def merge_sort_stupid(fin: io.BufferedIOBase, fout: io.BufferedIOBase, memory_size: int, left=0, count=None):
fout.seek(0)
if count is None:
count = content_length(fin, preserve_pos=False)
if count <= memory_size:
go_to_pos(fin, left)
write_content(fout, sorted(read_content(fin, count=count)), batch_size=memory_size)
return
with tmp_file() as left_f, tmp_file() as right_f:
merge_sort_stupid(fin, left_f, memory_size, left, count=count // 2)
merge_sort_stupid(fin, right_f, memory_size, left + count // 2, count=count - count // 2)
left_f.seek(0)
right_f.seek(0)
write_content(fout, heapq.merge(read_content(left_f, batch_size=memory_size // 2),
read_content(right_f, batch_size=memory_size // 2)),
batch_size=memory_size)
def _to_sorted_blocks(fin: io.BufferedIOBase, memory_size):
while True:
sorted_values = sorted(read_content(fin, memory_size))
if not sorted_values:
break
f = tmp_file()
write_content(f, sorted_values)
f.close()
yield f
def _check_sorted(self, source: io.BufferedIOBase, res: io.BufferedIOBase):
hashes_size = 2**20
def h(value):
return hash(value) % hashes_size
source.seek(0)
source_content = list(itertools.repeat(0, hashes_size))
for v in read_content(source):
source_content[h(v)] += 1
res.seek(0)
res_content = list(itertools.repeat(0, hashes_size))
prev = None
for cur in read_content(res):
res_content[h(cur)] += 1
self.assertTrue(prev is None or prev <= cur)
prev = cur
self.assertTrue(source_content == res_content, 'Content differs')
def _merge_blocks(tmp_files, fout: io.BufferedIOBase, memory_size: int):
# let's make output buffer slightly larger
# we can use 3 times `memory_size` for buffers
buffer_size = 3 * memory_size // (len(tmp_files) + 2)
for i, f in enumerate(tmp_files):
f = open(f.name, 'a+b')
f.seek(0)
tmp_files[i] = f
generators = [read_content(f, batch_size=buffer_size) for f in tmp_files]
write_content(fout, heapq.merge(*generators), batch_size=2 * buffer_size)
for f in tmp_files:
f.close()
