summaryrefslogtreecommitdiffstats
path: root/source-builder/sb/asciidoc/filters/code/code-filter.py
diff options
context:
space:
mode:
authorChris Johns <chrisj@rtems.org>2017-08-07 09:59:49 +1000
committerChris Johns <chrisj@rtems.org>2017-08-07 09:59:49 +1000
commit8b96e17c8abf61d97dd224b23370dc148f32fe3c (patch)
treee8eb043159d145ffbbbf9c23ef872226de5ab059 /source-builder/sb/asciidoc/filters/code/code-filter.py
parent4.12: Fix SIS patch URL (diff)
downloadrtems-source-builder-8b96e17c8abf61d97dd224b23370dc148f32fe3c.tar.bz2
doc: Remove in source documentation and the asciidoc package
The RSB documentation is now in ReST format and part of the RTEMS Documentation project. See https://docs.rtems.org/. Remove support for the GPL based asciidoc tool and remove the asciidoc package from the RSB. Add the Python Markdown package and update the reporter to use Markdown for HTML generation. The resuling HTML report is a single self contained file. Closes #3047.
Diffstat (limited to 'source-builder/sb/asciidoc/filters/code/code-filter.py')
-rwxr-xr-xsource-builder/sb/asciidoc/filters/code/code-filter.py239
1 files changed, 0 insertions, 239 deletions
diff --git a/source-builder/sb/asciidoc/filters/code/code-filter.py b/source-builder/sb/asciidoc/filters/code/code-filter.py
deleted file mode 100755
index 473fd6e..0000000
--- a/source-builder/sb/asciidoc/filters/code/code-filter.py
+++ /dev/null
@@ -1,239 +0,0 @@
-#!/usr/bin/env python
-'''
-NAME
- code-filter - AsciiDoc filter to highlight language keywords
-
-SYNOPSIS
- code-filter -b backend -l language [ -t tabsize ]
- [ --help | -h ] [ --version | -v ]
-
-DESCRIPTION
- This filter reads source code from the standard input, highlights language
- keywords and comments and writes to the standard output.
-
- The purpose of this program is to demonstrate how to write an AsciiDoc
- filter -- it's much to simplistic to be passed off as a code syntax
- highlighter. Use the 'source-highlight-filter' instead.
-
-
-OPTIONS
- --help, -h
- Print this documentation.
-
- -b
- Backend output file format: 'docbook', 'linuxdoc', 'html', 'css'.
-
- -l
- The name of the source code language: 'python', 'ruby', 'c++', 'c'.
-
- -t tabsize
- Expand source tabs to tabsize spaces.
-
- --version, -v
- Print program version number.
-
-BUGS
- - Code on the same line as a block comment is treated as comment.
- Keywords inside literal strings are highlighted.
- - There doesn't appear to be an easy way to accomodate linuxdoc so
- just pass it through without markup.
-
-AUTHOR
- Written by Stuart Rackham, <srackham@gmail.com>
-
-URLS
- http://sourceforge.net/projects/asciidoc/
- http://www.methods.co.nz/asciidoc/
-
-COPYING
- Copyright (C) 2002-2006 Stuart Rackham. Free use of this software is
- granted under the terms of the GNU General Public License (GPL).
-'''
-
-import os, sys, re, string
-
-VERSION = '1.1.2'
-
-# Globals.
-language = None
-backend = None
-tabsize = 8
-keywordtags = {
- 'html':
- ('<strong>','</strong>'),
- 'css':
- ('<strong>','</strong>'),
- 'docbook':
- ('<emphasis role="strong">','</emphasis>'),
- 'linuxdoc':
- ('','')
-}
-commenttags = {
- 'html':
- ('<i>','</i>'),
- 'css':
- ('<i>','</i>'),
- 'docbook':
- ('<emphasis>','</emphasis>'),
- 'linuxdoc':
- ('','')
-}
-keywords = {
- 'python':
- ('and', 'del', 'for', 'is', 'raise', 'assert', 'elif', 'from',
- 'lambda', 'return', 'break', 'else', 'global', 'not', 'try', 'class',
- 'except', 'if', 'or', 'while', 'continue', 'exec', 'import', 'pass',
- 'yield', 'def', 'finally', 'in', 'print'),
- 'ruby':
- ('__FILE__', 'and', 'def', 'end', 'in', 'or', 'self', 'unless',
- '__LINE__', 'begin', 'defined?' 'ensure', 'module', 'redo', 'super',
- 'until', 'BEGIN', 'break', 'do', 'false', 'next', 'rescue', 'then',
- 'when', 'END', 'case', 'else', 'for', 'nil', 'retry', 'true', 'while',
- 'alias', 'class', 'elsif', 'if', 'not', 'return', 'undef', 'yield'),
- 'c++':
- ('asm', 'auto', 'bool', 'break', 'case', 'catch', 'char', 'class',
- 'const', 'const_cast', 'continue', 'default', 'delete', 'do', 'double',
- 'dynamic_cast', 'else', 'enum', 'explicit', 'export', 'extern',
- 'false', 'float', 'for', 'friend', 'goto', 'if', 'inline', 'int',
- 'long', 'mutable', 'namespace', 'new', 'operator', 'private',
- 'protected', 'public', 'register', 'reinterpret_cast', 'return',
- 'short', 'signed', 'sizeof', 'static', 'static_cast', 'struct',
- 'switch', 'template', 'this', 'throw', 'true', 'try', 'typedef',
- 'typeid', 'typename', 'union', 'unsigned', 'using', 'virtual', 'void',
- 'volatile', 'wchar_t', 'while')
-}
-block_comments = {
- 'python': ("'''","'''"),
- 'ruby': None,
- 'c++': ('/*','*/')
-}
-inline_comments = {
- 'python': '#',
- 'ruby': '#',
- 'c++': '//'
-}
-
-def print_stderr(line):
- sys.stderr.write(line+os.linesep)
-
-def sub_keyword(mo):
- '''re.subs() argument to tag keywords.'''
