Hacked By AnonymousFox
class Format(object):
__slots__ = 'impl', 'mod'
def __init__(self, impl):
self.impl = impl
self.mod = impl.mod
def get_formatted_row(self, row):
fr = self.get_stat_data(row)
rows = []
rs = row.name.split('\n')
subsequent_indent = len(fr)*' '
rows.extend(self.mod.wrap(
fr+rs[0],
width=self.mod.line_length,
subsequent_indent=subsequent_indent))
for r in rs[1:]:
rows.extend(self.mod.wrap(
r,
width=self.mod.line_length,
initial_indent=subsequent_indent,
subsequent_indent=subsequent_indent))
return '\n'.join(rows)
def load_statrow_csk(self, r):
impl = self.impl
count, size, kind = r.split(' ', 2)
count = int(count)
size = int(size)
impl.cum_size += size
return StatRow(count, size, kind, impl.cur_index, impl.cum_size)
def load_statrow_sk(self, r):
impl = self.impl
size, kind = r.split(' ', 1)
size = int(size)
impl.cum_size += size
return StatRow(1, size, kind, impl.cur_index, impl.cum_size)
def _oh_get_num_lines(self):
return self.impl.numrows
def _oh_get_label(self):
return self.get_label()
def _oh_get_row_header(self):
impl = self.impl
if not (impl.count or impl.size):
return ''
sh = self.get_stat_header()
return self.mod.fill(
sh + self.impl.kindheader,
width=self.mod.line_length,
subsequent_indent=' '*len(sh))
def _oh_get_more_msg(self, start_lineno, end_lineno):
nummore = self.impl.numrows-(end_lineno+1)
return "<%d more rows. Type e.g. '_.more' to view.>" % nummore
def _oh_get_line_iter(self):
for numrows, row in enumerate(self.impl.get_rows()):
yield self.get_formatted_row(row)
class SetFormat(Format):
__slots__ = ()
def get_label(self):
impl = self.impl
if impl.count != 1:
s = 's'
else:
s = ''
return 'Partition of a set of %d object%s. Total size = %d bytes.' % (
impl.count, s, impl.size)
def get_rowdata(self, row):
return '%d %d %s' % (row.count, row.size, row.name)
def get_stat_header(self):
return (
' Index Count % Size % Cumulative % ')
def get_stat_data(self, row):
format = '%6d %6d %3d %8d %3d %9d %3d '
impl = self.impl
fr = format % (
row.index,
row.count, int('%.0f' % (row.count * 100.0/impl.count)),
row.size, int('%.0f' % (row.size * 100.0/impl.size)),
row.cumulsize, int('%.0f' % (row.cumulsize * 100.0/impl.size)),
)
return fr
def load_statrow(self, r):
return self.load_statrow_csk(r)
class IdFormat(Format):
__slots__ = ()
def get_label(self):
impl = self.impl
if impl.count != 1:
s = 's'
else:
s = ''
return (
'Set of %d %s object%s. Total size = %d bytes.' % (
impl.count, impl.kindname, s, impl.size))
return part
def get_rowdata(self, row):
return '%d %s' % (row.size, row.name)
def get_stat_header(self):
return (
' Index Size % Cumulative % ')
def get_stat_data(self, row):
impl = self.impl
format = '%6d %8d %5.1f %9d %5.1f '
fr = format % (
row.index,
row.size, (row.size * 100.0/impl.size),
row.cumulsize, row.cumulsize * 100.0/impl.size,
)
return fr
def load_statrow(self, r):
return self.load_statrow_sk(r)
class DiffFormat(Format):
__slots__ = ()
def _percent_of_b(self, size):
if self.impl.b_size != 0:
return '%9.3g' % (size*100.0/self.impl.b_size,)
else:
return ' (n.a.)'
def get_label(self):
impl = self.impl
x = (
'Summary of difference operation (A-B).\n' +
' Count Size\n' +
' A %6d %8d\n' % (impl.count+impl.b_count, impl.size+impl.b_size) +
' B %6d %8d\n' % (impl.b_count, impl.b_size) +
' A-B %6d %8d = %s %% of B\n' % (impl.count, impl.size, self._percent_of_b(impl.size)))
if impl.count or impl.size:
x += '\nDifferences by kind, largest absolute size diffs first.'
