def detect_paragraph_type(txt):
'''
Tries to determine the paragraph type of the document.
block: Paragraphs are separated by a blank line.
single: Each line is a paragraph.
print: Each paragraph starts with a 2+ spaces or a tab
and ends when a new paragraph is reached.
unformatted: most lines have hard line breaks, few/no blank lines or indents
returns block, single, print, unformatted
'''
txt = txt.replace('\r\n', '\n')
txt = txt.replace('\r', '\n')
txt_line_count = len(re.findall('(?mu)^\s*.+$', txt))
# Check for hard line breaks - true if 55% of the doc breaks in the same region
docanalysis = DocAnalysis('txt', txt)
hardbreaks = docanalysis.line_histogram(.55)
if hardbreaks:
# Determine print percentage
tab_line_count = len(re.findall('(?mu)^(\t|\s{2,}).+$', txt))
print_percent = tab_line_count / float(txt_line_count)
# Determine block percentage
empty_line_count = len(re.findall('(?mu)^\s*$', txt))
block_percent = empty_line_count / float(txt_line_count)
# Compare the two types - the type with the larger number of instances wins
# in cases where only one or the other represents the vast majority of the document neither wins
if print_percent >= block_percent:
if .15 <= print_percent <= .75:
return 'print'
elif .15 <= block_percent <= .75:
return 'block'
# Assume unformatted text with hardbreaks if nothing else matches
return 'unformatted'
# return single if hardbreaks is false
return 'single'
def detect_paragraph_type(txt):
'''
Tries to determine the paragraph type of the document.
block: Paragraphs are separated by a blank line.
single: Each line is a paragraph.
print: Each paragraph starts with a 2+ spaces or a tab
and ends when a new paragraph is reached.
unformatted: most lines have hard line breaks, few/no blank lines or indents
returns block, single, print, unformatted
'''
txt = txt.replace('\r\n', '\n')
txt = txt.replace('\r', '\n')
txt_line_count = len(re.findall(r'(?mu)^\s*.+$', txt))
# Check for hard line breaks - true if 55% of the doc breaks in the same region
docanalysis = DocAnalysis('txt', txt)
hardbreaks = docanalysis.line_histogram(.55)
if hardbreaks:
# Determine print percentage
tab_line_count = len(re.findall(r'(?mu)^(\t|\s{2,}).+$', txt))
print_percent = tab_line_count / float(txt_line_count)
# Determine block percentage
empty_line_count = len(re.findall(r'(?mu)^\s*$', txt))
block_percent = empty_line_count / float(txt_line_count)
# Compare the two types - the type with the larger number of instances wins
# in cases where only one or the other represents the vast majority of the document neither wins
if print_percent >= block_percent:
if .15 <= print_percent <= .75:
return 'print'
elif .15 <= block_percent <= .75:
return 'block'
# Assume unformatted text with hardbreaks if nothing else matches
return 'unformatted'
# return single if hardbreaks is false
return 'single'
def __call__(self, html):
self.log.debug("********* Heuristic processing HTML *********")
# Count the words in the document to estimate how many chapters to look for and whether
# other types of processing are attempted
try:
self.totalwords = self.get_word_count(html)
except:
self.log.warn("Can't get wordcount")
if self.totalwords < 50:
self.log.warn("flow is too short, not running heuristics")
return html
is_abbyy = self.is_abbyy(html)
if is_abbyy:
html = self.abbyy_processor(html)
# Arrange line feeds and </p> tags so the line_length and no_markup functions work correctly
html = self.arrange_htm_line_endings(html)
# self.dump(html, 'after_arrange_line_endings')
if self.cleanup_required():
# ##### Check Markup ######
#
# some lit files don't have any <p> tags or equivalent (generally just plain text between
# <pre> tags), check and mark up line endings if required before proceeding
# fix indents must run after this step
if self.no_markup(html, 0.1):
self.log.debug("not enough paragraph markers, adding now")
# markup using text processing
html = self.markup_pre(html)
# Replace series of non-breaking spaces with text-indent
if getattr(self.extra_opts, 'fix_indents', False):
html = self.fix_nbsp_indents(html)
if self.cleanup_required():
# fix indents must run before this step, as it removes non-breaking spaces
html = self.cleanup_markup(html)
