Sun, 21 Dec 2014 15:36:59 +0100
Added capability to use 'Sphinx' to preview ReST files.
# -*- coding: utf-8 -*- # Copyright (c) 2014 Detlev Offenbach <detlev@die-offenbachs.de> # """ Module implementing a previewer widget for HTML, Markdown and ReST files. """ from __future__ import unicode_literals try: # Only for Py2 import StringIO as io # __IGNORE_EXCEPTION__ except ImportError: import io # __IGNORE_WARNING__ import os import threading import re import shutil import tempfile import sys from PyQt5.QtCore import pyqtSlot, pyqtSignal, Qt, QUrl, QSize, QThread from PyQt5.QtWidgets import QWidget from PyQt5.QtWebKitWidgets import QWebPage from E5Gui.E5Application import e5App from .Ui_PreviewerHTML import Ui_PreviewerHTML import Utilities import Preferences class PreviewerHTML(QWidget, Ui_PreviewerHTML): """ Class implementing a previewer widget for HTML, Markdown and ReST files. """ def __init__(self, parent=None): """ Constructor @param parent reference to the parent widget (QWidget) """ super(PreviewerHTML, self).__init__(parent) self.setupUi(self) self.jsCheckBox.setChecked( Preferences.getUI("ShowFilePreviewJS")) self.ssiCheckBox.setChecked( Preferences.getUI("ShowFilePreviewSSI")) self.previewView.page().setLinkDelegationPolicy( QWebPage.DelegateAllLinks) self.__scrollBarPositions = {} self.__vScrollBarAtEnd = {} self.__hScrollBarAtEnd = {} self.__processingThread = PreviewProcessingThread() self.__processingThread.htmlReady.connect(self.__setHtml) self.__previewedPath = None self.__previewedEditor = None def shutdown(self): """ Public method to perform shutdown actions. """ self.__processingThread.wait() @pyqtSlot(bool) def on_jsCheckBox_clicked(self, checked): """ Private slot to enable/disable JavaScript. @param checked state of the checkbox (boolean) """ Preferences.setUI("ShowFilePreviewJS", checked) self.__setJavaScriptEnabled(checked) def __setJavaScriptEnabled(self, enable): """ Private method to enable/disable JavaScript. @param enable flag indicating the enable state (boolean) """ self.jsCheckBox.setChecked(enable) settings = self.previewView.settings() settings.setAttribute(settings.JavascriptEnabled, enable) self.processEditor() @pyqtSlot(bool) def on_ssiCheckBox_clicked(self, checked): """ Private slot to enable/disable SSI. @param checked state of the checkbox (boolean) """ Preferences.setUI("ShowFilePreviewSSI", checked) self.processEditor() def processEditor(self, editor=None): """ Public slot to process an editor's text. @param editor editor to be processed (Editor) """ if editor is None: editor = self.__previewedEditor else: self.__previewedEditor = editor if editor is not None: fn = editor.getFileName() if fn: extension = os.path.normcase(os.path.splitext(fn)[1][1:]) else: extension = "" if extension in \ Preferences.getEditor("PreviewHtmlFileNameExtensions") or \ editor.getLanguage() == "HTML": language = "HTML" elif extension in \ Preferences.getEditor("PreviewMarkdownFileNameExtensions"): language = "Markdown" elif extension in \ Preferences.getEditor("PreviewRestFileNameExtensions"): language = "ReST" else: self.__setHtml(fn, self.tr( "<p>No preview available for this type of file.</p>")) return if fn: project = e5App().getObject("Project") if project.isProjectFile(fn): rootPath = project.getProjectPath() else: rootPath = os.path.dirname(os.path.abspath(fn)) else: rootPath = "" self.__processingThread.process( fn, language, editor.text(), self.ssiCheckBox.isChecked(), rootPath, Preferences.getEditor("PreviewRestUseSphinx")) def __setHtml(self, filePath, html): """ Private method to set the HTML to the view and restore the scroll bars positions. @param filePath file path of the previewed editor (string) @param html processed HTML text ready to be shown (string) """ self.__saveScrollBarPositions() self.__previewedPath = Utilities.normcasepath( Utilities.fromNativeSeparators(filePath)) self.previewView.page().mainFrame().contentsSizeChanged.connect( self.__restoreScrollBarPositions) self.previewView.setHtml(html, baseUrl=QUrl.fromLocalFile(filePath)) @pyqtSlot(str) def on_previewView_titleChanged(self, title): """ Private slot to handle a change of the title. @param title new title (string) """ if title: self.titleLabel.setText(self.tr("Preview - {0}").format(title)) else: self.titleLabel.setText(self.tr("Preview")) def __saveScrollBarPositions(self): """ Private method to save scroll bar positions for a previewed editor. """ frame = self.previewView.page().mainFrame() if frame.contentsSize() == QSize(0, 0): return # no valid data, nothing to save pos = frame.scrollPosition() self.__scrollBarPositions[self.__previewedPath] = pos self.__hScrollBarAtEnd[self.__previewedPath] = \ frame.scrollBarMaximum(Qt.Horizontal) == pos.x() self.