return x
def get_rowdata(self, row):
return '%d %d %s' % (row.count, row.size, row.name)
def get_stat_header(self):
return (
' Index Count Size Cumulative % of B ')
def get_stat_data(self, row):
impl = self.impl
format = '%6d %6d %8d %9d %s '
fr = format % (
row.index,
row.count,
row.size,
row.cumulsize,
self._percent_of_b(row.cumulsize),
)
return fr
def load_statrow(self, r):
return self.load_statrow_csk(r)
class StatRow(object):
__slots__ = 'count', 'size', 'name', 'index', 'cumulsize'
def __init__(self, count, size, name, index=None, cumulsize=None):
self.count = count
self.size = size
self.name = name
self.index = index
self.cumulsize = cumulsize
class PartRow(StatRow):
__slots__ = 'set', 'kind'
def __init__(self, count, size, name, index, cumulsize, set, kind):
self.count = count
self.size = size
self.name = name
self.index = index
self.cumulsize = cumulsize
self.set = set
self.kind = kind
class Stat:
def __init__(self, mod, get_trows, firstheader=''):
self.mod = mod
self._hiding_tag_ = mod._hiding_tag_
self.get_trows = get_trows
self.firstheader = firstheader
self.it = iter(get_trows())
self.cur_index = 0
self.cum_size = 0
self.rows = []
r = self.get_next()
while r and not r.startswith('.r:'):
name = r[1:r.index(':')]
value = r[r.index(':')+1:].strip()
try:
value = int(value)
except ValueError:
pass
setattr(self, name, value)
r = self.get_next()
self.format_name = self.format
self.format_class = getattr(self.mod, self.format)
self.format = self.format_class(self)
self.timemade = float(self.timemade)
self.mod.OutputHandling.setup_printing(self, self.format)
def __getitem__(self, idx):
if isinstance(idx, int):
if idx < 0:
idx = self.numrows + idx
if not (0 <= idx < self.numrows):
raise IndexError('Stat index out of range.')
rows = [self.get_row(idx)]
elif isinstance(idx, slice):
start, stop, step = idx.indices(self.numrows)
rows = [self.get_row(idx) for idx in range(start, stop, step)]
else:
raise IndexError('Stat indices must be integers or slices.')
count = 0
size = 0
for r in rows:
count += r.count
size += r.size
trows = [
'.loader: _load_stat',
'.format: %s' % self.format_name,
'.timemade: %f' % self.timemade,
'.count: %d' % count,
'.size: %d' % size,
'.kindheader: %s' % self.kindheader,
'.kindname: %s' % self.kindname,
'.numrows: %d' % len(rows),
]
if getattr(self, 'b_count', None) is not None:
trows.append('.b_count: %d' % self.b_count)
trows.append('.b_size: %d' % self.b_size)
for r in rows:
trows.append('.r: %s' % self.format.get_rowdata(r))
return self.mod.load(trows)
def __len__(self):
return self.numrows
def __sub__(self, other):
if not isinstance(other, Stat):
raise TypeError(
'Can only take difference with other Stat instance.')
