185 lines
5.1 KiB
Python
185 lines
5.1 KiB
Python
# Copyright (c) 2010-2024 openpyxl
|
|
|
|
import re
|
|
from openpyxl.descriptors import (
|
|
Strict,
|
|
Integer,
|
|
String,
|
|
Typed,
|
|
)
|
|
from openpyxl.utils import quote_sheetname, absolute_coordinate
|
|
from openpyxl.utils.cell import SHEET_TITLE, SHEETRANGE_RE, RANGE_EXPR
|
|
|
|
from .cell_range import MultiCellRange
|
|
|
|
COL_RANGE = r"""(?P<cols>[$]?(?P<min_col>[a-zA-Z]{1,3}):[$]?(?P<max_col>[a-zA-Z]{1,3}))"""
|
|
COL_RANGE_RE = re.compile(COL_RANGE)
|
|
ROW_RANGE = r"""(?P<rows>[$]?(?P<min_row>\d+):[$]?(?P<max_row>\d+))"""
|
|
ROW_RANGE_RE = re.compile(ROW_RANGE)
|
|
TITLES_REGEX = re.compile("""{0}{1}?,?{2}?,?""".format(SHEET_TITLE, ROW_RANGE, COL_RANGE),
|
|
re.VERBOSE)
|
|
PRINT_AREA_RE = re.compile(f"({SHEET_TITLE})?(?P<cells>{RANGE_EXPR})", re.VERBOSE)
|
|
|
|
class ColRange(Strict):
|
|
"""
|
|
Represent a range of at least one column
|
|
"""
|
|
|
|
min_col = String()
|
|
max_col = String()
|
|
|
|
|
|
def __init__(self, range_string=None, min_col=None, max_col=None):
|
|
if range_string is not None:
|
|
match = COL_RANGE_RE.match(range_string)
|
|
if not match:
|
|
raise ValueError(f"{range_string} is not a valid column range")
|
|
min_col, max_col = match.groups()[1:]
|
|
self.min_col = min_col
|
|
self.max_col = max_col
|
|
|
|
|
|
def __eq__(self, other):
|
|
if isinstance(other, self.__class__):
|
|
return (self.min_col == other.min_col
|
|
and
|
|
self.max_col == other.max_col)
|
|
elif isinstance(other, str):
|
|
return (str(self) == other
|
|
or
|
|
f"{self.min_col}:{self.max_col}")
|
|
return False
|
|
|
|
|
|
def __repr__(self):
|
|
return f"Range of columns from '{self.min_col}' to '{self.max_col}'"
|
|
|
|
|
|
def __str__(self):
|
|
return f"${self.min_col}:${self.max_col}"
|
|
|
|
|
|
class RowRange(Strict):
|
|
"""
|
|
Represent a range of at least one row
|
|
"""
|
|
|
|
min_row = Integer()
|
|
max_row = Integer()
|
|
|
|
def __init__(self, range_string=None, min_row=None, max_row=None):
|
|
if range_string is not None:
|
|
match = ROW_RANGE_RE.match(range_string)
|
|
if not match:
|
|
raise ValueError(f"{range_string} is not a valid row range")
|
|
min_row, max_row = match.groups()[1:]
|
|
self.min_row = min_row
|
|
self.max_row = max_row
|
|
|
|
|
|
def __eq__(self, other):
|
|
if isinstance(other, self.__class__):
|
|
return (self.min_row == other.min_row
|
|
and
|
|
self.max_row == other.max_row)
|
|
elif isinstance(other, str):
|
|
return (str(self) == other
|
|
or
|
|
f"{self.min_row}:{self.max_row}")
|
|
return False
|
|
|
|
def __repr__(self):
|
|
return f"Range of rows from '{self.min_row}' to '{self.max_row}'"
|
|
|
|
|
|
def __str__(self):
|
|
return f"${self.min_row}:${self.max_row}"
|
|
|
|
|
|
class PrintTitles(Strict):
|
|
"""
|
|
Contains at least either a range of rows or columns
|
|
"""
|
|
|
|
cols = Typed(expected_type=ColRange, allow_none=True)
|
|
rows = Typed(expected_type=RowRange, allow_none=True)
|
|
title = String()
|
|
|
|
|
|
def __init__(self, cols=None, rows=None, title=""):
|
|
self.cols = cols
|
|
self.rows = rows
|
|
self.title = title
|
|
|
|
|
|
@classmethod
|
|
def from_string(cls, value):
|
|
kw = dict((k, v) for match in TITLES_REGEX.finditer(value)
|
|
for k, v in match.groupdict().items() if v)
|
|
|
|
if not kw:
|
|
raise ValueError(f"{value} is not a valid print titles definition")
|
|
|
|
cols = rows = None
|
|
|
|
if "cols" in kw:
|
|
cols = ColRange(kw["cols"])
|
|
if "rows" in kw:
|
|
rows = RowRange(kw["rows"])
|
|
|
|
title = kw.get("quoted") or kw.get("notquoted")
|
|
|
|
return cls(cols=cols, rows=rows, title=title)
|
|
|
|
|
|
def __eq__(self, other):
|
|
if isinstance(other, self.__class__):
|
|
return (self.cols == other.cols
|
|
and
|
|
self.rows == other.rows
|
|
and
|
|
self.title == other.title)
|
|
elif isinstance(other, str):
|
|
return str(self) == other
|
|
return False
|
|
|
|
def __repr__(self):
|
|
return f"Print titles for sheet {self.title} cols {self.rows}, rows {self.cols}"
|
|
|
|
|
|
def __str__(self):
|
|
title = quote_sheetname(self.title)
|
|
titles = ",".join([f"{title}!{value}" for value in (self.rows, self.cols) if value])
|
|
return titles or ""
|
|
|
|
|
|
class PrintArea(MultiCellRange):
|
|
|
|
|
|
@classmethod
|
|
def from_string(cls, value):
|
|
new = []
|
|
for m in PRINT_AREA_RE.finditer(value): # can be multiple
|
|
coord = m.group("cells")
|
|
if coord:
|
|
new.append(coord)
|
|
return cls(new)
|
|
|
|
|
|
def __init__(self, ranges=(), title=""):
|
|
self.title = ""
|
|
super().__init__(ranges)
|
|
|
|
|
|
def __str__(self):
|
|
if self.ranges:
|
|
return ",".join([f"{quote_sheetname(self.title)}!{absolute_coordinate(str(range))}"
|
|
for range in self.sorted()])
|
|
return ""
|
|
|
|
|
|
def __eq__(self, other):
|
|
super().__eq__(other)
|
|
if isinstance(other, str):
|
|
return str(self) == other
|