__vScrollBarAtEnd[self.__previewedPath] = \ frame.scrollBarMaximum(Qt.Vertical) == pos.y() def __restoreScrollBarPositions(self): """ Private method to restore scroll bar positions for a previewed editor. """ try: self.previewView.page().mainFrame().contentsSizeChanged.disconnect( self.__restoreScrollBarPositions) except TypeError: # not connected, simply ignore it pass if self.__previewedPath not in self.__scrollBarPositions: return frame = self.previewView.page().mainFrame() frame.setScrollPosition( self.__scrollBarPositions[self.__previewedPath]) if self.__hScrollBarAtEnd[self.__previewedPath]: frame.setScrollBarValue( Qt.Horizontal, frame.scrollBarMaximum(Qt.Horizontal)) if self.__vScrollBarAtEnd[self.__previewedPath]: frame.setScrollBarValue( Qt.Vertical, frame.scrollBarMaximum(Qt.Vertical)) @pyqtSlot(QUrl) def on_previewView_linkClicked(self, url): """ Private slot handling the clicking of a link. @param url url of the clicked link (QUrl) """ e5App().getObject("UserInterface").launchHelpViewer(url.toString()) class PreviewProcessingThread(QThread): """ Class implementing a thread to process some text into HTML usable by the previewer view. @signal htmlReady(str,str) emitted with the file name and processed HTML to signal the availability of the processed HTML """ htmlReady = pyqtSignal(str, str) DefaultStaticPath = os.path.join( os.path.abspath(os.path.dirname(__file__)), 'sphinx_default') StaticRegexp = re.compile(r'(src|href)=["\']_static([\s\w/\.]+?)["\']', re.IGNORECASE) def __init__(self, parent=None): """ Constructor @param parent reference to the parent object (QObject) """ super(PreviewProcessingThread, self).__init__() self.__lock = threading.Lock() def process(self, filePath, language, text, ssiEnabled, rootPath, useSphinx): """ Public method to convert the given text to HTML. @param filePath file path of the text (string) @param language language of the text (string) @param text text to be processed (string) @param ssiEnabled flag indicating to do some (limited) SSI processing (boolean) @param rootPath root path to be used for SSI processing (str) @param useSphinx flag indicating to use Sphinx to generate the ReST preview (boolean) """ with self.__lock: self.__filePath = filePath self.__language = language self.__text = text self.__ssiEnabled = ssiEnabled self.__rootPath = rootPath self.__haveData = True self.__useSphinx = useSphinx if not self.isRunning(): self.start(QThread.LowPriority) def run(self): """ Public thread method to convert the stored data. """ while True: # exits with break with self.__lock: filePath = self.__filePath language = self.__language text = self.__text ssiEnabled = self.__ssiEnabled rootPath = self.__rootPath useSphinx = self.__useSphinx self.__haveData = False html = self.__getHtml(language, text, ssiEnabled, filePath, rootPath, useSphinx) with self.__lock: if not self.__haveData: self.htmlReady.emit(filePath, html) break # else - next iteration def __getHtml(self, language, text, ssiEnabled, filePath, rootPath, useSphinx): """ Private method to process the given text depending upon the given language. @param language language of the text (string) @param text to be processed (string) @param ssiEnabled flag indicating to do some (limited) SSI processing (boolean) @param filePath file path of the text (string) @param rootPath root path to be used for SSI processing (str) @param useSphinx flag indicating to use Sphinx to generate the ReST preview (boolean) @return processed HTML text (string) """ if language == "HTML": if ssiEnabled: return self.__processSSI(text, filePath, rootPath) else: return text elif language == "Markdown": return self.__convertMarkdown(text) elif language == "ReST": return self.__convertReST(text, useSphinx) else: return self.tr( "<p>No preview available for this type of file.</p>") def __processSSI(self, txt, filename, root): """ Private method to process the given text for SSI statements. Note: Only a limited subset of SSI statements are supported. @param txt text to be processed (string) @param filename name of the file associated with the given text (string) @param root directory of the document root (string) @return processed HTML (string) """ if not filename: return txt # SSI include incRe = re.compile( r"""<!