mirror of
https://github.com/wxWidgets/Phoenix.git
synced 2026-01-07 04:20:07 +01:00
* Changed imports to use either absolute or explicit relative imports. Implicit relative imports are no longer allowed. * Changes to accomodate standard library classes or modues moving to other locations, or being removed entirely. * Changes related to print becoming a function, execfile being removed, u'' no longer allowed, and other syntax related issues. * Working around C APIs that have changed or simply vanished. (PyInt, PyString, PyBytes, etc.) * Dealing with text file objects using strings vs binary file objects using bytes, auto-encoding, and etc. * Replacing the use of PyCObject with PyCapsule and dealing with an apparent bug where PyCapsule objects can't be imported from submodules within a package. git-svn-id: https://svn.wxwidgets.org/svn/wx/wxPython/Phoenix/trunk@71554 c3d73ce0-8a6f-49c7-b76d-6d57e0e08775
117 lines
3.3 KiB
Python
117 lines
3.3 KiB
Python
#---------------------------------------------------------------------------
|
|
# Name: etgtools/generators.py
|
|
# Author: Robin Dunn
|
|
#
|
|
# Created: 3-Nov-2010
|
|
# Copyright: (c) 2011 by Total Control Software
|
|
# License: wxWindows License
|
|
#---------------------------------------------------------------------------
|
|
|
|
"""
|
|
Just some base classes and stubs for the various generators
|
|
"""
|
|
|
|
import sys
|
|
|
|
class WrapperGeneratorBase(object):
|
|
def __init__(self):
|
|
pass
|
|
def generate(self, module, destFile=None):
|
|
raise NotImplementedError
|
|
|
|
|
|
class DocsGeneratorBase(object):
|
|
def __init__(self):
|
|
pass
|
|
def generate(self, module):
|
|
raise NotImplementedError
|
|
|
|
|
|
class StubbedDocsGenerator(DocsGeneratorBase):
|
|
def generate(self, module):
|
|
pass
|
|
|
|
|
|
class SphinxGenerator(DocsGeneratorBase):
|
|
def generate(self, module):
|
|
pass
|
|
|
|
#---------------------------------------------------------------------------
|
|
# helpers
|
|
|
|
def nci(text, numSpaces=0, stripLeading=True):
|
|
"""
|
|
Normalize Code Indents
|
|
|
|
First use the count of leading spaces on the first line and remove that
|
|
many spaces from the front of all lines, and then indent each line by
|
|
adding numSpaces spaces. This is used so we can convert the arbitrary
|
|
indents that might be used by the tweaker code into what is expected for
|
|
the context we are generating for.
|
|
"""
|
|
def _getLeadingSpaceCount(line):
|
|
count = 0
|
|
for c in line:
|
|
assert c != '\t', "Use spaces for indent, not tabs"
|
|
if c != ' ':
|
|
break
|
|
count += 1
|
|
return count
|
|
|
|
def _allSpaces(text):
|
|
for c in text:
|
|
if c != ' ':
|
|
return False
|
|
return True
|
|
|
|
|
|
lines = text.rstrip().split('\n')
|
|
if stripLeading:
|
|
numStrip = _getLeadingSpaceCount(lines[0])
|
|
else:
|
|
numStrip = 0
|
|
|
|
for idx, line in enumerate(lines):
|
|
assert _allSpaces(line[:numStrip]), "Indentation inconsistent with first line"
|
|
lines[idx] = ' '*numSpaces + line[numStrip:]
|
|
|
|
newText = '\n'.join(lines) + '\n'
|
|
return newText
|
|
|
|
#---------------------------------------------------------------------------
|
|
|
|
if sys.version_info < (3,):
|
|
# For Python 2.x we'll convert any unicode text values to strings before
|
|
# adding them to the buffer
|
|
from StringIO import StringIO
|
|
class Utf8EncodingStream(StringIO):
|
|
def write(self, text):
|
|
if isinstance(text, unicode):
|
|
text = text.encode('utf-8')
|
|
return StringIO.write(self, text)
|
|
|
|
else:
|
|
# For Python 3.x we'll keep it all as str (unicode) objects and let the
|
|
# conversion to bytes happen when the text is written to the actual
|
|
# file.
|
|
from io import StringIO
|
|
class Utf8EncodingStream(StringIO):
|
|
pass
|
|
|
|
|
|
|
|
def textfile_open(filename, mode='rt'):
|
|
"""
|
|
Simple wrapper around open() that will open normally on Python 2.x and on
|
|
Python 2.3 will add the encoding parameter. The mode parameter must
|
|
include the 't' to put the stream into text mode.
|
|
"""
|
|
assert 't' in mode
|
|
if sys.version_info < (3,):
|
|
return open(filename, mode)
|
|
else:
|
|
return open(filename, mode, encoding='utf-8')
|
|
|
|
|
|
#---------------------------------------------------------------------------
|