Add initial files from envbot v0.1-beta1
This commit is contained in:
22
tools/bashdoc/ChangeLog
Normal file
22
tools/bashdoc/ChangeLog
Normal file
@@ -0,0 +1,22 @@
|
||||
Sun Oct 14 17:24:19 CEST 2007; Arvid Norlander
|
||||
The last two days I have changed so much on bashdoc that I lost count of it
|
||||
Some nice things:
|
||||
Cleaned up code.
|
||||
XHTML and CSS.
|
||||
Made bashdoc able to document variables.
|
||||
Fix many bugs.
|
||||
|
||||
Sun Aug 13 13:08:12 PDT 2006; Unknown
|
||||
Fix bug which caused either extra garbage around function links in src or
|
||||
skipped some interlinks.
|
||||
Release bashdoc-0.1.8
|
||||
|
||||
Sun Jan 22 16:22:45 EST 2006; Unknown
|
||||
ChangeLog: Started
|
||||
src2html: changed sed to fix the weird invalid range end erors
|
||||
Changes suggested by Jaka Kranjc
|
||||
generate-smgl-docs: Add handy example doc generation
|
||||
script, written by and for SourceMage
|
||||
generate-smgl-docs: Patch by Jaka Kranjc
|
||||
generate-smgl-docs: Add -q to bashdoc call to limit useless output
|
||||
Release: bashdoc-0.1.7
|
||||
726
tools/bashdoc/bashdoc.sh
Executable file
726
tools/bashdoc/bashdoc.sh
Executable file
@@ -0,0 +1,726 @@
|
||||
#!/usr/bin/env bash
|
||||
# -*- coding: utf-8 -*-
|
||||
#--------------------------
|
||||
## @Synopsis Reads specialy formated shell scripts and creates docs
|
||||
## @Copyright Copyright 2003, Paul Mahon
|
||||
## @Copyright Copyright 2007, Arvid Norlander
|
||||
## @License GPL v2
|
||||
## Parses comments between lines of '#---'
|
||||
## Lines to be parsed start with ##. All tags start with @.
|
||||
## Lines without a tag are considered simple description of the section.
|
||||
## If the line following the comment block doesn't start with 'function'
|
||||
## the it's assumed that the comment is for the whole file. Only the first
|
||||
## non-function comment block will be used, the other will be ignored.
|
||||
## <p>
|
||||
## Multiple identical tags are allowed, the contents are appended and separated
|
||||
## with a space. @param tags are treated specials and are assumed to be in order.
|
||||
## <p>
|
||||
## There is an additional <@function FUNCTION_NAME> tag that can be embeded
|
||||
## in any bashdoc comment. It will be transformed into a link to that function.
|
||||
## Note, this will only work for functions that are defined in the same script.
|
||||
## <p><pre>
|
||||
## Usage: [OPTIONS] [--] script [ script ...]
|
||||
## -p, --project project Name of the project
|
||||
## -o, --output directory Specifies the directory you want the resulting html to go into
|
||||
## -c, --nocss Do not write default CSS file.
|
||||
## -e, --exclusive tag Only output if the block has this tag
|
||||
## -q, --quiet Quiet the output
|
||||
## -h, --help Display this help and exit
|
||||
## -V, --version Output version information and exit
|
||||
## -- No more arguments, only scripts
|
||||
## script The script you want documented
|
||||
##</pre>
|
||||
##
|
||||
#--------------------------
|
||||
|
||||
# Make env sane
|
||||
unset LC_CTYPE LC_NUMERIC LC_TIME LC_COLLATE LC_MONETARY
|
||||
unset LC_MESSAGES LC_PAPER LC_NAME LC_ADDRESS
|
||||
unset LC_TELEPHONE LC_MEASUREMENT LC_IDENTIFICATION
|
||||
export LC_ALL=C
|
||||
export LANG=C
|
||||
|
||||
# Check bash version. We need at least 3.2.x
|
||||
# Lets not use anything like =~ here because
|
||||
# that may not work on old bash versions.
|
||||
if [[ "$(awk -F. '{print $1 $2}' <<< $BASH_VERSION)" -lt 32 ]]; then
|
||||
echo "Sorry your bash version is too old!"
