blake2: compile error with -march=bdver2

[floppym]

BPO	29504
Nosy	@tiran, @ned-deily, @floppym
Files	build.log: build log

^{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/tiran'
closed_at = <Date 2017-12-27.19:25:19.367>
created_at = <Date 2017-02-08.17:29:23.149>
labels = ['extension-modules', 'build', '3.7']
title = 'blake2: compile error with -march=bdver2'
updated_at = <Date 2017-12-27.19:25:19.366>
user = 'https://github.com/floppym'

bugs.python.org fields:

activity = <Date 2017-12-27.19:25:19.366>
actor = 'christian.heimes'
assignee = 'christian.heimes'
closed = True
closed_date = <Date 2017-12-27.19:25:19.367>
closer = 'christian.heimes'
components = ['Extension Modules']
creation = <Date 2017-02-08.17:29:23.149>
creator = 'floppymaster'
dependencies = []
files = ['46598']
hgrepos = []
issue_num = 29504
keywords = []
message_count = 8.0
messages = ['287356', '287357', '287406', '287407', '289963', '295655', '309043', '309098']
nosy_count = 5.0
nosy_names = ['christian.heimes', 'ned.deily', 'devurandom', 'floppymaster', 'Joseph Mitzen']
pr_nums = []
priority = 'high'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'compile error'
url = 'https://bugs.python.org/issue29504'
versions = ['Python 3.6', 'Python 3.7']

[floppym]

When compiling python-3.6.0 with -march=bdver2, the blake2 module fails to build.

In file included from /var/tmp/portage/dev-lang/python-3.6.0/work/Python-3.6.0/Modules/_blake2/impl/blake2s-round.h:70:0,
from /var/tmp/portage/dev-lang/python-3.6.0/work/Python-3.6.0/Modules/_blake2/impl/blake2s.c:40,
from /var/tmp/portage/dev-lang/python-3.6.0/work/Python-3.6.0/Modules/_blake2/blake2s_impl.c:32:
/var/tmp/portage/dev-lang/python-3.6.0/work/Python-3.6.0/Modules/_blake2/impl/blake2s-load-xop.h:29:4: error: expected identifier or '(' before 'return'
return _mm_blendv_epi8(t0, s1, mask);
^
/var/tmp/portage/dev-lang/python-3.6.0/work/Python-3.6.0/Modules/_blake2/impl/blake2s-load-xop.h:30:1: error: expected identifier or '(' before '}' token
}/
^
/var/tmp/portage/dev-lang/python-3.6.0/work/Python-3.6.0/Modules/_blake2/impl/blake2s-load-xop.h:30:3: error: expected identifier or '(' before '/' token
}/
^
In file included from /var/tmp/portage/dev-lang/python-3.6.0/work/Python-3.6.0/Modules/_blake2/blake2s_impl.c:32:0:
/var/tmp/portage/dev-lang/python-3.6.0/work/Python-3.6.0/Modules/_blake2/impl/blake2s.c: In function 'blake2s_init0':
/var/tmp/portage/dev-lang/python-3.6.0/work/Python-3.6.0/Modules/_blake2/impl/blake2s.c:167:38: error: 'blake2s_IV' undeclared (first use in this function)
for( i = 0; i < 8; ++i ) S->h[i] = blake2s_IV[i];
^
/var/tmp/portage/dev-lang/python-3.6.0/work/Python-3.6.0/Modules/_blake2/impl/blake2s.c:167:38: note: each undeclared identifier is reported only once for each function it appears in
/var/tmp/portage/dev-lang/python-3.6.0/work/Python-3.6.0/Modules/_blake2/impl/blake2s.c: In function 'PyBlake2_blake2s_init_param':
/var/tmp/portage/dev-lang/python-3.6.0/work/Python-3.6.0/Modules/_blake2/impl/blake2s.c:176:44: error: 'blake2s_IV' undeclared (first use in this function)
const uint8_t * v = ( const uint8_t * )( blake2s_IV );
^

[floppym]

Downstream bug report: https://bugs.gentoo.org/show_bug.cgi?id=608586

[tiran]

It's a bug in Blake2's reference implementation with nested C++ comments (/* /* */ */). BLAKE2/BLAKE2@259e61d#diff-ac1a341ec0ed38dc5d219572f4282287 fixed the bug. For Python 3.6.1 I'm going to update Python's copy of blake2.

[tiran]

I don't have time to fix the issue right away. Python's copy of blake2 needs some manual massaging and tweaking.

[ned-deily]

This didn't make it in time for 3.6.1; deferred to 3.6.2.

[ned-deily]

What's the status of this issue? If it's not going to be fixed prior to 3.6.2rc1, I'm going to downgrade it from "release blocker" as we've already released 3.6.1 without it.

[JosephMitzen]

Is this ever going to get merged? The gentoo bug report link includes a patch that simply removes the nested comments, which is all that is needed to get the code to compile properly on Bulldozer processors. There's no "manual massaging" or anything else necessary to fix the bug. I've tested it out and it works.

[tiran]

The issue has been fixed on Nov 1st in PR #4173 (master) and #4214 (3.6.4).