ruby--ruby/lib/rdoc/parsers/parse_c.rb

336 lines
10 KiB
Ruby
Raw Normal View History

# We attempt to parse C extension files. Basically we look for
# the standard patterns that you find in extensions: <tt>rb_define_class,
# rb_define_method</tt> and so on. We also try to find the corresponding
# C source for the methods and extract comments, but if we fail
# we don't worry too much.
#
# The comments associated with a Ruby method are extracted from the C
# comment block associated with the routine that _implements_ that
# method, that is to say the method whose name is given in the
# <tt>rb_define_method</tt> call. For example, you might write:
#
# /*
# * Returns a new array that is a one-dimensional flattening of this
# * array (recursively). That is, for every element that is an array,
# * extract its elements into the new array.
# *
# * s = [ 1, 2, 3 ] #=> [1, 2, 3]
# * t = [ 4, 5, 6, [7, 8] ] #=> [4, 5, 6, [7, 8]]
# * a = [ s, t, 9, 10 ] #=> [[1, 2, 3], [4, 5, 6, [7, 8]], 9, 10]
# * a.flatten #=> [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
# */
# static VALUE
# rb_ary_flatten(ary)
# VALUE ary;
# {
# ary = rb_obj_dup(ary);
# rb_ary_flatten_bang(ary);
# return ary;
# }
#
# ...
#
# void
# Init_Array()
# {
# ...
# rb_define_method(rb_cArray, "flatten", rb_ary_flatten, 0);
#
# Here RDoc will determine from the rb_define_method line that there's a
# method called "flatten" in class Array, and will look for the implementation
# in the method rb_ary_flatten. It will then use the comment from that
# method in the HTML output. This method must be in the same source file
# as the rb_define_method.
#
# C classes can be diagramed (see /tc/dl/ruby/ruby/error.c), and RDoc
# integrates C and Ruby source into one tree
# Classes and modules built in to the interpreter. We need
# these to define superclasses of user objects
require "rdoc/code_objects"
require "rdoc/parsers/parserfactory"
module RDoc
KNOWN_CLASSES = {
"rb_cObject" => "Object",
"rb_cArray" => "Array",
"rb_cBignum" => "Bignum",
"rb_cClass" => "Class",
"rb_cDir" => "Dir",
"rb_cData" => "Data",
"rb_cFalseClass" => "FalseClass",
"rb_cFile" => "File",
"rb_cFixnum" => "Fixnum",
"rb_cFloat" => "Float",
"rb_cHash" => "Hash",
"rb_cInteger" => "Integer",
"rb_cIO" => "IO",
"rb_cModule" => "Module",
"rb_cNilClass" => "NilClass",
"rb_cNumeric" => "Numeric",
"rb_cProc" => "Proc",
"rb_cRange" => "Range",
"rb_cRegexp" => "Regexp",
"rb_cString" => "String",
"rb_cSymbol" => "Symbol",
"rb_cThread" => "Thread",
"rb_cTime" => "Time",
"rb_cTrueClass" => "TrueClass",
"rb_cStruct" => "Struct",
"rb_eException" => "Exception",
"rb_eStandardError" => "StandardError",
"rb_eSystemExit" => "SystemExit",
"rb_eInterrupt" => "Interrupt",
"rb_eSignal" => "Signal",
"rb_eFatal" => "Fatal",
"rb_eArgError" => "ArgError",
"rb_eEOFError" => "EOFError",
"rb_eIndexError" => "IndexError",
"rb_eRangeError" => "RangeError",
"rb_eIOError" => "IOError",
"rb_eRuntimeError" => "RuntimeError",
"rb_eSecurityError" => "SecurityError",
"rb_eSystemCallError" => "SystemCallError",
"rb_eTypeError" => "TypeError",
"rb_eZeroDivError" => "ZeroDivError",
"rb_eNotImpError" => "NotImpError",
"rb_eNoMemError" => "NoMemError",
"rb_eFloatDomainError" => "FloatDomainError",
"rb_eScriptError" => "ScriptError",
"rb_eNameError" => "NameError",
"rb_eSyntaxError" => "SyntaxError",
"rb_eLoadError" => "LoadError",
"rb_mKernel" => "Kernel",
"rb_mComparable" => "Comparable",
"rb_mEnumerable" => "Enumerable",
"rb_mPrecision" => "Precision",
"rb_mErrno" => "Errno",
"rb_mFileTest" => "FileTest",
"rb_mGC" => "GC",
"rb_mMath" => "Math",
"rb_mProcess" => "Process"
}
# See rdoc/c_parse.rb
class C_Parser
extend ParserFactory
parse_files_matching(/\.(c|cc|cpp|CC)$/)
# prepare to parse a C file
def initialize(top_level, file_name, body, options)
@known_classes = KNOWN_CLASSES.dup
@body = body
@options = options
@top_level = top_level
@classes = Hash.new
end
# Extract the classes/modules and methods from a C file
# and return the corresponding top-level object
def scan
remove_commented_out_lines
do_classes
do_methods
do_includes
@top_level
