io.BytesIO doesn't support the buffer protocol

[pitrou]

BPO	5506
Nosy	@birkenfeld, @amauryfa, @pitrou, @benjaminp
Files	bytesiobuf2.patch

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = <Date 2010-09-06.18:50:12.023>
created_at = <Date 2009-03-18.10:40:11.270>
labels = ['type-feature', 'library', 'expert-IO']
title = "io.BytesIO doesn't support the buffer protocol"
updated_at = <Date 2010-09-06.18:50:12.022>
user = 'https://github.com/pitrou'

bugs.python.org fields:

activity = <Date 2010-09-06.18:50:12.022>
actor = 'pitrou'
assignee = 'none'
closed = True
closed_date = <Date 2010-09-06.18:50:12.023>
closer = 'pitrou'
components = ['Library (Lib)', 'IO']
creation = <Date 2009-03-18.10:40:11.270>
creator = 'pitrou'
dependencies = []
files = ['18732']
hgrepos = []
issue_num = 5506
keywords = ['patch']
message_count = 7.0
messages = ['83740', '85504', '85507', '115304', '115476', '115479', '115728']
nosy_count = 4.0
nosy_names = ['georg.brandl', 'amaury.forgeotdarc', 'pitrou', 'benjamin.peterson']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue5506'
versions = ['Python 3.2']

[pitrou]

It may be logical for BytesIO to support the buffer protocol (readable
/and/ writable).

[birkenfeld]

Is this still unimplemented?

[pitrou]

Yes.
I don't remember which use case I was thinking about when I suggested
this, but it may be useful to e.g. write() all the data to a file object
without actually making a copy (getvalue() does).

[pitrou]

Thinking about it, I'm not sure this would be a good idea to do this by default. There is an ambiguity where it's not obvious if the buffer would contain the whole data or only the bytes after the current position.

I think perhaps an explicit method (getbuffer()?) could be used to export a buffer-compatible object, without copying. Perhaps a memoryview; this would imply defining bf_releasebuffer without bf_getbuffer.

[pitrou]

Here is a patch implementing getbuffer(), together with tests.

[pitrou]

There was an unused "weakreflist" member.

[pitrou]

This was committed in r84562.