pmacs3/buffer/__init__.py

801 lines
27 KiB
Python

from util import defaultdict
import codecs, datetime, grp, os, pwd, re, shutil, stat, string
import fcntl, select, pty, threading
#import aes, dirutil, regex, highlight, lex, term
import dirutil, regex, highlight, lex, term
from point import Point
from subprocess import Popen, PIPE, STDOUT
from keyinput import MAP
# undo/redo stack constants
ACT_NONE = -1
ACT_NORM = 0
ACT_UNDO = 1
ACT_REDO = 2
STACK_LIMIT = 1024
def hasher(data):
try:
import hashlib
m = hashlib.md5(data)
except:
import md5
m = md5.new(data)
return m
class ReadOnlyError(Exception): pass
class FileChangedError(Exception): pass
class FileGoneError(Exception): pass
# used for undo/redo stacks when text will need to be added back
class AddMove(object):
def __init__(self, buffer, p, lines):
self.buffer = buffer
self.p = p
self.lines = lines
self.undo_id = buffer.undo_id
def restore(self, act=ACT_UNDO):
assert act == ACT_UNDO or act == ACT_REDO
self.buffer.insert_lines(self.p, self.lines, act)
def getpos(self):
return self.p
# used for undo/redo stacks when text will need to be removed
class DelMove(object):
def __init__(self, buffer, p1, p2):
self.buffer = buffer
self.p1 = p1
self.p2 = p2
self.undo_id = buffer.undo_id
def restore(self, act):
assert act == ACT_UNDO or act == ACT_REDO
self.buffer.delete(self.p1, self.p2, act)
def getpos(self):
return self.p1
# abstract class
class Buffer(object):
btype = 'generic'
modename = None
mac_re = re.compile('\r(?!\n)')
unix_re = re.compile('(?<!\r)\n')
win_re = re.compile('\r\n')
def __init__(self, stack_limit=STACK_LIMIT):
self.lines = [""]
self.windows = []
self.undo_id = 1
self.undo_stack = []
self.redo_stack = []
self.stack_limit = stack_limit
self.nl = '\n'
self.modified = False
self.highlights = {}
self.settings = {}
self.indentlvl = 4
self.writetabs = False
self.metadata = {}
def _detect_nl_type(self, data):
mac_c = len(self.mac_re.findall(data))
unix_c = len(self.unix_re.findall(data))
win_c = len(self.win_re.findall(data))
if (unix_c and mac_c) or (unix_c and win_c) or (mac_c and win_c):
# warn the user?
pass
if unix_c >= win_c and unix_c >= mac_c:
return '\n'
elif mac_c >= win_c:
return '\r'
else:
return '\r\n'
# basic file operation stuff
def _open_file_r(self, path):
path = os.path.realpath(path)
if not os.path.isfile(path):
raise Exception("Path '%s' does not exist" % (path))
if not os.access(path, os.R_OK):
raise Exception("Path '%s' cannot be read" % (path))
f = open(path, 'r')
return f
def _open_file_w(self, path):
if os.path.isfile(path):
raise Exception("Path '%s' already exists" % (path))
d = os.path.dirname(path)
if not os.access(d, os.R_OK):
raise Exception("Dir '%s' cannot be read" % (path))
if not os.access(d, os.W_OK):
raise Exception("Dir '%s' cannot be written" % (path))
f = open(path, 'w')
return f
def _temp_path(self, path):
(dirname, basename) = os.path.split(path)
return os.path.join(dirname, ".__%s__pmacs" % (basename))
# undo/redo stack
def _stack_trim(self, stack):
if self.stack_limit:
while len(stack) > self.stack_limit:
stack.pop(0)
def add_to_stack(self, move, act):
if act == ACT_NONE:
pass
elif act == ACT_NORM:
self.redo_stack = []
self.undo_stack.append(move)
self._stack_trim(self.undo_stack)
elif act == ACT_UNDO:
self.redo_stack.append(move)
self._stack_trim(self.redo_stack)
elif act == ACT_REDO:
self.undo_stack.append(move)
self._stack_trim(self.undo_stack)
else:
raise Exception("Invalid act: %d" % (act))
def undo(self):
if len(self.undo_stack):
undo_id = self.undo_stack[-1].undo_id
pos = None
while self.undo_stack and self.undo_stack[-1].undo_id == undo_id:
move = self.undo_stack.pop(-1)
move.restore(ACT_UNDO)
pos = move.getpos()
return pos
else:
raise Exception("Nothing to Undo!")
