Fix misalignment in fastsearch_memchr_1char

[andreas-schwab]

BPO	19537
Nosy	@loewis, @pitrou, @vstinner, @larryhastings, @skrah, @serhiy-storchaka, @andreas-schwab
Files	pyasciiobject-align.patch: Add explicit padding to PyASCIIObject

^{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 2014-03-23.21:57:20.154>
created_at = <Date 2013-11-09.16:32:08.481>
labels = ['interpreter-core', 'type-bug']
title = 'Fix misalignment in fastsearch_memchr_1char'
updated_at = <Date 2014-03-24.22:08:05.478>
user = 'https://github.com/andreas-schwab'

bugs.python.org fields:

activity = <Date 2014-03-24.22:08:05.478>
actor = 'BreamoreBoy'
assignee = 'none'
closed = True
closed_date = <Date 2014-03-23.21:57:20.154>
closer = 'pitrou'
components = ['Interpreter Core']
creation = <Date 2013-11-09.16:32:08.481>
creator = 'schwab'
dependencies = []
files = ['34589']
hgrepos = []
issue_num = 19537
keywords = ['patch']
message_count = 28.0
messages = ['202487', '202537', '202541', '202551', '202554', '202573', '204358', '204367', '204403', '206249', '206250', '206253', '206254', '206285', '206311', '214603', '214634', '214635', '214639', '214640', '214651', '214652', '214673', '214674', '214675', '214728', '214735', '214745']
nosy_count = 9.0
nosy_names = ['loewis', 'pitrou', 'vstinner', 'larry', 'skrah', 'BreamoreBoy', 'python-dev', 'serhiy.storchaka', 'schwab']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue19537'
versions = ['Python 3.3', 'Python 3.4']

[andreas-schwab]

sizeof(foo) is never a good approximation for the alignment of foo

[serhiy-storchaka]

Current code assumes that PyUnicode_DATA() is aligned to PyUnicode_KIND() bytes. If this is not true on some platform it will be easer to add a padding than rewrite a code. A lot of code depends on this assumption. See also bpo-14422.

I afraid that proposed patch may slow down a search.

[pitrou]

I don't understand in which concrete situation the current code could be wrong. The start of the unicode string should always be aligned, due to how unicode objects are allocated.

[andreas-schwab]

(gdb) p sizeof(PyASCIIObject)
$1 = 22

[skrah]

Andreas Schwab <report@bugs.python.org> wrote:

(gdb) p sizeof(PyASCIIObject)
$1 = 22

m68k again? ;)

[andreas-schwab]

There is nothing wrong with that.

[skrah]

(gdb) p sizeof(PyASCIIObject)
$1 = 22

If the only issue is that the size should be 24 instead of 22,
could we perhaps add an unused uint16_t to the struct just
for this architecture?

[serhiy-storchaka]

What is sizeof(PyUnicodeObject)?

[andreas-schwab]

(gdb) p sizeof(PyUnicodeObject)
$1 = 38

[serhiy-storchaka]

Antoine, if sizeof(PyUnicodeObject) is 38, the start of UCS4 unicode string is not aligned. This should be fixed.

[pitrou]

Antoine, if sizeof(PyUnicodeObject) is 38, the start of UCS4 unicode
string is not aligned. This should be fixed.

Probably. Does anyone want to propose a patch?

[andreas-schwab]

How about adding explicit padding to the bitfield in PyASCIIObject?

[pitrou]

How about adding explicit padding to the bitfield in PyASCIIObject?

That sounds ok to me, but it must be explicitly for 68k (or use a
sufficiently smart scheme not to affect already aligned architectures).