is_pdftohtml = self.is_pdftohtml(html)
if is_pdftohtml:
self.line_open = "<(?P<outer>p)[^>]*>(\\s*<[ibu][^>]*>)?\\s*"
self.line_close = "\\s*(</[ibu][^>]*>\\s*)?</(?P=outer)>"
# ADE doesn't render <br />, change to empty paragraphs
# html = re.sub('<br[^>]*>', u'<p>\u00a0</p>', html)
# Determine whether the document uses interleaved blank lines
self.blanks_between_paragraphs = self.analyze_blanks(html)
# detect chapters/sections to match xpath or splitting logic
if getattr(self.extra_opts, 'markup_chapter_headings', False):
html = self.markup_chapters(html, self.totalwords,
self.blanks_between_paragraphs)
# self.dump(html, 'after_chapter_markup')
if getattr(self.extra_opts, 'italicize_common_cases', False):
html = self.markup_italicis(html)
# If more than 40% of the lines are empty paragraphs and the user has enabled delete
# blank paragraphs then delete blank lines to clean up spacing
if self.blanks_between_paragraphs and getattr(
self.extra_opts, 'delete_blank_paragraphs', False):
self.log.debug("deleting blank lines")
self.blanks_deleted = True
html = self.multi_blank.sub(
'\n<p class="softbreak" style="margin-top:.5em; page-break-before:avoid; text-align:center"> </p>',
html)
html = self.blankreg.sub('', html)
# Determine line ending type
# Some OCR sourced files have line breaks in the html using a combination of span & p tags
# span are used for hard line breaks, p for new paragraphs. Determine which is used so
# that lines can be un-wrapped across page boundaries
format = self.analyze_line_endings(html)
# Check Line histogram to determine if the document uses hard line breaks, If 50% or
# more of the lines break in the same region of the document then unwrapping is required
docanalysis = DocAnalysis(format, html)
hardbreaks = docanalysis.line_histogram(.50)
self.log.debug("Hard line breaks check returned " +
unicode_type(hardbreaks))
# Calculate Length
unwrap_factor = getattr(self.extra_opts, 'html_unwrap_factor', 0.4)
length = docanalysis.line_length(unwrap_factor)
self.log.debug("Median line length is " + unicode_type(length) +
", calculated with " + format + " format")
# ##### Unwrap lines ######
if getattr(self.extra_opts, 'unwrap_lines', False):
# only go through unwrapping code if the histogram shows unwrapping is required or if the user decreased the default unwrap_factor
if hardbreaks or unwrap_factor < 0.4:
self.log.debug("Unwrapping required, unwrapping Lines")
# Dehyphenate with line length limiters
dehyphenator = Dehyphenator(self.extra_opts.verbose, self.log)
html = dehyphenator(html, 'html', length)
html = self.punctuation_unwrap(length, html, 'html')
if getattr(self.extra_opts, 'dehyphenate', False):
# dehyphenate in cleanup mode to fix anything previous conversions/editing missed
self.log.debug("Fixing hyphenated content")
dehyphenator = Dehyphenator(self.extra_opts.verbose, self.log)
html = dehyphenator(html, 'html_cleanup', length)
html = dehyphenator(html, 'individual_words', length)
# If still no sections after unwrapping mark split points on lines with no punctuation
if self.html_preprocess_sections < self.min_chapters and getattr(
self.extra_opts, 'markup_chapter_headings', False):
self.log.debug(
"Looking for more split points based on punctuation,"
" currently have " +
unicode_type(self.html_preprocess_sections))
chapdetect3 = re.compile(
r'<(?P<styles>(p|div)[^>]*)>\s*(?P<section>(<span[^>]*>)?\s*(?!([\W]+\s*)+)'
r'(<[ibu][^>]*>){0,2}\s*(<span[^>]*>)?\s*(<[ibu][^>]*>){0,2}\s*(<span[^>]*>)?\s*'
r'.?(?=[a-z#\-*\s]+<)([a-z#-*]+\s*){1,5}\s*\s*(</span>)?(</[ibu]>){0,2}\s*'
r'(</span>)?\s*(</[ibu]>){0,2}\s*(</span>)?\s*</(p|div)>)',
re.IGNORECASE)
html = chapdetect3.sub(self.chapter_break, html)
if getattr(self.extra_opts, 'renumber_headings', False):
# search for places where a first or second level heading is immediately followed by another
# top level heading. demote the second heading to h3 to prevent splitting between chapter
# headings and titles, images, etc
doubleheading = re.compile(
r'(?P<firsthead><h(1|2)[^>]*>.+?</h(1|2)>\s*(<(?!h\d)[^>]*>\s*)*)<h(1|2)(?P<secondhead>[^>]*>.+?)</h(1|2)>',
re.IGNORECASE)
html = doubleheading.sub(
'\\g<firsthead>' + '\n<h3' + '\\g<secondhead>' + '</h3>', html)
