Issue 24550: maxint on 64 bit platforms breaks os.read

Messages (5) msg246063 - (view) Author: Brian Mingus (Brian Mingus) Date: 2015-07-02 05:27

The lower range for this bug may be anything greater than 32 bit maxint. Other modules such as multiprocessing are limited passing objects of size 32 bit maxint, even on 64 bit systems, likely due to this issue. I have demonstrated this by modifying multiprocessing/connection.py to use longs in send and recv, which it surfaces the following error (note that read is os.read).

Traceback (most recent call last):
  File "/usr/lib/python3.4/multiprocessing/process.py", line 254, in _bootstrap
    self.run()
  File "/usr/lib/python3.4/multiprocessing/process.py", line 93, in run
    self._target(*self._args, **self._kwargs)
  File "/usr/lib/python3.4/multiprocessing/pool.py", line 103, in worker
    initializer(*initargs)
  File "/usr/local/lib/python3.4/dist-packages/gensim-0.11.1_1-py3.4-linux-x86_64.egg/gensim/models/ldamulticore.py", line 266, in worker_e_step
    chunk_no, chunk, worker_lda = input_queue.get()
  File "/usr/lib/python3.4/multiprocessing/queues.py", line 96, in get
    res = self._recv_bytes()
  File "/usr/lib/python3.4/multiprocessing/connection.py", line 216, in recv_bytes
    buf = self._recv_bytes(maxlength)
  File "/usr/lib/python3.4/multiprocessing/connection.py", line 420, in _recv_bytes
    return self._recv(size)
  File "/usr/lib/python3.4/multiprocessing/connection.py", line 383, in _recv
    chunk = read(handle, remaining)
OverflowError: signed integer is greater than maximum


Which can be traced back to this cpython code:

https://github.com/python/cpython/blob/3.4/Modules/posixmodule.c#L8048-L8065

msg265403 - (view) Author: Serhiy Storchaka (serhiy.storchaka) * Date: 2016-05-12 12:53

The 32-bit size limit in multiprocessing is issue17560. AFAIK read() and write() now support 64-bit size on 64-bit systems.

msg265438 - (view) Author: Martin Panter (martin.panter) * Date: 2016-05-13 02:31

Yes 3.5 should do 64-bit reads (if you have enough memory) thanks to revision 0c57aba6b1a3 (Argument Clinic conversion):

>>> os.read(0, 2**32)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
MemoryError
>>> os.read(0, 2**63)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
OverflowError: Python int too large to convert to C ssize_t

I understand 3.4 is only open to security fixes, not changes like this. But 2.7 is affected:

>>> os.read(0, 2**32)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
OverflowError: signed integer is greater than maximum

I guess this could be considered a bug in 2.7 (not an enhancement), but I’m not really sure.

msg265445 - (view) Author: Serhiy Storchaka (serhiy.storchaka) * Date: 2016-05-13 05:41

There is more specific issue for 32-bit reads in 2.7 (with patch): issue21199.

I think this issue can be closed as a duplicate of two other issues.

msg266037 - (view) Author: Martin Panter (martin.panter) * Date: 2016-05-22 01:24

Sorry in Python 3.5 the change was actually Issue 21932 (not Arg Clinic). But Victor said that change shouldn’t go into Python 2.

BTW Issue 21199 is about Python 2’s file.read() method, not os.read().

History Date User Action Args 2022-04-11 14:58:18adminsetgithub: 68738 2016-05-22 01:24:06martin.pantersetsuperseder: os.read() must use Py_ssize_t for the size parameter
resolution: duplicate -> wont fix
messages: + msg266037 2016-05-13 05:41:25serhiy.storchakasetstatus: open -> closed
resolution: duplicate
messages: + msg265445

stage: resolved

2016-05-13 02:31:18martin.pantersetnosy: + martin.panter

messages: + msg265438
versions: + Python 2.7, - Python 3.4

2016-05-12 12:53:09serhiy.storchakasetnosy: + vstinner, serhiy.storchaka
messages: + msg265403
2015-07-02 05:27:49Brian Minguscreate