def redo(self):
if len(self.redo_stack):
undo_id = self.redo_stack[-1].undo_id
pos = None
while self.redo_stack and self.redo_stack[-1].undo_id == undo_id:
move = self.redo_stack.pop(-1)
move.restore(ACT_REDO)
pos = move.getpos()
return pos
else:
raise Exception("Nothing to Redo!")
# window-buffer communication
def add_window(self, w):
if w not in self.windows:
self.windows.append(w)
modename = w.mode.name
if modename not in self.highlights and w.mode.lexer is not None:
self.highlights[modename] = highlight.Highlighter(w.mode.lexer)
self.highlights[modename].highlight(self.lines)
if modename not in self.settings:
self.settings[modename] = {}
def remove_window(self, w):
if w in self.windows:
self.windows.remove(w)
modename = w.mode.name
if w.mode.name in self.highlights:
for w2 in self.windows:
if w2.mode.name == w.mode.name:
return
del self.highlights[w.mode.name]
def _region_add(self, p1, p2, lines, act):
move = DelMove(self, p1, p2)
self.add_to_stack(move, act)
for name in self.highlights:
self.highlights[name].relex_add(self.lines, p1.y, p1.x, lines)
for w in self.windows:
w.region_added(p1, lines)
def _region_del(self, p1, p2, lines, act):
move = AddMove(self, p1, lines)
self.add_to_stack(move, act)
for name in self.highlights:
self.highlights[name].relex_del(self.lines, p1.y, p1.x, p2.y, p2.x)
for w in self.windows:
w.region_removed(p1, p2)
# internal validation
def _validate_point(self, p):
self._validate_xy(p.x, p.y)
def _validate_xy(self, x, y):
assert y >= 0 and y < len(self.lines), \
"xy1: %d >= 0 and %d < %d" % (y, y, len(self.lines))
assert x >= 0 and x <= len(self.lines[y]), \
"xy2: %d >= 0 and %d <= %d" % (x, x, len(self.lines[y]))
def _validate_y(self, y):
assert y >= 0 and y < len(self.lines), \
"y: %d >= 0 and %d < %d" % (y, y, len(self.lines))
# deal with the actual logical document string
def num_chars(self):
n = 0
for line in self.lines[:-1]:
n += len(line) + 1
n += len(self.lines[-1])
return n
def num_lines(self):
return len(self.lines)
def make_string(self):
if self.writetabs:
lines = []
for line in self.lines:
i = 0
while i < len(line) and line[i] == ' ':
i += 1
j, k = i // self.indentlvl, i % self.indentlvl
lines.append(('\t' * j) + (' ' * k) + line[i:])
return self.nl.join(lines)
else:
return self.nl.join(self.lines)
# methods to be overridden by subclasses
def name(self):
return "Generic"
def close(self):
pass
def open(self):
pass
def changed(self):
return self.modified
def changed_on_disk(self):
return False
def reload(self):
raise Exception("%s reload: Unimplemented" % (self.name()))
def save_as(self, path, force=False):
# check to see if the path exists, and if we're prepared to overwrite it
# if yes to both, get its mode so we can preserve the path's permissions
mode = None
if os.path.exists(path):
if force:
mode = os.stat(self.path)[0]
else:
# XYZ
raise Exception("oh no! %r already exists" % path)
# create the string that we're going to write into the file
data = self.write_filter(self.make_string())
# create a safe temporary path to write to, and write out data to it
temp_path = self._temp_path()
f2 = self._open_file_w(temp_path)
f2.write(data)
f2.close()
# move the temporary file to the actual path; maybe change permissions
shutil.move(temp_path, path)
if mode:
os.chmod(path, mode)
# the file has not been modified now
self.modified = False
def readonly(self):
return False
def read_filter(self, data):
return data
def write_filter(self, data):
return data
