Can't compile gdb with Python 3.6

[stratakis]

BPO	28898
Nosy	@vstinner, @benjaminp, @ned-deily, @stratakis
Files	fix-gdb-compilation.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 2016-12-16.07:51:07.749>
created_at = <Date 2016-12-07.18:14:39.455>
labels = ['interpreter-core', '3.7']
title = "Can't compile gdb with Python 3.6"
updated_at = <Date 2016-12-16.07:51:07.747>
user = 'https://github.com/stratakis'

bugs.python.org fields:

activity = <Date 2016-12-16.07:51:07.747>
actor = 'ned.deily'
assignee = 'none'
closed = True
closed_date = <Date 2016-12-16.07:51:07.749>
closer = 'ned.deily'
components = ['Interpreter Core']
creation = <Date 2016-12-07.18:14:39.455>
creator = 'cstratak'
dependencies = []
files = ['45791']
hgrepos = []
issue_num = 28898
keywords = ['3.6regression']
message_count = 10.0
messages = ['282649', '282695', '282696', '282713', '282753', '282754', '282768', '282772', '283378', '283381']
nosy_count = 5.0
nosy_names = ['vstinner', 'benjamin.peterson', 'ned.deily', 'python-dev', 'cstratak']
pr_nums = []
priority = None
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue28898'
versions = ['Python 3.6', 'Python 3.7']

[stratakis]

Trying to compile gdb, with python support and by having it depend on Python 3.6 produces an error that the HAVE_LONG_LONG has been redefined [0].

This seems to have been introduced by this commit [1].

I'm in no way expert on gdb, but from what I could deduct from the build logs, the #define HAVE_LONG_LONG line in pyport.h [2] is the issue, as gdb also sets the same constant to 1 [3][4]

I build python and gdb as rpm's, but I can't verify if it happens as well outside this environment. Could someone more experienced with gdb take a look at it?

The issue seems to be fixed by wrapping the #define HAVE_LONG_LONG with an #ifndef HAVE_LONG_LONG

Attaching a patch for consideration

[0] https://copr-be.cloud.fedoraproject.org/results/cstratak/python3.6/fedora-26-x86_64/00485858-gdb/build.log.gz

[1] https://hg.python.org/cpython/rev/cf6e9968ebb7/

[2] https://hg.python.org/cpython/file/tip/Include/pyport.h#l42

[3] https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=blob;f=gdb/config.in;h=3f8a72326b5d32db6ad8966c05045e248f955498;hb=HEAD#l282

[4] https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=blob;f=gdb/configure;h=6df88d9049b546a9e906b943f67374dc90d90d6d;hb=HEAD#l11662

[python-dev]

New changeset 4745d801cae2 by Benjamin Peterson in branch '3.6':
guard HAVE_LONG_LONG definition to prevent redefinition (bpo-28898)
https://hg.python.org/cpython/rev/4745d801cae2

New changeset 1afc3f4f5502 by Benjamin Peterson in branch 'default':
merge 3.6 (bpo-28898)
https://hg.python.org/cpython/rev/1afc3f4f5502

[benjaminp]

I think we should take this for 3.6.

[vstinner]

The pushed change 4745d801cae2 LGTM for a cherry-pick in 3.6.0.

[vstinner]

Oh, the problem has introduced a crash on OpenIndiana:
---
Fatal Python error: Aborted

Current thread 0x0000000000000001 (most recent call first):
File "/export/home/buildbot/64bits/3.x.cea-indiana-amd64/build/Lib/ctypes/init.py", line 273 in _reset_cache
...
---

http://buildbot.python.org/all/builders/AMD64%20OpenIndiana%203.x/builds/37/steps/test/logs/stdio

[vstinner]

The OpenIndiana uses gcc and has the long long type. Extract of the ./configure script:
---
checking for gcc... gcc
(...)
checking size of long long... 8
---
http://buildbot.python.org/all/builders/AMD64%20OpenIndiana%203.x/builds/37/steps/configure/logs/stdio

[benjaminp]

Why do you think this change caused that breakage? That bot looks like it's been broken for days. This change should have no behavior change whatsoever for Python itself.

[vstinner]

Why do you think this change caused that breakage? That bot looks like it's been broken for days. This change should have no behavior change whatsoever for Python itself.

It's right that the buildbot is broken for years:
https://mail.python.org/pipermail/python-dev/2016-September/146538.html

But the error changed, since the build 37, tests started to fail with a very specific error: "Fatal Python error: Aborted". The build 37 contains your change 1afc3f4f5502 and two other changes which only modify the Doc/ directory.

Sorry, I don't know more :-/ Maybe it's a false alarm, but it's worth to investigate.

[python-dev]

New changeset d6c8803c55b4 by Benjamin Peterson in branch '3.6':
guard HAVE_LONG_LONG definition to prevent redefinition (bpo-28898)
https://hg.python.org/cpython/rev/d6c8803c55b4

New changeset 9d0765c22bed by Ned Deily in branch '3.6':
Issue bpo-28898: add Misc/NEWS entry
https://hg.python.org/cpython/rev/9d0765c22bed

[ned-deily]

[cherrypicked for 3.6.0rc2]