|
||||
echo "You need at least version 3.2 of bash"
|
||||
echo "Please install a newer version:"
|
||||
echo " * Either use your distro's packages"
|
||||
echo " * Or see http://www.gnu.org/software/bash/"
|
||||
exit 2
|
||||
fi
|
||||
|
||||
# To make set -x more usable
|
||||
export PS4='(${BASH_SOURCE}:${LINENO}): ${FUNCNAME[0]} : '
|
||||
|
||||
VERSION="0.1.8"
|
||||
HEADERS="<!-- Generated by bashdoc version $VERSION, on $(date +'%Y-%m-%d'). -->
|
||||
<link rel=\"stylesheet\" href=\"style.css\" type=\"text/css\" />
|
||||
<meta http-equiv=\"Content-Type\" content=\"text/html; charset=UTF-8\" />"
|
||||
|
||||
GOOD=$'\e[32;01m'
|
||||
WARN=$'\e[33;01m'
|
||||
BAD=$'\e[31;01m'
|
||||
NORMAL=$'\e[0m'
|
||||
|
||||
#--------------------------
|
||||
## Output error message
|
||||
## @param Message
|
||||
## @Stderr Formated message
|
||||
#--------------------------
|
||||
print_error () {
|
||||
echo -e " ${BAD}*${NORMAL} $*" >&2
|
||||
}
|
||||
#--------------------------
|
||||
## Output warning message
|
||||
## @param Message
|
||||
## @Stderr Formated message
|
||||
#--------------------------
|
||||
print_warn () {
|
||||
echo -e " ${WARN}*${NORMAL} $*" >&2
|
||||
}
|
||||
#--------------------------
|
||||
## Output info message
|
||||
## @param Message
|
||||
## @Stderr Formated message
|
||||
#--------------------------
|
||||
print_info () {
|
||||
echo -e " ${GOOD}*${NORMAL} $*" >&2
|
||||
}
|
||||
#--------------------------
|
||||
## Output debug message
|
||||
## @param Message
|
||||
## @Stderr Formated message
|
||||
#--------------------------
|
||||
print_debug () {
|
||||
echo -e " $*" >&2
|
||||
}
|
||||
|
||||
#--------------------------
|
||||
## @Arguments -r: recursive, -o [directory]: output html
|
||||
## Parses arguments for this script
|
||||
## @Gobals RECURSIVE, OUT_DIR
|
||||
#--------------------------
|
||||
function args()
|
||||
{
|
||||
local retVal=0
|
||||
QUIET=0
|
||||
while true ; do
|
||||
case $1 in
|
||||
-p|--project)
|
||||
PROJECT="$2"
|
||||
(( retVal+=2 ))
|
||||
shift 2
|
||||
;;
|
||||
-o|--output)
|
||||
OUT_DIR="$2"
|
||||
(( retVal+=2 ))
|
||||
shift 2
|
||||
;;
|
||||
-c|--nocss)
|
||||
NOCSS="1"
|
||||
(( retVal+=2 ))
|
||||
shift 1
|
||||
;;
|
||||
-h|--help)
|
||||
usage
|
||||
exit 0
|
||||
;;
|
||||
-V|--version)
|
||||
version
|
||||
exit 0
|
||||
;;
|
||||
-e|--exclusive)
|
||||
EXCLUSIVE="${2%%=*}"
|
||||
EXCLUSIVE_VAL="${2#*=}"
|
||||
(( retVal+=2 ))
|
||||
shift 2
|
||||
;;
|
||||
-q|--quiet)
|
||||
(( QUIET+=1 ))
|
||||
(( retVal+=1 ))
|
||||
shift 1
|
||||
;;
|
||||
--)
|
||||
(( retVal++ ))
|
||||
return $retVal
|
||||
;;
|
||||
-*)
|
||||
usage
|
||||
exit 0
|
||||
;;
|
||||
*)
|
||||
[[ -e $1 ]] && return $retVal
|
||||
echo "$1 doesn't exist."