# If scene break formatting is enabled, find all blank paragraphs that definitely aren't scenebreaks,
# style it with the 'whitespace' class. All remaining blank lines are styled as softbreaks.
# Multiple sequential blank paragraphs are merged with appropriate margins
# If non-blank scene breaks exist they are center aligned and styled with appropriate margins.
if getattr(self.extra_opts, 'format_scene_breaks', False):
self.log.debug('Formatting scene breaks')
html = re.sub('(?i)<div[^>]*>\\s*<br(\\s?/)?>\\s*</div>',
'<p></p>', html)
html = self.detect_scene_breaks(html)
html = self.detect_whitespace(html)
html = self.detect_soft_breaks(html)
blanks_count = len(self.any_multi_blank.findall(html))
if blanks_count >= 1:
html = self.merge_blanks(html, blanks_count)
detected_scene_break = re.compile(
r'<p class="scenebreak"[^>]*>.*?</p>')
scene_break_count = len(detected_scene_break.findall(html))
# If the user has enabled scene break replacement, then either softbreaks
# or 'hard' scene breaks are replaced, depending on which is in use
# Otherwise separator lines are centered, use a bit larger margin in this case
replacement_break = getattr(self.extra_opts,
'replace_scene_breaks', None)
if replacement_break:
replacement_break = self.markup_user_break(replacement_break)
if scene_break_count >= 1:
html = detected_scene_break.sub(replacement_break, html)
html = re.sub('<p\\s+class="softbreak"[^>]*>\\s*</p>',
replacement_break, html)
else:
html = re.sub('<p\\s+class="softbreak"[^>]*>\\s*</p>',
replacement_break, html)
if self.deleted_nbsps:
# put back non-breaking spaces in empty paragraphs so they render correctly
html = self.anyblank.sub(
'\n' + r'\g<openline>' + '\u00a0' + r'\g<closeline>', html)
return html
def __call__(self, html):
self.log.debug("********* Heuristic processing HTML *********")
# Count the words in the document to estimate how many chapters to look for and whether
# other types of processing are attempted
try:
self.totalwords = self.get_word_count(html)
except:
self.log.warn("Can't get wordcount")
if self.totalwords < 50:
self.log.warn("flow is too short, not running heuristics")
return html
is_abbyy = self.is_abbyy(html)
if is_abbyy:
html = self.abbyy_processor(html)
# Arrange line feeds and </p> tags so the line_length and no_markup functions work correctly
html = self.arrange_htm_line_endings(html)
# self.dump(html, 'after_arrange_line_endings')
if self.cleanup_required():
# ##### Check Markup ######
#
# some lit files don't have any <p> tags or equivalent (generally just plain text between
# <pre> tags), check and mark up line endings if required before proceeding
# fix indents must run after this step
if self.no_markup(html, 0.1):
self.log.debug("not enough paragraph markers, adding now")
# markup using text processing
html = self.markup_pre(html)
# Replace series of non-breaking spaces with text-indent
if getattr(self.extra_opts, 'fix_indents', False):
html = self.fix_nbsp_indents(html)
if self.cleanup_required():
# fix indents must run before this step, as it removes non-breaking spaces
html = self.cleanup_markup(html)
is_pdftohtml = self.is_pdftohtml(html)
if is_pdftohtml:
self.line_open = "<(?P<outer>p)[^>]*>(\\s*<[ibu][^>]*>)?\\s*"
self.line_close = "\\s*(</[ibu][^>]*>\\s*)?</(?P=outer)>"
# ADE doesn't render <br />, change to empty paragraphs
# html = re.sub('<br[^>]*>', u'<p>\u00a0</p>', html)
# Determine whether the document uses interleaved blank lines
self.blanks_between_paragraphs = self.analyze_blanks(html)
# detect chapters/sections to match xpath or splitting logic
if getattr(self.extra_opts, 'markup_chapter_headings', False):
html = self.markup_chapters(html, self.totalwords, self.blanks_between_paragraphs)
# self.dump(html, 'after_chapter_markup')
if getattr(self.extra_opts, 'italicize_common_cases', False):
html = self.markup_italicis(html)
# If more than 40% of the lines are empty paragraphs and the user has enabled delete
# blank paragraphs then delete blank lines to clean up spacing
if self.blanks_between_paragraphs and getattr(self.extra_opts, 'delete_blank_paragraphs', False):
self.log.debug("deleting blank lines")
self.blanks_deleted = True
html = self.multi_blank.sub('\n<p class="softbreak" style="margin-top:.5em; page-break-before:avoid; text-align:center"> </p>', html)
html = self.blankreg.sub('', html)
# Determine line ending type
# Some OCR sourced files have line breaks in the html using a combination of span & p tags
# span are used for hard line breaks, p for new paragraphs. Determine which is used so
# that lines can be un-wrapped across page boundaries
format = self.analyze_line_endings(html)