--#include[ \t]+(virtual|file)=[\"']([^\"']+)[\"']\s*-->""", re.IGNORECASE) baseDir = os.path.dirname(os.path.abspath(filename)) docRoot = root if root != "" else baseDir while True: incMatch = incRe.search(txt) if incMatch is None: break if incMatch.group(1) == "virtual": incFile = Utilities.normjoinpath(docRoot, incMatch.group(2)) elif incMatch.group(1) == "file": incFile = Utilities.normjoinpath(baseDir, incMatch.group(2)) else: incFile = "" if os.path.exists(incFile): try: f = open(incFile, "r") incTxt = f.read() f.close() except (IOError, OSError): # remove SSI include incTxt = "" else: # remove SSI include incTxt = "" txt = txt[:incMatch.start(0)] + incTxt + txt[incMatch.end(0):] return txt def __convertReST(self, text, useSphinx): """ Private method to convert ReST text into HTML. @param text text to be processed (string) @param useSphinx flag indicating to use Sphinx to generate the ReST preview (boolean) @return processed HTML (string) """ if useSphinx: return self.__convertReSTSphinx(text) else: return self.__convertReSTDocutils(text) def __convertReSTSphinx(self, text): """ Private method to convert ReST text into HTML using 'sphinx'. @param text text to be processed (string) @return processed HTML (string) """ try: from sphinx.application import Sphinx # __IGNORE_EXCEPTION__ except ImportError: return self.tr( """<p>ReStructuredText preview requires the""" """ <b>sphinx</b> package.<br/>Install it with""" """ your package manager,'pip install Sphinx' or see""" """ <a href="http://pypi.python.org/pypi/Sphinx">""" """this page.</a></p>""" """<p>Alternatively you may disable Sphinx usage""" """ on the Editor, Filehandling configuration page.</p>""") tempDir = tempfile.mkdtemp(prefix='eric-rest-') try: filename = 'sphinx_preview' basePath = os.path.join(tempDir, filename) fh = open(basePath + '.rst', 'w', encoding='utf-8') fh.write(text) fh.close() overrides = {'html_add_permalinks': False, 'html_copy_source': False, 'html_title': 'Sphinx preview', 'html_use_index': False, 'html_use_modindex': False, 'html_use_smartypants': True, 'master_doc': filename } app = Sphinx(srcdir=tempDir, confdir=None, outdir=tempDir, doctreedir=tempDir, buildername='html', confoverrides=overrides, status=None, warning=io.StringIO()) app.build(force_all=True, filenames=None) fh = open(basePath + '.html', 'r', encoding='utf-8') html = fh.read() fh.close() finally: shutil.rmtree(tempDir) # Replace the "_static/..." references inserted by Sphinx with absolute # links to the specified DefaultStaticPath replacement. def replace(m): return '{0}="file://{1}{2}"'.format( m.group(1), self.DefaultStaticPath, m.group(2)) html = re.sub(self.StaticRegexp, replace, html) return html def __convertReSTDocutils(self, text): """ Private method to convert ReST text into HTML using 'docutils'. @param text text to be processed (string) @return processed HTML (string) """ if 'sphinx' in sys.modules: # Make sure any Sphinx polution of docutils has been removed. unloadKeys = [k for k in sys.modules.keys() if k.startswith(('docutils', 'sphinx'))] for key in unloadKeys: sys.modules.pop(key) try: import docutils.core # __IGNORE_EXCEPTION__ except ImportError: return self.tr( """<p>ReStructuredText preview requires the""" """ <b>python-docutils</b> package.<br/>Install it with""" """ your package manager, 'pip install docutils' or see""" """ <a href="http://pypi.python.org/pypi/docutils">""" """this page.</a></p>""") return docutils.core.publish_string(text, writer_name='html')\ .decode("utf-8") def __convertMarkdown(self, text): """ Private method to convert Markdown text into HTML. @param text text to be processed (string) @return processed HTML (string) """ try: import markdown # __IGNORE_EXCEPTION__ except ImportError: return self.tr( """<p>Markdown preview requires the <b>python-markdown</b> """ """package.<br/>Install it with your package manager or see """ """<a href="http://pythonhosted.org/Markdown/install.html">""" """installation instructions.</a></p>""") try: import mdx_mathjax # __IGNORE_EXCEPTION__ __IGNORE_WARNING__ except ImportError: #mathjax doesn't require import statement if installed as extension pass extensions = ['fenced_code', 'nl2br', 'extra'] # version 2.0 supports only extension names, not instances if markdown.version_info[0] > 2 or \ (markdown.version_info[0] == 2 and markdown.version_info[1] > 0): class _StrikeThroughExtension(markdown.Extension): """ Class is placed here, because it depends on imported markdown, and markdown import is lazy. (see http://achinghead.com/ python-markdown-adding-insert-delete.html this page for details) """ DEL_RE = r'(~~)(.*?)~~' def extendMarkdown(self, md, md_globals): # Create the del pattern del_tag = markdown.inlinepatterns.SimpleTagPattern( self.DEL_RE, 'del') # Insert del pattern into markdown parser md.inlinePatterns.add('del', del_tag, '>not_strong') extensions.append(_StrikeThroughExtension()) try: return markdown.markdown(text, extensions + ['mathjax']) except (ImportError, ValueError): # markdown raises ValueError or ImportError, depends on version # It is not clear, how to distinguish missing mathjax from other # errors. So keep going without mathjax. return markdown.markdown(text, extensions)