|
||||
usage
|
||||
exit 1
|
||||
;;
|
||||
esac
|
||||
done
|
||||
}
|
||||
|
||||
#-------------------------
|
||||
## Version for this script
|
||||
## @Stdout Version information
|
||||
#-------------------------
|
||||
function version()
|
||||
{
|
||||
echo "bashdoc $VERSION - Generate HTML documentation from bash scripts"
|
||||
echo ''
|
||||
echo 'Copyright (C) 2003 Paul Mahon'
|
||||
echo 'Copyright (C) 2007 Arvid Norlander'
|
||||
echo 'This is free software; see the source for copying conditions. There is NO'
|
||||
echo 'warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.'
|
||||
echo ''
|
||||
echo 'Written by Paul Mahon and modified by Arvid Norlander'
|
||||
}
|
||||
|
||||
#-------------------------
|
||||
## Usage for this script
|
||||
## @Stdout Usage information
|
||||
#-------------------------
|
||||
function usage()
|
||||
{
|
||||
cat <<- EOF
|
||||
bashdoc generates HTML documentation from bash scripts.
|
||||
|
||||
Usage: $(basename $0) [OPTIONS] [--] script [script ...]
|
||||
|
||||
Options:
|
||||
-p, --project project Name of the project
|
||||
-o, --output directory Specifies the directory you want the resulting html to go into
|
||||
-c, --nocss Do not write default CSS file.
|
||||
-e, --exclusive tag Only output if the block has this tag
|
||||
-q, --quiet Quiet the output
|
||||
-h, --help Display this help and exit
|
||||
-V, --version Output version information and exit
|
||||
-- No more arguments, only scripts
|
||||
script The script you want documented
|
||||
|
||||
Examples:
|
||||
bashdoc.sh -p bashdoc -o docs/ bashdoc.sh Generate documentation for this program.
|
||||
bashdoc.sh -p appname -o docs/ -e Type=API someapp.sh Generate documentation for someapp.sh,
|
||||
exclude items that do not include the tag
|
||||
@Type API
|
||||
EOF
|
||||
}
|
||||
|
||||
|
||||
#--------------------------
|
||||
## Reads until it has read an entire comment block. A block starts with
|
||||
## <br><pre>#---</pre></br>
|
||||
## Alone on a line, and continues until the next
|
||||
## <br><pre>#---</pre></br>
|
||||
## All comment lines inside should have ## at the start or they
|
||||
## will be ignored.
|
||||
##
|
||||
## @return 0 Possibly more blocks
|
||||
## @return 1 Unexpected end of file
|
||||
## @return 2 Expected end of file, no more blocks
|
||||
## @Stdin Reads a chunk
|
||||
## @Stdout Block with starting '##' removed
|
||||
## @Globals paramDesc, retDesc, desc, block, split, out_comment_block
|
||||
#--------------------------
|
||||
function get_comment_block()
|
||||
{
|
||||
local inComment commentBlock lastLine=""
|
||||
commentBlock=""
|
||||
while read LINE ; do
|
||||
(( srcLine++ ))
|
||||
if [[ ${LINE:0:4} == '#---' ]] ; then
|
||||
if [[ $inComment ]] ; then
|
||||
out_comment_block="$commentBlock"
|
||||
# I'm not sure why this is needed but it fixes incorrect line number
|
||||
(( srcLine++ ))
|
||||
return 0
|
||||
else
|
||||
inComment=yes
|
||||
fi
|
||||
elif [[ ${LINE:0:2} != '##' ]] && [[ $inComment ]] ; then
|
||||
[[ $QUIET -lt 1 ]] && print_warn "Line $srcLine of $FILE isn't a doc comment! Ignoring."