# point retrieval
def get_buffer_start(self):
return Point(0, 0)
def get_buffer_end(self):
return Point(len(self.lines[-1]), len(self.lines) - 1)
# data retrieval
def get_sublines(self, p1, p2):
self._validate_point(p1)
self._validate_point(p2)
assert p1 <= p2, "p1.x (%d) > p2.x (%d)" % (p1.x, p2.x)
lines = []
x = p1.x
for i in range(p1.y, p2.y):
lines.append(self.lines[i][x:])
x = 0
lines.append(self.lines[p2.y][x:p2.x])
return lines
def get_substring(self, p1, p2):
lines = self.get_sublines(p1, p2)
return '\n'.join(lines)
# buffer set
def set_lines(self, lines, force=False):
if not force and self.readonly():
raise Exception("set_data: buffer is readonly")
start = self.get_buffer_start()
end = self.get_buffer_end()
self.delete(start, end, force=force)
self.insert_lines(start, lines, force=force)
self.modified = True
def set_data(self, data, force=False):
lines = data.split('\n')
self.set_lines(lines, force)
# append into buffer
def append_lines(self, lines, act=ACT_NORM, force=False):
p = self.get_buffer_end()
self.insert_lines(p, lines, act, force)
def append_string(self, s, act=ACT_NORM, force=False):
lines = s.split("\n")
self.insert_lines(lines, act, force)
# insertion into buffer
def insert_lines(self, p, lines, act=ACT_NORM, force=False):
llen = len(lines)
assert llen > 0
if not force and self.readonly():
raise ReadOnlyError("buffer is read-only")
p2 = p.vadd(len(lines[-1]), llen - 1)
if llen > 1:
self.lines.insert(p.y + 1, [])
self.lines[p.y + 1] = lines[-1] + self.lines[p.y][p.x:]
self.lines[p.y] = self.lines[p.y][:p.x] + lines[0]
for i in range(1, llen - 1):
self.lines.insert(p.y + i, lines[i])
else:
self.lines[p.y] = self.lines[p.y][:p.x] + lines[-1] + self.lines[p.y][p.x:]
self._region_add(p, p2, lines, act)
self.modified = True
def insert_string(self, p, s, act=ACT_NORM, force=False):
lines = s.split("\n")
self.insert_lines(p, lines, act, force)
# deletion from buffer
def delete(self, p1, p2, act=ACT_NORM, force=False):
"""delete characters from p1 up to p2 from the buffer"""
if not force and self.readonly():
raise ReadOnlyError("buffer is read-only")
self._validate_point(p1)
self._validate_point(p2)
if p1 == p2:
return
assert p1 < p2, "p1 %r > p2 %r" % (p1, p2)
lines = self.get_sublines(p1, p2)
line1 = self.lines[p1.y]
line2 = self.lines[p2.y]
self.lines[p1.y:p2.y+1] = ["%s%s" % (line1[:p1.x], line2[p2.x:])]
self._region_del(p1, p2, lines, act)
self.modified = True
def delete_char(self, p, act=ACT_NORM, force=False):
if p.x == len(self.lines[p.y]):
p2 = Point(0, p.y + 1)
else:
p2 = Point(p.x + 1, p.y)
self.delete(p, p2, act=act, force=force)
def overwrite_char(self, p, c, act=ACT_NORM, force=False):
self.delete_char(p, act=act, force=force)
self.insert_string(p, c, act=act, force=force)
def delete_line(self, y, act=ACT_NORM, force=False):
line = self.lines[y]
p1 = Point(0, y)
if y < len(self.lines) - 1:
p2 = Point(0, y + 1)
else:
p2 = Point(len(self.lines[-1]), y)
self.delete(p1, p2, act, force)
# random
def is_whitespace(self, y):
return regex.whitespace.match(self.lines[y])
def count_leading_whitespace(self, y):
m = regex.leading_whitespace.match(self.lines[y])
return m.end()
def detect_indent_level(self, y1, y2):
x = None
for y in range(y1, y2):
if self.is_whitespace(y):
continue
c = self.count_leading_whitespace(y)
if x is None:
x = c
else:
x = min(x, c)
return x or 0
# generic window functionality
def forward(self, p):
if p.x < len(self.lines[p.y]):
