unable to serialize Infinity or NaN on ARM using marshal

[doko42]

BPO	1762561
Nosy	@tim-one, @loewis, @doko42, @mdickinson, @nirs, @djc
Files	arm-float.diff: patch for 2.5 arm-float2.diff arm-oabi-float-2.7.patch: Same patch for 2.7

^{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 = 'https://github.com/mdickinson'
closed_at = <Date 2009-02-14.10:20:41.510>
created_at = <Date 2007-07-28.08:41:18.000>
labels = ['interpreter-core', 'build']
title = 'unable to serialize Infinity or NaN on ARM using marshal'
updated_at = <Date 2012-04-22.17:16:30.598>
user = 'https://github.com/doko42'

bugs.python.org fields:

activity = <Date 2012-04-22.17:16:30.598>
actor = 'loewis'
assignee = 'mark.dickinson'
closed = True
closed_date = <Date 2009-02-14.10:20:41.510>
closer = 'mark.dickinson'
components = ['Interpreter Core']
creation = <Date 2007-07-28.08:41:18.000>
creator = 'doko'
dependencies = []
files = ['8140', '13076', '24818']
hgrepos = []
issue_num = 1762561
keywords = ['patch']
message_count = 18.0
messages = ['52959', '55628', '70141', '81893', '81914', '81922', '81943', '81945', '81950', '81974', '81983', '81986', '81988', '82003', '155589', '158201', '158962', '158974']
nosy_count = 9.0
nosy_names = ['tim.peters', 'loewis', 'doko', 'mark.dickinson', 'nirs', 'djc', 'aurel32', 'Arfrever', 'mirell']
pr_nums = []
priority = 'normal'
resolution = 'wont fix'
stage = 'patch review'
status = 'closed'
superseder = None
type = 'compile error'
url = 'https://bugs.python.org/issue1762561'
versions = ['Python 2.7']

[doko42]

forwarded from http://bugs.debian.org/434905, patch by
Aurelien Jarno.

python2.5 is unable to serialize a floating point variable which equals Infinity or NaN on ARM using marshal.

The problem is that python2.5 wants to manage the floating points in memory by itself, and assume that they are only little and big endian IEEE formats. Otherwise it fallback to a generic code, which does not handle NaN or Infinity.

The attached patched fixed that by adding mixed-endian IEEE format, as it can be found on ARM old-ABI.

[loewis]

I'm not sure like the naming of the format. "mixed-endian" could mean
anything. I doubt IEEE specifies this as a possible byte representation
(but then, I'm uncertain whether IEEE specifies big-endian and
little-endian, either).

One option would be to call it "ARM, mixed-endian", assuming this can be
only found on ARM.

Another option would be to call it "IEEE,
bytes-big-words-little-endian", describing precisely how the format
works (IIUC).

Tim, any opinion?

[aurel32]

AFAIK, this "mixed-endian" format is only used on little endian ARM
(old-ABI only).

That is true that IEEE 754 does not specify any format. I used the big
and little endian code as a template to add the "ARM format", hence
IEEE in the name. "mixed-endian" is the term usually used to describe
this format in the ARM community. I am not opposed to any other name.

OTOH, if you consider that IEEE does not specify any format, "IEEE,
little-endian" and "IEEE, big-endian" are also not correct.

[mirell]

This still occurs in the latest SVN checkout, and is preventing Python
from building on ARMV4L and ARMV5L OABI, dying during the test_float.py
compilation. The patch appears to solve this problem, however I have not
run a full suite of tests to ensure it doesn't have any regression issues.

Since this has been broken since 2007, could this be looked at in
getting merged?

[mdickinson]

I'll take a look at this, provided Tim doesn't mind me stealing his
issue. (Please steal it back if so.)

Mark, could you please post the output from test_float?