if self.kindheader != other.kindheader:
raise ValueError('Mismatching table kind header, %r vs %r.' % (
self.kindheader, other.kindheader))
rows = []
otab = {}
stab = {}
for r in other.get_rows():
o = otab.get(r.name)
if o:
otab[r.name] = StatRow(
r.count+o.count, r.size+o.size, r.name, o.index, None)
else:
otab[r.name] = r
for r in self.get_rows():
o = otab.get(r.name)
if o:
del otab[r.name]
count = r.count - o.count
size = r.size - o.size
else:
count = r.count
size = r.size
if count == 0 and size == 0:
continue
sr = stab.get(r.name)
if sr:
sr.count += count
sr.size += size
else:
sr = StatRow(count, size, r.name)
stab[sr.name] = sr
rows.append(sr)
rs = list(otab.values())
rs.sort(key=lambda x: x.index) # Preserve orig. order
for r in rs:
sr = StatRow(-r.count, -r.size, r.name)
assert sr.name not in stab
rows.append(sr)
rows.sort(key=lambda x: abs(x.size), reverse=True)
cumulcount = 0
cumulsize = 0
for r in rows:
cumulcount += r.count
cumulsize += r.size
r.cumulsize = cumulsize
trows = [
'.loader: _load_stat',
'.format: DiffFormat',
'.timemade: %f' % self.mod.time.time(),
'.b_count: %d' % other.count,
'.b_size: %d' % other.size,
'.count: %d' % cumulcount,
'.size: %d' % cumulsize,
'.kindheader: %s' % self.kindheader,
'.kindname: %s' % self.kindname,
'.numrows: %d' % len(rows),
]
for r in rows:
trows.append('.r: %d %d %s' % (r.count, r.size, r.name))
return self.mod.load(trows)
def dump(self, fn, mode='a'):
if not hasattr(fn, 'write'):
f = open(fn, mode)
else:
f = fn
try:
for r in self.get_trows():
if not r[-1:] == '\n':
r += '\n'
f.write(r)
end = '.end: .loader: %s\n' % self.loader
if r != end:
f.write(end)
finally:
if f is not fn:
f.close()
def get_next(self):
try:
r = next(self.it)
except StopIteration:
r = None
else:
r = r.rstrip('\n')
self.last = r
return r
def get_row(self, idx):
while idx >= len(self.rows):
self.parse_next_row()
return self.rows[idx]
def get_rows(self, idx=0):
while idx < self.numrows:
try:
row = self.get_row(idx)
except IndexError:
return
else:
yield row
idx += 1
def get_rows_of_kinds(self, kinds):
# Return the rows with names in sequence kinds of unique names
# in that order. None if no such kind.
kindtab = {}
N = len(kinds)
res = [None] * len(kinds)
for i, kind in enumerate(kinds):
kindtab[kind] = i
assert len(kindtab) == N
n = 0
for row in self.get_rows():
idx = kindtab.get(row.name)
if idx is not None:
res[idx] = row
n += 1
if n >= N:
break
return res
def get_rows_n_and_other(self, N, sortby='Size'):
# Get N rows, the largest first
# mix in an '<Other>' row at a sorted position
# Size is either size if sortby = 'Size',
# or count if sortby = 'Count'.
# Returns a NEW LIST (caller may modify/sort it)
if sortby not in ('Size', 'Count'):
raise ValueError("Argument 'sortby' must be 'Size' or 'Count'.")
# Rows are already sorted by Size, largest first.
# If they want by Count, we need to resort them.
rows = self.get_rows()
if sortby == 'Count':
rows = list(rows)
rows.sort(key=lambda x: x.count, reverse=True)
retrows = []
cumulcount = 0
cumulsize = 0
for (i, r) in enumerate(rows):
if i >= N:
othercount = self.count - cumulcount
othersize = self.size - cumulsize
other = StatRow(othercount,
othersize,
'<Other>')
if sortby == 'Size':
for (i, r) in enumerate(retrows):
if r.size < othersize:
retrows[i:i] = [other]
break
else:
retrows.append(other)
elif sortby == 'Count':
for (i, r) in enumerate(retrows):
if r.count < othercount:
retrows[i:i] = [other]
break
else:
retrows.append(other)
else:
assert 0
break
cumulcount += r.count
cumulsize += r.size
retrows.append(r)
else:
assert cumulcount == self.count
assert cumulsize == self.size
return retrows
def parse_next_row(self):
r = self.last
if not r:
raise IndexError('Row index out of range.')
if r.startswith('.r: '):
r = r[4:]
sr = self.format.load_statrow(r)
self.cur_index += 1
self.rows.append(sr)
self.get_next()
return
elif r.startswith('.end'):
raise IndexError('Row index out of range.')