return Point(p.x + 1, p.y)
elif p.y < len(self.lines) - 1:
return Point(0, p.y + 1)
else:
return p
def backward(self, p):
if p.x > 0:
return Point(p.x - 1, p.y)
elif p.y > 0:
x = len(self.lines[p.y - 1])
return Point(x, p.y - 1)
else:
return p
def end_of_line(self, p):
return Point(len(self.lines[p.y]), p.y)
def start_of_line(self, p):
return Point(0, p.y)
def previous_line(self, p):
if p.y > 0:
return Point(p.x, p.y - 1)
else:
return p
def next_line(self, p):
if p.y < len(self.lines) - 1:
return Point(p.x, p.y + 1)
else:
return p
def left_delete(self, p):
(x, y) = p.xy()
if x > 0:
self.delete_char(Point(x - 1, y))
elif y > 0:
x = len(self.lines[y - 1])
self.delete_char(Point(x, y - 1))
def right_delete(self, p):
if (p.y < len(self.lines) - 1 or
p.x < len(self.lines[-1])):
self.delete_char(p)
class InterpreterPipeError(Exception):
pass
class InterpreterBuffer(Buffer):
_basename = 'Interpreter'
def create_name(cls, parent):
if hasattr(parent, 'path'):
return '*%s:%s*' % (cls._basename, parent.name())
else:
return '*%s*' % cls._basename
create_name = classmethod(create_name)
btype = 'interpreter'
readre = re.compile('^([A-Z]+):(.*)\n$')
def __init__(self, parent, app):
self.application = app
if parent and hasattr(parent, 'path'):
self.parent = parent
else:
self.parent = None
Buffer.__init__(self)
cmd = self.get_cmd()
env = dict(os.environ)
env.update(self.get_env())
f = open('/dev/null', 'w')
#f = open('pipe-errors', 'w')
self.pipe = Popen(cmd, stdin=PIPE, stdout=PIPE, stderr=f, env=env)
self.prompt = '***'
self.clear()
self.pipe_read()
self._name = self.create_name(parent)
def name(self):
return self._name
def get_env(self):
return {}
def get_cmd(self):
raise Exception('unimplemented')
def pipe_readline(self):
if self.pipe.poll() is not None:
raise InterpreterPipeError('broken pipe')
line = self.pipe.stdout.readline()
m = self.readre.match(line)
if m:
return (m.group(1), m.group(2))
else:
return (None, line.rstrip())
def pipe_read(self):
lines = []
while True:
(type_, value) = self.pipe_readline()
if type_ == 'PROMPT':
self.prompt = value.strip() + ' '
break
value.rstrip()
if value:
lines.append(value)
if lines:
output = '\n'.join(lines) + '\n'
p = self.get_buffer_end()
self.insert_string(p, output, force=True)
def pipe_write(self, s):
self.pipe.stdin.write("%s\n" % s)
self.pipe.stdin.flush()
def completions(self, word):
self.pipe_write("COMPLETE:%s" % word)
candidates = self.pipe_read_completions()
self.pipe_read()
return candidates
def pipe_read_completions(self):
try:
(typ_, value) = self.pipe_readline()
assert typ_ == 'COMPLETIONS', '%r %r' % (typ_, value)
candidates = [x for x in value.split('|') if x]
return candidates
except:
return []
def clear(self):
self.set_data('', force=True)
def changed(self):
return False
def readonly(self):
return True
class IperlBuffer(InterpreterBuffer):
_basename = 'IPerl'
btype = 'iperl'
modename = 'iperl'
def create_name(cls, parent):
if parent and hasattr(parent, 'path'):
if parent.path.endswith('.pm'):
return '*%s:%s*' % (cls._basename, parent.name())
else:
raise Exception("not a perl module")
else:
return '*%s*' % cls._basename
create_name = classmethod(create_name)
def get_cmd(self):
if self.parent:
return ('iperl', '-p', '-r', self.parent.path)
else:
return ('iperl', '-p')
def get_env(self):
lib = ':'.join(self.application.config.get('perl.libs', []))
return {'PERL5LIB': lib}
def readline_completions(self, x1, x2, line):
