AIM-PIbd-32-Kurbanova-A-A/aimenv/Lib/site-packages/pythonwin/pywin/idle/FormatParagraph.py
2024-10-02 22:15:59 +04:00

167 lines
5.6 KiB
Python

# Extension to format a paragraph
# Does basic, standard text formatting, and also understands Python
# comment blocks. Thus, for editing Python source code, this
# extension is really only suitable for reformatting these comment
# blocks or triple-quoted strings.
# Known problems with comment reformatting:
# * If there is a selection marked, and the first line of the
# selection is not complete, the block will probably not be detected
# as comments, and will have the normal "text formatting" rules
# applied.
# * If a comment block has leading whitespace that mixes tabs and
# spaces, they will not be considered part of the same block.
# * Fancy comments, like this bulleted list, arent handled :-)
import re
class FormatParagraph:
menudefs = [
(
"edit",
[
("Format Paragraph", "<<format-paragraph>>"),
],
)
]
keydefs = {
"<<format-paragraph>>": ["<Alt-q>"],
}
unix_keydefs = {
"<<format-paragraph>>": ["<Meta-q>"],
}
def __init__(self, editwin):
self.editwin = editwin
def close(self):
self.editwin = None
def format_paragraph_event(self, event):
text = self.editwin.text
first, last = self.editwin.get_selection_indices()
if first and last:
data = text.get(first, last)
comment_header = ""
else:
first, last, comment_header, data = find_paragraph(
text, text.index("insert")
)
if comment_header:
# Reformat the comment lines - convert to text sans header.
lines = data.split("\n")
lines = map(lambda st, l=len(comment_header): st[l:], lines)
data = "\n".join(lines)
# Reformat to 70 chars or a 20 char width, whichever is greater.
format_width = max(70 - len(comment_header), 20)
newdata = reformat_paragraph(data, format_width)
# re-split and re-insert the comment header.
newdata = newdata.split("\n")
# If the block ends in a \n, we dont want the comment
# prefix inserted after it. (Im not sure it makes sense to
# reformat a comment block that isnt made of complete
# lines, but whatever!) Can't think of a clean soltution,
# so we hack away
block_suffix = ""
if not newdata[-1]:
block_suffix = "\n"
newdata = newdata[:-1]
builder = lambda item, prefix=comment_header: prefix + item
newdata = "\n".join([builder(d) for d in newdata]) + block_suffix
else:
# Just a normal text format
newdata = reformat_paragraph(data)
text.tag_remove("sel", "1.0", "end")
if newdata != data:
text.mark_set("insert", first)
text.undo_block_start()
text.delete(first, last)
text.insert(first, newdata)
text.undo_block_stop()
else:
text.mark_set("insert", last)
text.see("insert")
def find_paragraph(text, mark):
lineno, col = list(map(int, mark.split(".")))
line = text.get("%d.0" % lineno, "%d.0 lineend" % lineno)
while text.compare("%d.0" % lineno, "<", "end") and is_all_white(line):
lineno = lineno + 1
line = text.get("%d.0" % lineno, "%d.0 lineend" % lineno)
first_lineno = lineno
comment_header = get_comment_header(line)
comment_header_len = len(comment_header)
while get_comment_header(line) == comment_header and not is_all_white(
line[comment_header_len:]
):
lineno = lineno + 1
line = text.get("%d.0" % lineno, "%d.0 lineend" % lineno)
last = "%d.0" % lineno
# Search back to beginning of paragraph
lineno = first_lineno - 1
line = text.get("%d.0" % lineno, "%d.0 lineend" % lineno)
while (
lineno > 0
and get_comment_header(line) == comment_header
and not is_all_white(line[comment_header_len:])
):
lineno = lineno - 1
line = text.get("%d.0" % lineno, "%d.0 lineend" % lineno)
first = "%d.0" % (lineno + 1)
return first, last, comment_header, text.get(first, last)
def reformat_paragraph(data, limit=70):
lines = data.split("\n")
i = 0
n = len(lines)
while i < n and is_all_white(lines[i]):
i = i + 1
if i >= n:
return data
indent1 = get_indent(lines[i])
if i + 1 < n and not is_all_white(lines[i + 1]):
indent2 = get_indent(lines[i + 1])
else:
indent2 = indent1
new = lines[:i]
partial = indent1
while i < n and not is_all_white(lines[i]):
# XXX Should take double space after period (etc.) into account
words = re.split("(\s+)", lines[i])
for j in range(0, len(words), 2):
word = words[j]
if not word:
continue # Can happen when line ends in whitespace
if len((partial + word).expandtabs()) > limit and partial != indent1:
new.append(partial.rstrip())
partial = indent2
partial = partial + word + " "
if j + 1 < len(words) and words[j + 1] != " ":
partial = partial + " "
i = i + 1
new.append(partial.rstrip())
# XXX Should reformat remaining paragraphs as well
new.extend(lines[i:])
return "\n".join(new)
def is_all_white(line):
return re.match(r"^\s*$", line) is not None
def get_indent(line):
return re.match(r"^(\s*)", line).group()
def get_comment_header(line):
m = re.match(r"^(\s*#*)", line)
if m is None:
return ""
return m.group(1)