# Check Line histogram to determine if the document uses hard line breaks, If 50% or
# more of the lines break in the same region of the document then unwrapping is required
docanalysis = DocAnalysis(format, html)
hardbreaks = docanalysis.line_histogram(.50)
self.log.debug("Hard line breaks check returned "+unicode_type(hardbreaks))
# Calculate Length
unwrap_factor = getattr(self.extra_opts, 'html_unwrap_factor', 0.4)
length = docanalysis.line_length(unwrap_factor)
self.log.debug("Median line length is " + unicode_type(length) + ", calculated with " + format + " format")
# ##### Unwrap lines ######
if getattr(self.extra_opts, 'unwrap_lines', False):
# only go through unwrapping code if the histogram shows unwrapping is required or if the user decreased the default unwrap_factor
if hardbreaks or unwrap_factor < 0.4:
self.log.debug("Unwrapping required, unwrapping Lines")
# Dehyphenate with line length limiters
dehyphenator = Dehyphenator(self.extra_opts.verbose, self.log)
html = dehyphenator(html,'html', length)
html = self.punctuation_unwrap(length, html, 'html')
if getattr(self.extra_opts, 'dehyphenate', False):
# dehyphenate in cleanup mode to fix anything previous conversions/editing missed
self.log.debug("Fixing hyphenated content")
dehyphenator = Dehyphenator(self.extra_opts.verbose, self.log)
html = dehyphenator(html,'html_cleanup', length)
html = dehyphenator(html, 'individual_words', length)
# If still no sections after unwrapping mark split points on lines with no punctuation
if self.html_preprocess_sections < self.min_chapters and getattr(self.extra_opts, 'markup_chapter_headings', False):
self.log.debug("Looking for more split points based on punctuation,"
" currently have " + unicode_type(self.html_preprocess_sections))
chapdetect3 = re.compile(
r'<(?P<styles>(p|div)[^>]*)>\s*(?P<section>(<span[^>]*>)?\s*(?!([\W]+\s*)+)(<[ibu][^>]*>){0,2}\s*(<span[^>]*>)?\s*(<[ibu][^>]*>){0,2}\s*(<span[^>]*>)?\s*.?(?=[a-z#\-*\s]+<)([a-z#-*]+\s*){1,5}\s*\s*(</span>)?(</[ibu]>){0,2}\s*(</span>)?\s*(</[ibu]>){0,2}\s*(</span>)?\s*</(p|div)>)', re.IGNORECASE) # noqa
html = chapdetect3.sub(self.chapter_break, html)
if getattr(self.extra_opts, 'renumber_headings', False):
# search for places where a first or second level heading is immediately followed by another
# top level heading. demote the second heading to h3 to prevent splitting between chapter
# headings and titles, images, etc
doubleheading = re.compile(
r'(?P<firsthead><h(1|2)[^>]*>.+?</h(1|2)>\s*(<(?!h\d)[^>]*>\s*)*)<h(1|2)(?P<secondhead>[^>]*>.+?)</h(1|2)>', re.IGNORECASE)
html = doubleheading.sub('\\g<firsthead>'+'\n<h3'+'\\g<secondhead>'+'</h3>', html)
# If scene break formatting is enabled, find all blank paragraphs that definitely aren't scenebreaks,
# style it with the 'whitespace' class. All remaining blank lines are styled as softbreaks.
# Multiple sequential blank paragraphs are merged with appropriate margins
# If non-blank scene breaks exist they are center aligned and styled with appropriate margins.
if getattr(self.extra_opts, 'format_scene_breaks', False):
self.log.debug('Formatting scene breaks')
html = re.sub('(?i)<div[^>]*>\\s*<br(\\s?/)?>\\s*</div>', '<p></p>', html)
html = self.detect_scene_breaks(html)
html = self.detect_whitespace(html)
html = self.detect_soft_breaks(html)
blanks_count = len(self.any_multi_blank.findall(html))
if blanks_count >= 1:
html = self.merge_blanks(html, blanks_count)
detected_scene_break = re.compile(r'<p class="scenebreak"[^>]*>.*?</p>')
scene_break_count = len(detected_scene_break.findall(html))
# If the user has enabled scene break replacement, then either softbreaks
# or 'hard' scene breaks are replaced, depending on which is in use
# Otherwise separator lines are centered, use a bit larger margin in this case
replacement_break = getattr(self.extra_opts, 'replace_scene_breaks', None)
if replacement_break:
replacement_break = self.markup_user_break(replacement_break)
if scene_break_count >= 1:
html = detected_scene_break.sub(replacement_break, html)
html = re.sub('<p\\s+class="softbreak"[^>]*>\\s*</p>', replacement_break, html)
else:
html = re.sub('<p\\s+class="softbreak"[^>]*>\\s*</p>', replacement_break, html)
if self.deleted_nbsps:
# put back non-breaking spaces in empty paragraphs so they render correctly
html = self.anyblank.sub('\n'+r'\g<openline>'+u'\u00a0'+r'\g<closeline>', html)
return html
