mirror of
https://github.com/ankitects/anki.git
synced 2025-09-19 14:32:22 -04:00
315 lines
9.1 KiB
Python
315 lines
9.1 KiB
Python
# -*- coding: utf-8 -*-
|
|
# Copyright: Damien Elmes <anki@ichi2.net>
|
|
# License: GNU AGPL, version 3 or later; http://www.gnu.org/licenses/agpl.html
|
|
|
|
import re, os, random, time, types, math, htmlentitydefs, subprocess, \
|
|
tempfile, shutil
|
|
|
|
try:
|
|
import hashlib
|
|
md5 = hashlib.md5
|
|
except ImportError:
|
|
import md5
|
|
md5 = md5.new
|
|
|
|
from anki.lang import _, ngettext
|
|
import locale, sys
|
|
|
|
if sys.version_info[1] < 5:
|
|
def format_string(a, b):
|
|
return a % b
|
|
locale.format_string = format_string
|
|
|
|
# Time handling
|
|
##############################################################################
|
|
|
|
def intTime():
|
|
return int(time.time())
|
|
|
|
timeTable = {
|
|
"years": lambda n: ngettext("%s year", "%s years", n),
|
|
"months": lambda n: ngettext("%s month", "%s months", n),
|
|
"days": lambda n: ngettext("%s day", "%s days", n),
|
|
"hours": lambda n: ngettext("%s hour", "%s hours", n),
|
|
"minutes": lambda n: ngettext("%s minute", "%s minutes", n),
|
|
"seconds": lambda n: ngettext("%s second", "%s seconds", n),
|
|
}
|
|
|
|
afterTimeTable = {
|
|
"years": lambda n: ngettext("%s year<!--after-->", "%s years<!--after-->", n),
|
|
"months": lambda n: ngettext("%s month<!--after-->", "%s months<!--after-->", n),
|
|
"days": lambda n: ngettext("%s day<!--after-->", "%s days<!--after-->", n),
|
|
"hours": lambda n: ngettext("%s hour<!--after-->", "%s hours<!--after-->", n),
|
|
"minutes": lambda n: ngettext("%s minute<!--after-->", "%s minutes<!--after-->", n),
|
|
"seconds": lambda n: ngettext("%s second<!--after-->", "%s seconds<!--after-->", n),
|
|
}
|
|
|
|
shortTimeTable = {
|
|
"years": _("%sy"),
|
|
"months": _("%smo"),
|
|
"days": _("%sd"),
|
|
"hours": _("%sh"),
|
|
"minutes": _("%sm"),
|
|
"seconds": _("%ss"),
|
|
}
|
|
|
|
def fmtTimeSpan(time, pad=0, point=0, short=False, after=False):
|
|
"Return a string representing a time span (eg '2 days')."
|
|
(type, point) = optimalPeriod(time, point)
|
|
time = convertSecondsTo(time, type)
|
|
if not point:
|
|
time = math.floor(time)
|
|
if short:
|
|
fmt = shortTimeTable[type]
|
|
else:
|
|
if after:
|
|
fmt = afterTimeTable[type](_pluralCount(time, point))
|
|
else:
|
|
fmt = timeTable[type](_pluralCount(time, point))
|
|
timestr = "%(a)d.%(b)df" % {'a': pad, 'b': point}
|
|
return locale.format_string("%" + (fmt % timestr), time)
|
|
|
|
def optimalPeriod(time, point):
|
|
if abs(time) < 60:
|
|
type = "seconds"
|
|
point -= 1
|
|
elif abs(time) < 3599:
|
|
type = "minutes"
|
|
elif abs(time) < 60 * 60 * 24:
|
|
type = "hours"
|
|
elif abs(time) < 60 * 60 * 24 * 30:
|
|
type = "days"
|
|
elif abs(time) < 60 * 60 * 24 * 365:
|
|
type = "months"
|
|
point += 1
|
|
else:
|
|
type = "years"
|
|
point += 1
|
|
return (type, max(point, 0))
|
|
|
|
def convertSecondsTo(seconds, type):
|
|
if type == "seconds":