|
||||
[[ $QUIET -lt 1 ]] && print_warn "Line in question is: $LINE"
|
||||
elif [[ $inComment ]] ; then
|
||||
commentBlock="$commentBlock"$'\n'${LINE####}
|
||||
fi
|
||||
done
|
||||
#If we make it out here, we hit the end of the file
|
||||
if [[ $commentBlock ]] ; then
|
||||
#If there is a comment block started, then it never ended
|
||||
[[ $QUIET -lt 2 ]] && print_error "Unfinished comment block:"
|
||||
[[ $QUIET -lt 2 ]] && print_error "$commentBlock"
|
||||
return 1
|
||||
else
|
||||
return 2
|
||||
fi
|
||||
}
|
||||
|
||||
|
||||
#-----------------------
|
||||
## Parses the comments from stdin. Also reads the (non-commented)
|
||||
## function name. Mostly uses <@function parse_block> and
|
||||
## <@function output_parsed_block> to do the read work.
|
||||
## @Stdin Reads line after comment block
|
||||
## @Globals paramDesc, retDesc, desc, block, split, out_comment_block
|
||||
#-----------------------
|
||||
function parse_comments()
|
||||
{
|
||||
|
||||
#We use a lot of $( echo ... ) in here to trim the blanks
|
||||
|
||||
local funcLine funcName
|
||||
paramDesc=()
|
||||
retDesc=()
|
||||
local FIRST_BLOCK="yes"
|
||||
local skipRead
|
||||
local outBlock=""
|
||||
local lastOutBlock=""
|
||||
srcLine=0
|
||||
# 1 = function
|
||||
# 2 = variable
|
||||
itemtype=0
|
||||
while true ; do
|
||||
paramNames=()
|
||||
paramDesc=()
|
||||
split=()
|
||||
retDesc=()
|
||||
desc=""
|
||||
itemtype=0
|
||||
unset out_comment_block
|
||||
get_comment_block
|
||||
[[ $? -gt 0 ]] && break
|
||||
block="$out_comment_block"
|
||||
|
||||
if [[ $skipRead ]] ; then
|
||||
skipRead=""
|
||||
else
|
||||
funcLine=""
|
||||
funcName=""
|
||||
read funcLine
|
||||
fi
|
||||
# Is it a (global) variable?
|
||||
# Check before function to catch arrays.
|
||||
if [[ ${funcLine} =~ ^(declare -r +)?([a-zA-Z_][a-zA-Z0-9_]*)=.+$ ]]; then
|
||||
varName="${BASH_REMATCH[@]: -1}"
|
||||
itemtype=2
|
||||
# Is it a function?
|
||||
elif [[ ${funcLine%%[[:blank:]]*} == function ]] || [[ ${funcLine} =~ ^[^\ ]+\ *\(\)\ *\{?$ ]]; then
|
||||
funcName=$( echo ${funcLine#function} )
|
||||
funcName=$( echo ${funcName%%()*} )
|
||||
itemtype=1
|
||||
fi
|
||||
if [[ $funcName ]] || [[ $varName ]] || [[ $FIRST_BLOCK ]] ; then
|
||||
# Only bother with this block if it is a function block or
|
||||
# the first script block
|
||||
|
||||
#This fills in paramDesc[*], tag_*, retDesc
|
||||
parse_block
|
||||
lastOutBlock="$outBlock"
|
||||
outBlock=$(output_parsed_block)
|
||||
|
||||
if [[ $FIRST_BLOCK ]] && [[ ! $funcName ]] && [[ ! $varName ]]; then
|
||||
FIRST_BLOCK=""
|
||||
fi
|
||||
|
||||
if [[ $EXCLUSIVE ]] ; then
|
||||
# If this is first block, include it anyway.
|
||||
if [[ $funcName ]] || [[ $varName ]]; then
|
||||
local i="tag_${EXCLUSIVE}"
|
||||
if [[ ${!i} != $EXCLUSIVE_VAL ]] ; then
|
||||
if [[ $itemtype = 2 ]]; then
|
||||
funcName="$varName"
|
||||
fi
|
||||
print_debug "$funcName block ignored, no $EXCLUSIVE=$EXCLUSIVE_VAL tag."