end
#######
private
#######
# remove lines that are commented out that might otherwise get
# picked up when scanning for classes and methods
def remove_commented_out_lines
@body.gsub!(%r{//.*rb_define_}, '//')
end
def handle_class_module(var_name, class_mod, class_name, parent, in_module)
@known_classes[var_name] = class_name
parent_name = @known_classes[parent] || parent
if in_module
enclosure = @classes[in_module]
unless enclosure
$stderr.puts("Enclosing class/module '#{in_module}' for " +
"#{class_mod} #{class_name} not known")
return
end
else
enclosure = @top_level
end
if class_mod == "class"
cm = enclosure.add_class(NormalClass, class_name, parent_name)
else
cm = enclosure.add_module(NormalModule, class_name)
end
cm.record_location(enclosure.toplevel)
find_class_comment(class_name, cm)
@classes[var_name] = cm
end
def find_class_comment(class_name, class_meth)
if @body =~ %r{((?>/\*.*?\*/\s+))
(static\s+)?void\s+Init_#{class_name}\s*\(\)}xm
class_meth.comment = mangle_comment($1)
end
end
def do_classes
@body.scan(/(\w+)\s* = \s*rb_define_module\(\s*"(\w+)"\s*\)/mx) do
|var_name, class_name|
handle_class_module(var_name, "module", class_name, nil, nil)
end
@body.scan(/(\w+)\s* = \s*rb_define_module_under
\(
\s*(\w+),
\s*"(\w+)"
\)/mx) do
|var_name, in_module, class_name|
handle_class_module(var_name, "module", class_name, nil, in_module)
end
@body.scan(/(\w+)\s* = \s*rb_define_class
\(
\s*"(\w+)",
\s*(\w+)\s*
\)/mx) do
|var_name, class_name, parent|
handle_class_module(var_name, "class", class_name, parent, nil)
end
@body.scan(/(\w+)\s* = \s*rb_define_class_under
\(
\s*(\w+),
\s*"(\w+)",
\s*(\w+)\s*
\)/mx) do
|var_name, in_module, class_name, parent|
handle_class_module(var_name, "class", class_name, parent, in_module)
end
end
def do_methods
@body.scan(/rb_define_(singleton_method|method|module_function)\(\s*(\w+),
\s*"([^"]+)",
\s*(?:RUBY_METHOD_FUNC\(|VALUEFUNC\()?(\w+)\)?,
\s*(-?\w+)\s*\)/xm) do #"
|type, var_name, meth_name, meth_body, param_count|
next if meth_name == "initialize_copy"
class_name = @known_classes[var_name] || var_name
class_obj = find_class(var_name, class_name)
if class_obj
if meth_name == "initialize"
meth_name = "new"
type = "singleton_method"
end
meth_obj = AnyMethod.new("", meth_name)
meth_obj.singleton = type == "singleton_method"
p_count = (Integer(param_count) rescue -1)
if p_count < 0
meth_obj.params = "(...)"
elsif p_count == 0
meth_obj.params = "()"
else
meth_obj.params = "(" +
(1..p_count).map{|i| "p#{i}"}.join(", ") +
")"
end
find_body(meth_body, meth_obj)
class_obj.add_method(meth_obj)
end
end
end
# Find the C code corresponding to a c method
def find_body(meth_name, meth_obj)
if @body =~ %r{((?>/\*.*?\*/\s+))(static\s+)?VALUE\s+#{meth_name}
\s*(\(.*?\)).*?^}xm
comment, params = $1, $3
body_text = $&
# see if we can find the whole body
re = Regexp.escape(body_text) + "[^(]*^{.*?^}"
if Regexp.new(re, Regexp::MULTILINE).match(@body)
body_text = $&
end
# If the comment block contains a section that looks like
# call-seq:
# Array.new
# Array.new(10)
# use it for the parameters
if comment.sub!(/call-seq:(.*?)^\s*\*?\s*$/m, '')
seq = $1
seq.gsub!(/^\s*\*\s*/, '')
meth_obj.params = "!verb!" + seq
end
# meth_obj.params = params
meth_obj.start_collecting_tokens
meth_obj.add_token(RubyToken::Token.new(1,1).set_text(body_text))
meth_obj.comment = mangle_comment(comment)
end
end
# Look for includes of the form
# rb_include_module(rb_cArray, rb_mEnumerable);
def do_includes
@body.scan(/rb_include_module\(\s*(\w+?),\s*(\w+?)\s*\)/) do |c,m|
if cls = @classes[c]
m = KNOWN_CLASSES[m] || m
cls.add_include(Include.new(m, ""))
end
end
end
# Remove the /*'s and leading asterisks from C comments
def mangle_comment(comment)
comment.sub!(%r{/\*+}) { " " * $&.length }
comment.sub!(%r{\*+/}) { " " * $&.length }
comment.gsub!(/^[ \t]*\*/m) { " " * $&.length }
comment
end
def find_class(raw_name, name)
unless @classes[name]
if raw_name =~ /^rb_m/
@classes[name] = @top_level.add_module(NormalModule, name)
else
@classes[name] = @top_level.add_class(NormalClass, name, nil)
end
end
@classes[name]
end
end
end