Intelegentny_Pszczelarz/.venv/Lib/site-packages/markdown/extensions/tables.py

237 lines
7.9 KiB
Python
Raw Normal View History

2023-06-19 00:49:18 +02:00
"""
Tables Extension for Python-Markdown
====================================
Added parsing of tables to Python-Markdown.
See <https://Python-Markdown.github.io/extensions/tables>
for documentation.
Original code Copyright 2009 [Waylan Limberg](http://achinghead.com)
All changes Copyright 2008-2014 The Python Markdown Project
License: [BSD](https://opensource.org/licenses/bsd-license.php)
"""
from . import Extension
from ..blockprocessors import BlockProcessor
import xml.etree.ElementTree as etree
import re
PIPE_NONE = 0
PIPE_LEFT = 1
PIPE_RIGHT = 2
class TableProcessor(BlockProcessor):
""" Process Tables. """
RE_CODE_PIPES = re.compile(r'(?:(\\\\)|(\\`+)|(`+)|(\\\|)|(\|))')
RE_END_BORDER = re.compile(r'(?<!\\)(?:\\\\)*\|$')
def __init__(self, parser, config):
self.border = False
self.separator = ''
self.config = config
super().__init__(parser)
def test(self, parent, block):
"""
Ensure first two rows (column header and separator row) are valid table rows.
Keep border check and separator row do avoid repeating the work.
"""
is_table = False
rows = [row.strip(' ') for row in block.split('\n')]
if len(rows) > 1:
header0 = rows[0]
self.border = PIPE_NONE
if header0.startswith('|'):
self.border |= PIPE_LEFT
if self.RE_END_BORDER.search(header0) is not None:
self.border |= PIPE_RIGHT
row = self._split_row(header0)
row0_len = len(row)
is_table = row0_len > 1
# Each row in a single column table needs at least one pipe.
if not is_table and row0_len == 1 and self.border:
for index in range(1, len(rows)):
is_table = rows[index].startswith('|')
if not is_table:
is_table = self.RE_END_BORDER.search(rows[index]) is not None
if not is_table:
break
if is_table:
row = self._split_row(rows[1])
is_table = (len(row) == row0_len) and set(''.join(row)) <= set('|:- ')
if is_table:
self.separator = row
return is_table
def run(self, parent, blocks):
""" Parse a table block and build table. """
block = blocks.pop(0).split('\n')
header = block[0].strip(' ')
rows = [] if len(block) < 3 else block[2:]
# Get alignment of columns
align = []
for c in self.separator:
c = c.strip(' ')
if c.startswith(':') and c.endswith(':'):
align.append('center')
elif c.startswith(':'):
align.append('left')
elif c.endswith(':'):
align.append('right')
else:
align.append(None)
# Build table
table = etree.SubElement(parent, 'table')
thead = etree.SubElement(table, 'thead')
self._build_row(header, thead, align)
tbody = etree.SubElement(table, 'tbody')
if len(rows) == 0:
# Handle empty table
self._build_empty_row(tbody, align)
else:
for row in rows:
self._build_row(row.strip(' '), tbody, align)
def _build_empty_row(self, parent, align):
"""Build an empty row."""
tr = etree.SubElement(parent, 'tr')
count = len(align)
while count:
etree.SubElement(tr, 'td')
count -= 1
def _build_row(self, row, parent, align):
""" Given a row of text, build table cells. """
tr = etree.SubElement(parent, 'tr')
tag = 'td'
if parent.tag == 'thead':
tag = 'th'
cells = self._split_row(row)
# We use align here rather than cells to ensure every row
# contains the same number of columns.
for i, a in enumerate(align):
c = etree.SubElement(tr, tag)
try:
c.text = cells[i].strip(' ')
except IndexError: # pragma: no cover
c.text = ""
if a:
if self.config['use_align_attribute']:
c.set('align', a)
else:
c.set('style', f'text-align: {a};')
def _split_row(self, row):
""" split a row of text into list of cells. """
if self.border:
if row.startswith('|'):
row = row[1:]
row = self.RE_END_BORDER.sub('', row)
return self._split(row)
def _split(self, row):
""" split a row of text with some code into a list of cells. """
elements = []
pipes = []
tics = []
tic_points = []
tic_region = []
good_pipes = []
# Parse row
# Throw out \\, and \|
for m in self.RE_CODE_PIPES.finditer(row):
# Store ` data (len, start_pos, end_pos)
if m.group(2):
# \`+
# Store length of each tic group: subtract \
tics.append(len(m.group(2)) - 1)
# Store start of group, end of group, and escape length
tic_points.append((m.start(2), m.end(2) - 1, 1))
elif m.group(3):
# `+
# Store length of each tic group
tics.append(len(m.group(3)))
# Store start of group, end of group, and escape length
tic_points.append((m.start(3), m.end(3) - 1, 0))
# Store pipe location
elif m.group(5):
pipes.append(m.start(5))
# Pair up tics according to size if possible
# Subtract the escape length *only* from the opening.
# Walk through tic list and see if tic has a close.
# Store the tic region (start of region, end of region).
pos = 0
tic_len = len(tics)
while pos < tic_len:
try:
tic_size = tics[pos] - tic_points[pos][2]
if tic_size == 0:
raise ValueError
index = tics[pos + 1:].index(tic_size) + 1
tic_region.append((tic_points[pos][0], tic_points[pos + index][1]))
pos += index + 1
except ValueError:
pos += 1
# Resolve pipes. Check if they are within a tic pair region.
# Walk through pipes comparing them to each region.
# - If pipe position is less that a region, it isn't in a region
# - If it is within a region, we don't want it, so throw it out
# - If we didn't throw it out, it must be a table pipe
for pipe in pipes:
throw_out = False
for region in tic_region:
if pipe < region[0]:
# Pipe is not in a region
break
elif region[0] <= pipe <= region[1]:
# Pipe is within a code region. Throw it out.
throw_out = True
break
if not throw_out:
good_pipes.append(pipe)
# Split row according to table delimiters.
pos = 0
for pipe in good_pipes:
elements.append(row[pos:pipe])
pos = pipe + 1
elements.append(row[pos:])
return elements
class TableExtension(Extension):
""" Add tables to Markdown. """
def __init__(self, **kwargs):
self.config = {
'use_align_attribute': [False, 'True to use align attribute instead of style.'],
}
super().__init__(**kwargs)
def extendMarkdown(self, md):
""" Add an instance of TableProcessor to BlockParser. """
if '|' not in md.ESCAPED_CHARS:
md.ESCAPED_CHARS.append('|')
processor = TableProcessor(md.parser, self.getConfigs())
md.parser.blockprocessors.register(processor, 'table', 75)
def makeExtension(**kwargs): # pragma: no cover
return TableExtension(**kwargs)