|
||||
# Code duplication but hard to avoid
|
||||
for i in ${!tag_*} ; do
|
||||
unset $i
|
||||
done
|
||||
continue
|
||||
fi
|
||||
fi
|
||||
fi
|
||||
|
||||
for i in ${!tag_*} ; do
|
||||
unset $i
|
||||
done
|
||||
|
||||
if [[ $funcName ]]; then
|
||||
FUNC_LIST="$FUNC_LIST $funcName"
|
||||
elif [[ $varName ]]; then
|
||||
VAR_LIST="$VAR_LIST $varName"
|
||||
fi
|
||||
unset funcName varName
|
||||
echo "$outBlock"
|
||||
|
||||
else
|
||||
[[ $QUIET -lt 2 ]] && print_warn "Ignoring non-first non-function/variable comment block"
|
||||
[[ $QUIET -lt 1 ]] && print_warn "$block"
|
||||
fi
|
||||
done
|
||||
}
|
||||
|
||||
#---------------------
|
||||
## Create HTML from the non-special tags
|
||||
## @param var or func (is this for a variable or function)
|
||||
## @Stdout HTMLized tags
|
||||
#---------------------
|
||||
function output_parsed_tags() {
|
||||
local i
|
||||
for i in ${!tag_*} ; do
|
||||
# Convert _ in tags to space. Looks better.
|
||||
echo " <h3 class=\"othertag ${1}othertag ${i/tag_/tag-}\">$(sed 's/_/ /g' <<< "${i#tag_}")</h3>"
|
||||
# This may be fun, allow special formatting by tag.
|
||||
echo " <p class=\"othertag ${1}othertag ${i/tag_/tag-}\">"
|
||||
echo " ${!i}"
|
||||
echo " </p>"
|
||||
unset $i
|
||||
done
|
||||
}
|
||||
|
||||
#---------------------
|
||||
## Outputs the parsed information in a nice pretty format.
|
||||
## @Stdout formated documentation
|
||||
## @Globals paramDesc, retDesc, desc, block, split
|
||||
#---------------------
|
||||
function output_parsed_block()
|
||||
{
|
||||
echo "<hr />"
|
||||
if [[ $itemtype -eq 1 ]] && [[ $funcName ]]; then
|
||||
echo "<!-- Block for $funcName -->"
|
||||
echo " <h2 id=\"$funcName\" class=\"function\">function <strong>$funcName</strong>()</h2>"
|
||||
echo " <h3>Parameters:</h3>"
|
||||
echo " <ul class=\"paramerters\">"
|
||||
if [[ ${#paramDesc[*]} -gt 0 ]] ; then
|
||||
for(( i=0; i<"${#paramDesc[@]}"; i++ )) ; do
|
||||
echo " <li class=\"paramerters\">\$$[i+1]: ${paramDesc[i]}</li>"
|
||||
done
|
||||
else
|
||||
echo "<li>None</li>"
|
||||
fi
|
||||
echo " </ul>"
|
||||
if [[ ${#retDesc[*]} -gt 0 ]] ; then
|
||||
echo " <h3>Returns:</h3>"
|
||||
echo " <ul class=\"returns\">"
|
||||
for(( i=0; i<"${#retDesc[@]}"; i++ )) ; do
|
||||
echo " <li class=\"returns\">${retDesc[i]}</li>"
|
||||
done
|
||||
echo " </ul>"
|
||||
fi
|
||||
|
||||
output_parsed_tags func
|
||||
[[ $desc ]] && echo "<h3>Description</h3><p class=\"description funcdescription\">$desc</p>"
|
||||
elif [[ $itemtype -eq 2 ]]; then
|
||||
echo "<!-- Block for $varName -->"
|
||||
echo " <h2 id=\"$varName\" class=\"variable\">variable <strong>$varName</strong></h2>"
|
||||
output_parsed_tags var
|
||||
[[ $desc ]] && echo "<h3>Description</h3><p class=\"description vardescription\">$desc</p>"
|
||||
else
|
||||
echo '<!-- Header for whole script -->'
|
||||
echo "<h1>$FILE</h1>"
|
||||
echo " <p class=\"filedescription\">$desc</p>"
|
||||
echo "$desc" >> $SCRIPT_DESC
|
||||
|
||||
for i in ${!tag_*} ; do
|
||||
echo " <h3 class=\"fileothertag ${i/tag_/tag-}\">${i#tag_}</h3>"
|
||||
echo " <p class=\"fileothertag ${i/tag_/tag-}\">${!i}</p>"