else:
raise SyntaxError
class Partition:
def __init__(self, mod, set, er):
self.mod = mod
self.set = set
self.er = er
self._hiding_tag_ = mod._hiding_tag_
self.timemade = mod.time.time()
def __iter__(self):
# The default iteration is over the sets
# To iterate over rows (if more info is needed), get_rows() is available.
return self.get_sets()
def get_rows(self, rowindex=None):
# Iterator over rows
if rowindex is None:
rowindex = 0
while 1:
try:
row = self.get_row(rowindex)
except IndexError:
return
else:
yield row
rowindex += 1
def get_set(self, index):
if isinstance(index, slice):
start, stop, step = index.indices(self.numrows)
ns = self.get_nodeset(start, stop, step)
return self.mod.idset(ns, er=self.er)
else:
if index < 0:
index += self.numrows
return self.get_rowset(index)
def get_sets(self, index=None):
for idx in range(self.numrows):
yield self.get_rowset(idx)
def get_stat(self):
# Avoid any references into the set!
trows = list(self.get_trows())
def get_trows():
return trows
return self.mod._load_stat(get_trows)
def get_trows(self):
yield '.loader: _load_stat'
yield '.format: %s' % self.format.__class__.__name__
yield '.timemade: %f' % self.timemade
yield '.count: %d' % self.count
yield '.size: %d' % self.size
yield '.kindname: %s' % self.kindname
yield '.kindheader: %s' % self.kindheader
yield '.numrows: %d' % self.numrows
for row in self.get_rows():
yield '.r: %s' % self.format.get_rowdata(row)
def init_format(self, FormatClass):
self.format = FormatClass(self)
self.mod.OutputHandling.setup_printing(self, self.format)
class IdentityPartitionCluster(object):
# Contains objects of same size.
# to speed up management of identity partition
# - since otherwise we'd have to sort all the objects,
# on their string representation in worst case.
__slots__ = 'objects', 'locount', 'hicount', 'losize', 'obsize', 'issorted'
def __init__(self, objects, locount, count, losize, obsize):
self.objects = objects # tuple of objects in this segment
self.locount = locount # count BEFORE objects in this cluster
self.hicount = locount+count # count AFTER these objects
self.losize = losize # size BEFORE objects in this cluster
self.obsize = obsize # size of EACH object in this segment
self.issorted = False # indicates if .objects is sorted
class IdentityPartition(Partition):
def __init__(self, mod, set, er):
Partition.__init__(self, mod, set, er)
clusters = []
sizeclasses = mod.Size.classifier.partition_cli(set.nodes)
sizeclasses.sort()
sizeclasses.reverse()
totcount = 0
totsize = 0
for size, v in sizeclasses:
count = len(v)
clusters.append(IdentityPartitionCluster(
self.mod.observation_list(v), totcount, count, totsize, size))
totsize += size * count
totcount += count
assert totcount == set.count
self.cluidx = 0
self.clusters = clusters
self.count = totcount
self.kind = kind = set.byclodo.kind
self.kindheader = kind.fam.c_get_idpart_header(kind)
self.kindname = kind.fam.c_get_idpart_label(kind)
self.numrows = totcount
self.render = kind.fam.c_get_idpart_render(kind)
self.size = totsize
self.sortrender = kind.fam.c_get_idpart_sortrender(kind)
self.init_format(IdFormat)
def get_nodeset(self, start, stop, step):
return self.get_nodeset_cluster(start, stop, step)[0]
def get_nodeset_cluster(self, start, stop, step):
if step <= 0:
raise ValueError('Step must be positive.')