|
|
return seconds
|
|
elif type == "minutes":
|
|
return seconds / 60.0
|
|
elif type == "hours":
|
|
return seconds / 3600.0
|
|
elif type == "days":
|
|
return seconds / 86400.0
|
|
elif type == "months":
|
|
return seconds / 2592000.0
|
|
elif type == "years":
|
|
return seconds / 31536000.0
|
|
assert False
|
|
|
|
def _pluralCount(time, point):
|
|
if point:
|
|
return 2
|
|
return math.floor(time)
|
|
|
|
# Locale
|
|
##############################################################################
|
|
|
|
def fmtPercentage(float_value, point=1):
|
|
"Return float with percentage sign"
|
|
fmt = '%' + "0.%(b)df" % {'b': point}
|
|
return locale.format_string(fmt, float_value) + "%"
|
|
|
|
def fmtFloat(float_value, point=1):
|
|
"Return a string with decimal separator according to current locale"
|
|
fmt = '%' + "0.%(b)df" % {'b': point}
|
|
return locale.format_string(fmt, float_value)
|
|
|
|
# HTML
|
|
##############################################################################
|
|
|
|
def stripHTML(s):
|
|
s = re.sub("(?s)<style.*?>.*?</style>", "", s)
|
|
s = re.sub("(?s)<script.*?>.*?</script>", "", s)
|
|
s = re.sub("<.*?>", "", s)
|
|
s = entsToTxt(s)
|
|
return s
|
|
|
|
def stripHTMLMedia(s):
|
|
"Strip HTML but keep media filenames"
|
|
s = re.sub("<img src=[\"']?([^\"'>]+)[\"']? ?/?>", " \\1 ", s)
|
|
return stripHTML(s)
|
|
|
|
def minimizeHTML(s):
|
|
"Correct Qt's verbose bold/underline/etc."
|
|
s = re.sub('<span style="font-weight:600;">(.*?)</span>', '<b>\\1</b>',
|
|
s)
|
|
s = re.sub('<span style="font-style:italic;">(.*?)</span>', '<i>\\1</i>',
|
|
s)
|
|
s = re.sub('<span style="text-decoration: underline;">(.*?)</span>',
|
|
'<u>\\1</u>', s)
|
|
return s
|
|
|
|
def entsToTxt(html):
|
|
def fixup(m):
|
|
text = m.group(0)
|
|
if text[:2] == "&#":
|
|
# character reference
|
|
try:
|
|
if text[:3] == "&#x":
|
|
return unichr(int(text[3:-1], 16))
|
|
else:
|
|
return unichr(int(text[2:-1]))
|
|
except ValueError:
|
|
pass
|
|
else:
|
|
# named entity
|
|
try:
|
|
text = unichr(htmlentitydefs.name2codepoint[text[1:-1]])
|
|
except KeyError:
|
|
pass
|
|
return text # leave as is
|
|
return re.sub("&#?\w+;", fixup, html)
|
|
|
|
# IDs
|
|
##############################################################################
|
|
|
|
def hexifyID(id):
|
|
return "%x" % id
|
|
|
|
def dehexifyID(id):
|
|
return int(id, 16)
|
|
|
|
def ids2str(ids):
|
|
"""Given a list of integers, return a string '(int1,int2,...)'."""
|
|
return "(%s)" % ",".join(str(i) for i in ids)
|
|
|
|
# Tags
|
|
##############################################################################
|
|
|
|
def parseTags(tags):
|
|
"Parse a string and return a list of tags."
|
|
return [t for t in tags.split(" ") if t]
|
|
|
|
def joinTags(tags):
|
|
"Join tags into a single string, with leading and trailing spaces."
|
|
if not tags:
|
|
return u""
|
|
return u" %s " % u" ".join(tags)
|
|
|
|
def canonifyTags(tags):
|
|
"Strip leading/trailing/superfluous spaces and duplicates."