|
||||
unset $i
|
||||
done
|
||||
fi
|
||||
|
||||
}
|
||||
|
||||
#---------------
|
||||
## Does the real work of the parsing. Tags start with @. Special
|
||||
## tags are @return and @param. Doc lines without a tag are
|
||||
## considered description.
|
||||
## @Globals paramDesc, retDesc, desc, block, split
|
||||
#---------------
|
||||
function parse_block()
|
||||
{
|
||||
local tag
|
||||
local backIFS="$IFS"
|
||||
IFS=$'\n'
|
||||
for LINE in $block; do
|
||||
IFS="$backIFS"
|
||||
LINE=$( echo $LINE )
|
||||
if [[ ${LINE:0:1} == '@' ]] ; then
|
||||
split_tag split $LINE
|
||||
case ${split} in
|
||||
@param)
|
||||
#paramNames[${#paramNames[*]}]=${split[1]}
|
||||
paramDesc=( "${paramDesc[@]}" "${split[1]}" )
|
||||
;;
|
||||
@return)
|
||||
retDesc=( "${retDesc[@]}" "${split[1]}" )
|
||||
;;
|
||||
@*)
|
||||
tag=${split[0]#@}
|
||||
local i="tag_${tag}"
|
||||
if [[ ${!i} ]] ; then
|
||||
local varname="tag_${tag}"
|
||||
eval "tag_${tag}=\"\${!varname}"$'\n'"\${split[1]}\""
|
||||
else
|
||||
eval "tag_${tag}=\"\${split[1]}\""
|
||||
fi
|
||||
;;
|
||||
*)
|
||||
print_error "We shouldn't get here... it was a tag, but not a tag?"
|
||||
;;
|
||||
esac
|
||||
else
|
||||
desc="$desc"$'\n'"$LINE"
|
||||
fi
|
||||
done
|
||||
IFS="$backIFS"
|
||||
}
|
||||
|
||||
#----------------
|
||||
## Splits a line that starts with a tag into tag and data.
|
||||
## @param Variable you want the result put into. Array is format is ( tag, data ).
|
||||
## @param Tag
|
||||
## @param Data
|
||||
## @Globals The variable in $1 will get the results
|
||||
#----------------
|
||||
function split_tag()
|
||||
{
|
||||
local out="${1}" ; shift
|
||||
local tag=$( echo ${1} ) ; shift
|
||||
# local key=$( echo ${1} ) ; shift
|
||||
local value=$( echo $* )
|
||||
eval "$out=( \"\$tag\" \"\${value}\" )"
|
||||
}
|
||||
|
||||
#--------------------
|
||||
## Outputs a header for script pages
|
||||
## @Stdout html header
|
||||
## @param Script name
|
||||
#--------------------
|
||||
function script_header()
|
||||
{
|
||||
cat <<- EOF > $OUT_FILE
|
||||
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
|
||||
<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en">
|
||||
<head>
|
||||
$HEADERS
|
||||
<title>$1 - $PROJECT</title>
|
||||
</head>
|
||||
<body>
|
||||
<p class="right">
|
||||
<a href="script_list.html">Script Index</a>
|
||||
</p>
|
||||
EOF
|
||||
}
|
||||
|
||||
# Initialise project variables
|
||||
OUT_DIR=$( dirname $0 )
|
||||
NOCSS=0
|
||||
args "$@"
|
||||
shift $?
|
||||
[[ $OUT_DIR ]] || OUT_DIR="."
|
||||
|
||||
# Create output directory in case it doesn't exist
|
||||
mkdir -p "$OUT_DIR" || {
|
||||
print_error "Failed to create output directory."