self.pipe.stdin.write("READLINE:%d:%d:%s\n" % (x1, x2, line))
self.pipe.stdin.flush()
(typ_, value) = self.pipe_readline()
assert typ_ == 'COMPLETIONS', '%r %r' % (typ_, value)
candidates = [x for x in value.split('|') if x]
self.pipe_read()
return candidates
class IpythonBuffer(InterpreterBuffer):
_basename = 'IPython'
btype = 'ipython'
modename = 'ipython'
def get_cmd(self):
if self.parent:
return ('epython', '-p', '-r', self.parent.path)
else:
return ('epython', '-p')
def get_env(self):
return {'PYTHONPATH': self.application.config.get('python.lib', '.')}
class BinaryDataException(Exception):
pass
class FileBuffer(Buffer):
btype = 'file'
def __init__(self, path, name=None):
'''fb = FileBuffer(path)'''
Buffer.__init__(self)
self.path = os.path.realpath(path)
self.checksum = None
self.bytemark = ''
if name is None:
self._name = os.path.basename(self.path)
else:
self._name = name
if os.path.exists(self.path) and not os.access(self.path, os.W_OK):
self._readonly = True
else:
self._readonly = False
def readonly(self):
return self._readonly
def _open_file_r(self, path=None):
if path is None:
path = self.path
path = os.path.realpath(path)
self.path = path
if not os.path.isfile(path):
raise Exception("Path '%s' does not exist" % (path))
if not os.access(path, os.R_OK):
raise Exception("Path '%s' cannot be read" % (path))
f = open(path, 'r')
return f
def _open_file_w(self, path=None, preserve=True):
if path is None:
path = self.path
if preserve and os.path.isfile(path):
raise Exception("Path '%s' already exists" % (path))
d = os.path.dirname(path)
if not os.access(d, os.R_OK):
raise Exception("Dir '%s' cannot be read" % (path))
if not os.access(d, os.W_OK):
raise Exception("Dir '%s' cannot be written" % (path))
f = open(path, 'w')
return f
def _temp_path(self, path=None):
if path is None:
path = self.path
(dirname, basename) = os.path.split(path)
return os.path.join(dirname, ".__%s__pmacs" % (basename))
# methods for dealing with the underlying resource, etc.
def name(self):
return self._name
def path_exists(self):
return os.path.exists(self.path)
def store_checksum(self, data):
self.checksum = hasher(data)
def read(self):
if self.path_exists():
f = self._open_file_r()
data = f.read()
if '\t' in data:
self.writetabs = True
f.close()
self.store_checksum(data)
else:
data = ''
if data.startswith('\xEF\xBB\xBF'):
# utf-8 bytemark
self.bytemark = data[:3]
data = data[3:]
self.nl = self._detect_nl_type(data)
data = self.read_filter(data)
try:
data = data.decode('utf-8')
data = data.replace("\t", " ")
return data
except UnicodeDecodeError:
raise BinaryDataException("binary files are not supported")
def open(self):
data = self.read()
self.lines = data.split(self.nl)
def reload(self):
data = self.read()
self.set_data(data)
def changed_on_disk(self):
assert self.checksum is not None
f = open(self.path)
data = f.read()
f.close()
m = hasher(data)
return self.checksum.digest() != m.digest()
def save(self, force=False):
if self.readonly():
raise ReadOnlyError("can't save read-only file")
if self.checksum is not None and force is False:
# the file already existed and we took a checksum so make sure it's
# still the same right now
if not self.path_exists():
raise FileGoneError("oh no! %r disappeared!" % self.path)
if self.changed_on_disk():
raise FileChangedError("oh no! %r has changed on-disk!" % self.path)
exists = os.path.exists(self.path)
if exists:
temp_path = self._temp_path()
shutil.copyfile(self.path, temp_path)
try:
data = self.make_string()
if self.windows[0].mode.savetabs:
data = data.replace(" ", "\t")
data = self.write_filter(data)
f2 = self._open_file_w(self.path, preserve=False)
f2.write(self.bytemark + data)
f2.close()
except Exception, e:
if exists: shutil.copyfile(temp_path, self.path)
raise e
else:
self.store_checksum(data)
self.modified = False
if exists: os.unlink(temp_path)
def save_as(self, path):
self.path = path
self.save()
#class AesBuffer(FileBuffer):
# btype = 'aesfile'
# def __init__(self, path, password, name=None):
# '''fb = FileBuffer(path)'''
# FileBuffer.__init__(self, path, name)
# self.password = password
# def read_filter(self, data):
# return aes.decrypt_data(data, self.password)
# def write_filter(self, data):
# return aes.encrypt_data(data, self.password)
class Binary32Buffer(FileBuffer):
btype = 'bin32file'
grouppad = 2
groupsize = 8
numgroups = 2
bytepad = 1
data = None
wordsize = 4
def __init__(self, path, name=None):
'''fb = FileBuffer(path)'''
FileBuffer.__init__(self, path, name)
def _detect_nl_type(self, data):
return '\n'
def cursorx_to_datax(self, cy, cx):
bytespace = 2 + self.bytepad
groupspace = bytespace * self.groupsize - self.bytepad + self.grouppad
groupmod = (cx + self.grouppad) % groupspace
if groupmod < self.grouppad:
return None
groupdiv = (cx + 2) // groupspace
if groupdiv >= self.numgroups:
return None
bytemod = (cx + self.bytepad - groupdiv) % bytespace
if bytemod == 0:
return None
bytediv = ((cx + self.bytepad) % groupspace) // bytespace
ix = self.groupsize * groupdiv + bytediv
if ix < len(self.rawdata[cy]):
return ix
else:
return None
def datax_to_cursorx(self, ix):
groupsize = (((2 + self.bytepad) * self.groupsize) + self.grouppad)
maxsize = groupsize * self.numgroups - self.grouppad
if ix < maxsize:
return (ix // self.groupsize) * (self.grouppad - 1) + ix * (2 + self.bytepad)
else:
return None
def datax_to_cursory(self, ix):
return ix // (self.groupsize * self.numgroups)
def get_address(self, cy, ix):
return (cy * self.numgroups * self.groupsize) + ix
def overwrite_char(self, p, c, act=ACT_NORM, force=False):
ix = self.cursorx_to_datax(p.y, p.x)
if ix is None:
return
Buffer.overwrite_char(self, p, c, act, force)
cx = self.datax_to_cursorx(ix)
c = chr(int(self.lines[p.y][cx:cx + 2], 16))
rawline = self.rawdata[p.y]
self.rawdata[p.y] = rawline[0:ix] + c + rawline[ix + 1:]
def read_filter(self, data):
bytepad = ' ' * self.bytepad
grouppad = ' ' * self.grouppad
self.rawdata = []
lines = []
i = 0
while i < len(data):
self.rawdata.append(data[i:i + self.numgroups * self.groupsize])
j = 0
groups = []
while j < self.numgroups * self.groupsize and i + j < len(data):
bytes = []
for c in data[i + j:i + j + self.groupsize]:
bytes.append(string.hexdigits[ord(c) / 16] + string.hexdigits[ord(c) % 16])
groups.append(bytepad.join(bytes))
j += self.groupsize
lines.append(grouppad.join(groups))
i += self.numgroups * self.groupsize
if not self.rawdata:
self.rawdata = ['']
return '\n'.join(lines)
def write_filter(self, data):
return ''.join(self.rawdata)
# log is another singleton
log = None
class LogBuffer(Buffer):
btype = 'log'
def __new__(cls, *args, **kwargs):
global log
if log is None:
log = object.__new__(LogBuffer, *args, **kwargs)
return log
def __init__(self): Buffer.__init__(self)
def clear(self): log.set_data('', force=True)
def name(self): return '*Log*'
def changed(self): return False
def close(self): global log; log = None
def readonly(self): return True