ns = self.mod.mutnodeset()
if start >= stop:
return (ns, None)
clusters = self.clusters
lo = 0
hi = len(clusters)
cluidx = self.cluidx
while lo < hi:
clu = clusters[cluidx]
if clu.locount <= start:
if start < clu.hicount:
break
else:
lo = cluidx + 1
else:
hi = cluidx
cluidx = (lo + hi) // 2
else:
return (ns, None)
clu_to_return = clu
while 1:
objects = clu.objects
if start != clu.locount or stop < clu.hicount or step != 1:
if not clu.issorted:
sortrender = self.sortrender
if sortrender == 'IDENTITY':
ks = objects
else:
ks = [sortrender(x) for x in objects]
ks = [(kind, i) for i, kind in enumerate(ks)]
ks.sort()
clu.objects = objects = self.mod.observation_list(
[objects[i] for (kind, i) in ks])
clu.issorted = True
objects = objects[start-clu.locount:stop-clu.locount:step]
ns |= objects
self.cluidx = cluidx # memo till next call
start += len(objects)*step
if start >= stop:
break
for cluidx in range(cluidx + 1, len(clusters)):
clu = clusters[cluidx]
if clu.locount <= start < clu.hicount:
break
else:
break
return (ns, clu_to_return)
def get_row(self, rowidx):
ns, clu = self.get_nodeset_cluster(rowidx, rowidx+1, 1)
if not ns:
raise IndexError('Partition index out of range.')
vi = self.mod.idset(ns, er=self.er)
row = PartRow(1, clu.obsize, self.render(vi.theone),
rowidx, (rowidx+1-clu.locount)*clu.obsize + clu.losize,
vi, vi.kind)
return row
def get_rowset(self, rowidx):
ns = self.get_nodeset(rowidx, rowidx+1, 1)
if not ns:
raise IndexError('Partition index out of range.')
return self.mod.idset(ns, er=self.er)
class SetPartition(Partition):
def __init__(self, mod, set, er):
Partition.__init__(self, mod, set, er)
classifier = er.classifier
tosort = [(-part.size, classifier.get_tabrendering(kind, ''), kind, part)
for (kind, part) in classifier.partition(set.nodes)]
tosort.sort()
cumulsize = 0
rows = []
for (minusize, name, kind, part) in tosort:
size = -minusize
cumulsize += size
rows.append(PartRow(
part.count, size, name,
len(rows), cumulsize,
part, kind))
self.count = set.count
self.kindheader = classifier.get_tabheader('')
self.kindname = ''
self.numrows = len(rows)
self.rows = rows
self.size = cumulsize
self.init_format(SetFormat)
def get_nodeset(self, start, stop, step):
if step <= 0:
raise ValueError('Step must be positive.')
ns = self.mod.mutnodeset()
while start < stop:
ns |= self.rows[start].set.nodes
start += step
return ns
def get_row(self, idx):
try:
return self.rows[idx]
except IndexError:
raise IndexError('Partition index out of range.')
def get_rowset(self, idx):
row = self.get_row(idx)
set = row.set
try:
set._partition
except AttributeError:
set._partition = self.mod._fast_partition(self, row)
return set
class FastPartition(SetPartition):
def __init__(self, mod, parent, row):
Partition.__init__(self, mod, parent.set, parent.er)
self.count = row.count
self.kindheader = parent.kindheader
self.kindname = ''
self.numrows = 1
self.rows = [row]
self.size = row.size
self.init_format(SetFormat)
class _GLUECLAMP_:
_preload_ = ('_hiding_tag_',)
_chgable_ = ('line_length', 'backup_suffix')
_imports_ = (
'_parent.ImpSet:mutnodeset',
'_parent:OutputHandling',
'_parent.Use:Id',
'_parent.Use:Size',
'_parent.Use:idset',
'_parent.Use:load',
'_parent.View:_hiding_tag_',
'_parent.View:observation_list',
'_root.os:rename',
'_root.textwrap:fill',
'_root.textwrap:wrap',
'_root.textwrap:wrap',
'_root:time',
)
# 'Config'
line_length = 100
backup_suffix = '.old'
# Factory method
def partition(self, set, er):
if er.classifier is self.Id.classifier:
return IdentityPartition(self, set, er)
else:
return SetPartition(self, set, er)
# Private - Use.load is intended to be used directly.
def _load_stat(self, get_trows):
return Stat(self, get_trows)
def _fast_partition(self, parent, row):
return FastPartition(self, parent, row)
Hacked By AnonymousFox1.0, Coded By AnonymousFox