|
|
tags = [t.lstrip(":") for t in set(tags)]
|
|
return joinTags(sorted(tags))
|
|
|
|
def hasTag(tag, tags):
|
|
"True if TAG is in TAGS. Ignore case."
|
|
return tag.lower() in [t.lower() for t in tags]
|
|
|
|
def addTags(addtags, tags):
|
|
"Add tags if they don't exist."
|
|
currentTags = parseTags(tags)
|
|
for tag in parseTags(addtags):
|
|
if not hasTag(tag, currentTags):
|
|
currentTags.append(tag)
|
|
return canonifyTags(currentTags)
|
|
|
|
def delTags(deltags, tags):
|
|
"Delete tags if they don't exists."
|
|
currentTags = parseTags(tags)
|
|
for tag in parseTags(deltags):
|
|
# find tags, ignoring case
|
|
remove = []
|
|
for tx in currentTags:
|
|
if tag.lower() == tx.lower():
|
|
remove.append(tx)
|
|
# remove them
|
|
for r in remove:
|
|
currentTags.remove(r)
|
|
return canonifyTags(currentTags)
|
|
|
|
# Fields
|
|
##############################################################################
|
|
|
|
def joinFields(list):
|
|
return "\x1f".join(list)
|
|
|
|
def splitFields(string):
|
|
return string.split("\x1f")
|
|
|
|
# Checksums
|
|
##############################################################################
|
|
|
|
def checksum(data):
|
|
return md5(data).hexdigest()
|
|
|
|
def fieldChecksum(data):
|
|
# 32 bit unsigned number from first 8 digits of md5 hash
|
|
return int(checksum(data.encode("utf-8"))[:8], 16)
|
|
|
|
# Temp files
|
|
##############################################################################
|
|
|
|
_tmpdir = None
|
|
|
|
def tmpdir():
|
|
"A reusable temp folder which we clean out on each program invocation."
|
|
global _tmpdir
|
|
if not _tmpdir:
|
|
def cleanup():
|
|
shutil.rmtree(_tmpdir)
|
|
import atexit
|
|
atexit.register(cleanup)
|
|
_tmpdir = os.path.join(tempfile.gettempdir(), "anki_temp")
|
|
try:
|
|
shutil.rmtree(_tmpdir)
|
|
except (IOError, OSError):
|
|
pass
|
|
os.mkdir(_tmpdir)
|
|
return _tmpdir
|
|
|
|
def tmpfile(prefix=None, suffix=None):
|
|
return tempfile.mkstemp(dir=tmpdir(), prefix=prefix, suffix=suffix)
|
|
|
|
def namedtmp(name):
|
|
"Return tmpdir+name. Deletes any existing file."
|
|
path = os.path.join(tmpdir(), name)
|
|
try:
|
|
os.unlink(path)
|
|
except (OSError, IOError):
|
|
pass
|
|
return path
|
|
|
|
# Cmd invocation
|
|
##############################################################################
|
|
|
|
def call(argv, wait=True, **kwargs):
|
|
"Execute a command. If WAIT, return exit code."
|
|
# ensure we don't open a separate window for forking process on windows
|
|
if isWin:
|
|
si = subprocess.STARTUPINFO()
|
|
try:
|
|
si.dwFlags |= subprocess.STARTF_USESHOWWINDOW
|
|
except:
|
|
si.dwFlags |= subprocess._subprocess.STARTF_USESHOWWINDOW
|
|
else:
|
|
si = None
|
|
# run
|
|
try:
|
|
o = subprocess.Popen(argv, startupinfo=si, **kwargs)
|
|
except OSError:
|
|
# command not found
|
|
return -1
|
|
# wait for command to finish
|
|
if wait:
|
|
while 1:
|
|
try:
|
|
ret = o.wait()
|
|
except OSError:
|
|
# interrupted system call
|
|
continue
|
|
break
|
|
else:
|
|
ret = 0
|
|
return ret
|
|
|
|
# OS helpers
|
|
##############################################################################
|
|
|
|
isMac = sys.platform.startswith("darwin")
|
|
isWin = sys.platform.startswith("win32")
|