misc/libfreetype/src/tools/docmaker/formatter.py
changeset 9372 915436ff64ab
parent 9371 f3840de881bd
child 9373 b769a8e38cbd
--- a/misc/libfreetype/src/tools/docmaker/formatter.py	Thu Jul 25 23:16:06 2013 +0200
+++ /dev/null	Thu Jan 01 00:00:00 1970 +0000
@@ -1,188 +0,0 @@
-#  Formatter (c) 2002, 2004, 2007, 2008 David Turner <david@freetype.org>
-#
-
-from sources import *
-from content import *
-from utils   import *
-
-# This is the base Formatter class.  Its purpose is to convert
-# a content processor's data into specific documents (i.e., table of
-# contents, global index, and individual API reference indices).
-#
-# You need to sub-class it to output anything sensible.  For example,
-# the file tohtml.py contains the definition of the HtmlFormatter sub-class
-# used to output -- you guessed it -- HTML.
-#
-
-class  Formatter:
-
-    def  __init__( self, processor ):
-        self.processor   = processor
-        self.identifiers = {}
-        self.chapters    = processor.chapters
-        self.sections    = processor.sections.values()
-        self.block_index = []
-
-        # store all blocks in a dictionary
-        self.blocks = []
-        for section in self.sections:
-            for block in section.blocks.values():
-                self.add_identifier( block.name, block )
-
-                # add enumeration values to the index, since this is useful
-                for markup in block.markups:
-                    if markup.tag == 'values':
-                        for field in markup.fields:
-                            self.add_identifier( field.name, block )
-
-        self.block_index = self.identifiers.keys()
-        self.block_index.sort( index_sort )
-
-    def  add_identifier( self, name, block ):
-        if self.identifiers.has_key( name ):
-            # duplicate name!
-            sys.stderr.write(                                           \
-               "WARNING: duplicate definition for '" + name + "' in " + \
-               block.location() + ", previous definition in " +         \
-               self.identifiers[name].location() + "\n" )
-        else:
-            self.identifiers[name] = block
-
-    #
-    #  Formatting the table of contents
-    #
-    def  toc_enter( self ):
-        pass
-
-    def  toc_chapter_enter( self, chapter ):
-        pass
-
-    def  toc_section_enter( self, section ):
-        pass
-
-    def  toc_section_exit( self, section ):
-        pass
-
-    def  toc_chapter_exit( self, chapter ):
-        pass
-
-    def  toc_index( self, index_filename ):
-        pass
-
-    def  toc_exit( self ):
-        pass
-
-    def  toc_dump( self, toc_filename = None, index_filename = None ):
-        output = None
-        if toc_filename:
-            output = open_output( toc_filename )
-
-        self.toc_enter()
-
-        for chap in self.processor.chapters:
-
-            self.toc_chapter_enter( chap )
-
-            for section in chap.sections:
-                self.toc_section_enter( section )
-                self.toc_section_exit( section )
-
-            self.toc_chapter_exit( chap )
-
-        self.toc_index( index_filename )
-
-        self.toc_exit()
-
-        if output:
-            close_output( output )
-
-    #
-    #  Formatting the index
-    #
-    def  index_enter( self ):
-        pass
-
-    def  index_name_enter( self, name ):
-        pass
-
-    def  index_name_exit( self, name ):
-        pass
-
-    def  index_exit( self ):
-        pass
-
-    def  index_dump( self, index_filename = None ):
-        output = None
-        if index_filename:
-            output = open_output( index_filename )
-
-        self.index_enter()
-
-        for name in self.block_index:
-            self.index_name_enter( name )
-            self.index_name_exit( name )
-
-        self.index_exit()
-
-        if output:
-            close_output( output )
-
-    #
-    #  Formatting a section
-    #
-    def  section_enter( self, section ):
-        pass
-
-    def  block_enter( self, block ):
-        pass
-
-    def  markup_enter( self, markup, block = None ):
-        pass
-
-    def  field_enter( self, field, markup = None, block = None ):
-        pass
-
-    def  field_exit( self, field, markup = None, block = None ):
-        pass
-
-    def  markup_exit( self, markup, block = None ):
-        pass
-
-    def  block_exit( self, block ):
-        pass
-
-    def  section_exit( self, section ):
-        pass
-
-    def  section_dump( self, section, section_filename = None ):
-        output = None
-        if section_filename:
-            output = open_output( section_filename )
-
-        self.section_enter( section )
-
-        for name in section.block_names:
-            block = self.identifiers[name]
-            self.block_enter( block )
-
-            for markup in block.markups[1:]:   # always ignore first markup!
-                self.markup_enter( markup, block )
-
-                for field in markup.fields:
-                    self.field_enter( field, markup, block )
-                    self.field_exit( field, markup, block )
-
-                self.markup_exit( markup, block )
-
-            self.block_exit( block )
-
-        self.section_exit( section )
-
-        if output:
-            close_output( output )
-
-    def  section_dump_all( self ):
-        for section in self.sections:
-            self.section_dump( section )
-
-# eof