|
||||
exit 1
|
||||
}
|
||||
|
||||
if [[ $NOCSS = 0 ]]; then
|
||||
print_info "Writing CSS"
|
||||
# Copy stylesheet to output directory.
|
||||
cat <<- EOF > "${OUT_DIR}/style.css"
|
||||
/* Based on Trac CSS */
|
||||
body {
|
||||
background: #fff;
|
||||
color: #000;
|
||||
margin: 10px;
|
||||
padding: 0;
|
||||
}
|
||||
body, th, td {
|
||||
font: normal 13px verdana,arial,'Bitstream Vera Sans',helvetica,sans-serif;
|
||||
}
|
||||
h1, h2, h3, h4 {
|
||||
font-family: arial,verdana,'Bitstream Vera Sans',helvetica,sans-serif;
|
||||
font-weight: bold;
|
||||
letter-spacing: -0.018em;
|
||||
}
|
||||
h1 { font-size: 19px; margin: .15em 1em 0 0 }
|
||||
h2 { font-size: 16px; font-weight: normal; }
|
||||
h3 { font-size: 14px }
|
||||
hr { border: none; border-top: 1px solid #ccb; margin: 2em 0 }
|
||||
address { font-style: normal }
|
||||
img { border: none }
|
||||
tt { white-space: pre }
|
||||
:link, :visited {
|
||||
text-decoration: none;
|
||||
color: #b00;
|
||||
border-bottom: 1px dotted #bbb;
|
||||
}
|
||||
:link:hover, :visited:hover {
|
||||
background-color: #eee;
|
||||
color: #555;
|
||||
}
|
||||
h1 :link, h1 :visited ,h2 :link, h2 :visited, h3 :link, h3 :visited,
|
||||
h4 :link, h4 :visited, h5 :link, h5 :visited, h6 :link, h6 :visited {
|
||||
color: inherit;
|
||||
}
|
||||
/* Partly own stuff: */
|
||||
.nav body {
|
||||
margin: 0;
|
||||
padding: 0;
|
||||
background: inherit;
|
||||
color: inherit;
|
||||
}
|
||||
.nav ul { font-size: 11px; list-style: none; margin: 0; padding: 0; text-align: left }
|
||||
.nav li {
|
||||
display: block;
|
||||
padding: 0;
|
||||
margin: 0;
|
||||
white-space: nowrap;
|
||||
}
|
||||
/* Own stuff */
|
||||
.nav-header {
|
||||
font-weight: bold;
|
||||
}
|
||||
.right { text-align: right }
|
||||
.tag-Deprecated { color: #e00; }
|
||||
EOF
|
||||
else
|
||||
print_warn "Not writing a stylesheet. You will need to add your own by hand afterwards."
|
||||
fi
|
||||
|
||||
while [[ $# -gt 0 ]] ; do
|
||||
|
||||
#Initialise vars for this src
|
||||
FILE=$1
|
||||
[[ ! -f $FILE ]] || [[ ! -r $FILE ]] && {
|
||||
print_error "$FILE is not a file or is not readable, skipping."
|
||||
shift
|
||||
continue
|
||||
}
|
||||
print_info "Parsing $FILE"
|
||||
shift
|
||||
OUT_FILE=${FILE#/} #Remove leading /
|
||||
OUT_FILE="$OUT_DIR/${OUT_FILE//\//.}.html"
|
||||
FUNC_FILE="${OUT_FILE%.html}.funcs"
|
||||
VAR_FILE="${OUT_FILE%.html}.vars"
|
||||
SCRIPT_DESC="${OUT_FILE%.html}.desc"
|
||||
# Store real name (reuse in script list)
|
||||
REAL_NAME_FILE="${OUT_FILE%.html}.name"
|
||||
echo -n "${FILE#/}" > "$REAL_NAME_FILE"
|
||||
|
||||
FUNC_LIST=""
|
||||
VAR_LIST=""
|
||||
|
||||
#Start this src's html file
|
||||
script_header "$FILE"
|
||||
|
||||
# Parse and write out function list
|
||||
{
|
||||
parse_comments < $FILE
|
||||
echo "$FUNC_LIST" > $FUNC_FILE
|
||||
echo "$VAR_LIST" > $VAR_FILE
|
||||
# Convert references like <@function file,functioname> into links
|
||||
} | sed -e 's!<@[[:blank:]]*function \([^,>]*\)[[:blank:]]*>!<a href="#\1">\1</a>!g' \
|
||||
-e 's!<@[[:blank:]]*function \([^,>]*\),[[:blank:]]*\([^>]*\)[[:blank:]]*>!<a href="\1#\2">\1</a>!g' >> $OUT_FILE
|
||||
#Close off the html for this src
|
||||
cat <<- EOF >> $OUT_FILE
|
||||
</body>
|
||||
</html>
|
||||
EOF
|
||||
|
||||
done #Go on to next src
