Don't use AC_RUN_IFELSE to determine float endian

[rossburton]

BPO	34585
Nosy	@tim-one, @mdickinson, @benjaminp, @rossburton, @yan12125
PRs	[RFC] bpo-34585: Don't do runtime test to get float byte order #9085
Files	0001-Don-t-do-runtime-test-to-get-float-byte-order.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 2018-09-19.06:25:56.225>
created_at = <Date 2018-09-05.10:57:22.362>
labels = ['build', '3.8']
title = "Don't use AC_RUN_IFELSE to determine float endian"
updated_at = <Date 2018-09-19.06:49:08.815>
user = 'https://github.com/rossburton'

bugs.python.org fields:

activity = <Date 2018-09-19.06:49:08.815>
actor = 'benjamin.peterson'
assignee = 'none'
closed = True
closed_date = <Date 2018-09-19.06:25:56.225>
closer = 'benjamin.peterson'
components = ['Build']
creation = <Date 2018-09-05.10:57:22.362>
creator = 'rossburton'
dependencies = []
files = ['47788']
hgrepos = []
issue_num = 34585
keywords = ['patch']
message_count = 3.0
messages = ['324627', '325705', '325708']
nosy_count = 5.0
nosy_names = ['tim.peters', 'mark.dickinson', 'benjamin.peterson', 'rossburton', 'yan12125']
pr_nums = ['9085']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue34585'
versions = ['Python 3.8']

[rossburton]

Currently configure.ac uses AC_RUN_IFELSE to determine the byte order of floats and doubles. This hurts when cross-compiling because a default is set, resulting in Python silently falling back to sub-optimal codepaths.

A partial improvement would be to not set a fallback, to force the user to set the right byte order explicitly.

However this test can be done without running anything. autoconf-archive has a macro that uses a carefully constructed double that encodes to an ASCII string in the binary, which is then examined using grep. Evil genius.

Attached is a POC using this. Currently the autoconf-archive macro only handles big and little endian not the ARM OABI mixed-endian format, so configure.ac assumes if the byte order is unknown then it's the crazy mixed-endian.

To be honest OABI is so old now, I don't believe anyone actually uses it anymore: everyone still on ARMv4 should have moved to EABI many years ago, and the mixed-endian support could be removed from Py3 in the future.

[benjaminp]

New changeset 2a9c380 by Benjamin Peterson (Ross Burton) in branch 'master':
closes bpo-34585: Don't do runtime test to get float byte order. (GH-9085)
2a9c380

[benjaminp]

New changeset b3b8cb4 by Benjamin Peterson in branch 'master':
run autoconf (GH-9411)
b3b8cb4

[libTorrentUser]

Unfortunately ax_c_float_words_bigendian.m4 evil magical trickery does not work if certain compiler flags are used. Using gcc and -flto seems to be enough to break it.