tempoutfile; +done + +#format counter to have always x digits +nc=$(printf "%05d" $nc); +echo " $nc WARNINGS IN TOTAL IN ${#module_array[@]} MODULES" >>tempoutfile; + +#print a sorted version of the temporary outfile +sort -br tempoutfile +rm tempoutfile +nc=0 + +echo " "; +echo "################################################################################"; +echo "~~~ MODULES WITH RSTFILES ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"; +echo "################################################################################"; +echo " "; + +#print array content and append to temporary outfile +for i in "${!rstfile_array[@]}" +do + m=$i; + n=${rstfile_array[$i]}; + ((nc += n)) + #format counter to have always x digits + n=$(printf "%05d" $n); + echo " $m | $n" >>tempoutfile; +done + +#format counter to have always x digits +nc=$(printf "%05d" $nc); +#in case the name (e.g) index.rst is used multiple times in the same module warnings are combined +echo " $nc WARNINGS IN TOTAL IN APPROX. ${#rstfile_array[@]} RST FILES" >>tempoutfile; + +#print a sorted version of the temporary outfile +sort -b tempoutfile + +# clean up +rm tempoutfile +nc=0 + +echo " "; +echo "################################################################################"; +echo "~~~ RSTFILES ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"; +echo "################################################################################"; +echo " "; + +#print array content and append to temporary outfile +for i in "${!rstfile_array[@]}" +do + m=$i; + n=${rstfile_array[$i]}; + ((nc += n)) + #format counter to have always x digits + n=$(printf "%05d" $n); + echo " $n | $m" >>tempoutfile; +done + +#format counter to have always x digits +nc=$(printf "%05d" $nc); +#in case the name (e.g) index.rst is used multiple times in the same module warnings are combined +echo " $nc WARNINGS IN TOTAL IN APPROX. ${#rstfile_array[@]} RST FILES" >>tempoutfile; + +#print a sorted version of the temporary outfile +sort -br tempoutfile + +# clean up +rm tempoutfile +nc=0 + +echo " "; +exit + +### +### backup code for future extensions +### + +# +# Block_quote_ends_without_a_blank_line_unexpected_unindent +# Bullet_list_ends_without_a_blank_line_unexpected_unindent +# Citation_[\w-]_is_not_referenced +# Citation_unit_test_is_not_referenced +# Content_block_expected_for_the_code_directive_none_found +# Content_block_expected_for_the_container_directive_none_found +# Could_not_lex_literal_block_as_bash__Highlighting_skipped +# Could_not_lex_literal_block_as_console__Highlighting_skipped +# Could_not_lex_literal_block_as_guess__Highlighting_skipped +# Could_not_lex_literal_block_as_json__Highlighting_skipped +# Could_not_lex_literal_block_as_yaml__Highlighting_skipped +# Definition_list_ends_without_a_blank_line_unexpected_unindent +# document_isnt_included_in_any_toctree +# download_file_not_readable +# Duplicate_explicit_target_name +# duplicate_label +# Enumerated_list_ends_without_a_blank_line_unexpected_unindent +# Error_in_code_directive +# Error_in_code-block_directive +# Error_in_image_directive +# Explicit_markup_ends_without_a_blank_line_unexpected_unindent +# Field_list_ends_without_a_blank_line_unexpected_unindent +# Footnote_[0-9.*]_is_not_referenced +# image_file_not_readable +# Include_file +# Inconsistent_literal_block_quoting +# Inline_emphasis_start-string_without_end-string +# Inline_interpreted_text_or_phrase_reference_start-string_without_end-string +# Inline_strong_start-string_without_end-string +# Inline_substitution_reference_start-string_without_end-string +# Literal_block_ends_without_a_blank_line_unexpected_unindent +# Literal_block_expected_none_found +# Malformed_table +# Pygments_lexer_name_asn_is_not_known +# Title_level_inconsistent +# Title_overline__underline_mismatch +# Title_overline_too_short +# Title_underline_too_short +# toctree_contains_reference_to_nonexisting_document +# Too_many_autonumbered_footnote_references_only_0_corresponding_footnotes_available +# undecodable_source_characters_replacing_with +# undefined_label +# Unexpected_indentation +# Unknown_directive_type_clode-block +# unknown_document +# Unknown_target_name |