|
||||
|
||||
#Now for tying the scripts all together
|
||||
pushd $OUT_DIR >/dev/null
|
||||
|
||||
print_info "Writing function list"
|
||||
# Start page that will have all the function calls
|
||||
cat <<- EOF > function_list.html
|
||||
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
|
||||
<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en">
|
||||
<head>
|
||||
$HEADERS
|
||||
<title>Functions of $PROJECT</title>
|
||||
</head>
|
||||
<body class="nav">
|
||||
<ul class="nav">
|
||||
EOF
|
||||
|
||||
echo "<li class=\"nav nav-header\">Functions</li>" >> function_list.html
|
||||
# Merge function lists of all sources, sort by function name
|
||||
for i in *.funcs ; do
|
||||
for f in $( cat $i ) ; do
|
||||
echo "$f <li class=\"nav nav-function\"><a href=\"${i%.funcs}.html#$f\" target=\"main\">$f</a></li>"
|
||||
done
|
||||
done | sort | cut -d' ' -f2- >> function_list.html
|
||||
|
||||
echo "<li class=\"nav nav-header\">Variables</li>" >> function_list.html
|
||||
for i in *.vars ; do
|
||||
for v in $( cat $i ) ; do
|
||||
echo "$v <li class=\"nav nav-variable\"><a href=\"${i%.vars}.html#$v\" target=\"main\">$v</a></li>"
|
||||
done
|
||||
done | sort | cut -d' ' -f2- >> function_list.html
|
||||
|
||||
# Close off the html for the global function list
|
||||
cat <<- EOF >> function_list.html
|
||||
</ul>
|
||||
</body>
|
||||
</html>
|
||||
EOF
|
||||
|
||||
print_info "Writing script list"
|
||||
# Start the list of scripts
|
||||
TITLE="Scripts"
|
||||
[[ $PROJECT ]] && TITLE="$PROJECT Script Documentation"
|
||||
cat <<- EOF > script_list.html
|
||||
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
|
||||
<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en">
|
||||
<head>
|
||||
$HEADERS
|
||||
<title>Scripts of $PROJECT</title>
|
||||
</head>
|
||||
<body>
|
||||
<h1>$TITLE</h1>
|
||||
<hr />
|
||||
<dl>
|
||||
EOF
|
||||
|
||||
# List all the sources + descriptions, sort by script dir/name
|
||||
for i in *.name ; do
|
||||
name=${i%.name}
|
||||
echo "${name} $(cat "$i")"
|
||||
done | sort | while read LINE realname; do
|
||||
echo "<dt><a href=\"${LINE}.html\">$realname</a></dt>"
|
||||
echo "<dd>"
|
||||
cat ${LINE}.desc 2>/dev/null || { [[ $QUIET -lt 2 ]] && print_warn "$LINE has no description."; }
|
||||
echo "</dd>"
|
||||
done >> script_list.html
|
||||
|
||||
# Close off the html for the global script list
|
||||
cat <<- EOF >> script_list.html
|
||||
</dl>
|
||||
</body>
|
||||
</html>
|
||||
EOF
|
||||
|
||||
print_info "Writing index file"
|
||||
# Create the index file for the whole shbang
|
||||
cat <<- EOF > index.html
|
||||
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Frameset//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-frameset.dtd">
|
||||
<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en">
|
||||
<head>
|
||||
$HEADERS
|
||||
<title>BashDoc - $PROJECT</title>
|
||||
</head>
|
||||
<frameset cols="25%,*">
|
||||
<frame src="function_list.html" name="function_list" />
|
||||
<frame src="script_list.html" name="main" />
|
||||
</frameset>
|
||||
</html>
|
||||
EOF
|
||||
|
||||
# Remove the temporary .desc and .name files, leave the .func and .vars files, someone may want them later.
|
||||
rm *.desc
|
||||
rm *.name
|
||||
popd >/dev/null
|
||||
Reference in New Issue
Block a user