- word = mo.group('word')
- if word in keywords[language]:
- stag,etag = keywordtags[backend]
- return stag+word+etag
- else:
- return word
-
-def code_filter():
- '''This function does all the work.'''
- global language, backend
- inline_comment = inline_comments[language]
- blk_comment = block_comments[language]
- if blk_comment:
- blk_comment = (re.escape(block_comments[language][0]),
- re.escape(block_comments[language][1]))
- stag,etag = commenttags[backend]
- in_comment = 0 # True if we're inside a multi-line block comment.
- tag_comment = 0 # True if we should tag the current line as a comment.
- line = sys.stdin.readline()
- while line:
- line = string.rstrip(line)
- line = string.expandtabs(line,tabsize)
- # Escape special characters.
- line = string.replace(line,'&','&amp;')
- line = string.replace(line,'<','&lt;')
- line = string.replace(line,'>','&gt;')
- # Process block comment.
- if blk_comment:
- if in_comment:
- if re.match(r'.*'+blk_comment[1]+r'$',line):
- in_comment = 0
- else:
- if re.match(r'^\s*'+blk_comment[0]+r'.*'+blk_comment[1],line):
- # Single line block comment.
- tag_comment = 1
- elif re.match(r'^\s*'+blk_comment[0],line):
- # Start of multi-line block comment.
- tag_comment = 1
- in_comment = 1
- else:
- tag_comment = 0
- if tag_comment:
- if line: line = stag+line+etag
- else:
- if inline_comment:
- pos = string.find(line,inline_comment)
- else:
- pos = -1
- if pos >= 0:
- # Process inline comment.
- line = re.sub(r'\b(?P<word>\w+)\b',sub_keyword,line[:pos]) \
- + stag + line[pos:] + etag
- else:
- line = re.sub(r'\b(?P<word>\w+)\b',sub_keyword,line)
- sys.stdout.write(line + os.linesep)
- line = sys.stdin.readline()
-
-def usage(msg=''):
- if msg:
- print_stderr(msg)
- print_stderr('Usage: code-filter -b backend -l language [ -t tabsize ]')
- print_stderr(' [ --help | -h ] [ --version | -v ]')
-
-def main():
- global language, backend, tabsize
- # Process command line options.
- import getopt
- opts,args = getopt.getopt(sys.argv[1:],
- 'b:l:ht:v',
- ['help','version'])
- if len(args) > 0:
- usage()
- sys.exit(1)
- for o,v in opts:
- if o in ('--help','-h'):
- print __doc__
- sys.exit(0)
- if o in ('--version','-v'):
- print('code-filter version %s' % (VERSION,))
- sys.exit(0)
- if o == '-b': backend = v
- if o == '-l':
- v = string.lower(v)
- if v == 'c': v = 'c++'
- language = v
- if o == '-t':
- try:
- tabsize = int(v)
- except:
- usage('illegal tabsize')
- sys.exit(1)
- if tabsize <= 0:
- usage('illegal tabsize')
- sys.exit(1)
- if backend is None:
- usage('backend option is mandatory')
- sys.exit(1)
- if not keywordtags.has_key(backend):
- usage('illegal backend option')
- sys.exit(1)
- if language is None:
- usage('language option is mandatory')
- sys.exit(1)
- if not keywords.has_key(language):
- usage('illegal language option')
- sys.exit(1)
- # Do the work.
- code_filter()
-
-if __name__ == "__main__":
- try:
- main()
- except (KeyboardInterrupt, SystemExit):
- pass
- except:
- print_stderr("%s: unexpected exit status: %s" %
- (os.path.basename(sys.argv[0]), sys.exc_info()[1]))
- # Exit with previous sys.exit() status or zero if no sys.exit().
- sys.exit(sys.exc_info()[1])