classification
Title: libffi needs an update to support mips64, arm and armeabi on linux
Type: compile error Stage:
Components: Extension Modules Versions: Python 2.6
process
Status: closed Resolution: fixed
Dependencies: Superseder:
Assigned To: theller Nosy List: doko, theller (2)
Priority: Keywords

Created on 2007-10-18 09:23 by doko, last changed 2008-03-05 14:58 by theller.

Files
File name Uploaded Description Edit Remove
arm.diff doko, 2007-10-24 20:00
Messages (11)
msg56522 - (view) Author: Matthias Klose (doko) Date: 2007-10-18 09:23
libffi needs an update to support mips64, arm and armeabi on linux; the
current bits are all available in GCC trunk. as an alternative, those
systems should be default to --with-system-ffi, unless this is
xplicitely disabled.
msg56726 - (view) Author: Matthias Klose (doko) Date: 2007-10-24 20:00
committed the following patch to the trunk (lets the buildds test the
ctypes extension)
msg56753 - (view) Author: Thomas Heller (theller) Date: 2007-10-25 19:58
I'm unsure how to proceed with this.
Replacing the copy of libffi in the ctypes sources - I'm very afraid to
do that.  It has it's own configure system, written by someone else. 
Then it has several changes from various people for Python, which would
be overwritten.  All these on systems that I do not have direct access
to, and we do not have buildbots for all of them.

So, it looks like --with-system-ffi should be made the default (on
systems that have libffi installed?)

There are also test failures since I committed the c_longdouble patch
(on alpha debian, and S390 Debian, for example).  Maybe newer libffi
versions have a fix for that...

Last comment:  Experimental changes COULD be tested out in a branch, we
would only have to manually trigger the build from the buildbot web pages.
msg57000 - (view) Author: Thomas Heller (theller) Date: 2007-10-31 19:59
In the branches_ctypes-branch I hacked setup.py to always use an
installed libffi if one is found.  Then I triggered the trunk buildbots
which failed or crashed before in some c_longdouble tests; the tests
worked ok on them (ppc Debian unstable, and S-390 Debian).
msg57001 - (view) Author: Thomas Heller (theller) Date: 2007-10-31 20:00
I meant branches/ctypes_branch, of course.
msg57005 - (view) Author: Matthias Klose (doko) Date: 2007-10-31 21:26
oops, did the same for arm* on trunk, but in configure.in
msg57063 - (view) Author: Thomas Heller (theller) Date: 2007-11-02 19:51
I have now made --with-system-ffi default to "yes" for Linux/alpha*,
Linux/arm*, Linux/ppc*, and Linux/s390* machines.  If you think it is
needed for mips machines also please add this - there's no buildbot
where I can check the result.
msg60005 - (view) Author: Thomas Heller (theller) Date: 2008-01-16 20:09
Matthias, can this issue be set to 'fixed' and 'closed' now?
msg62139 - (view) Author: Matthias Klose (doko) Date: 2008-02-07 09:46
there's a buildbot for mips-linux, none for mips64, so lets close this one.
msg62638 - (view) Author: Thomas Heller (theller) Date: 2008-02-21 18:20
I'm reopening this.  libffi 3 has been released; I have created the
libffi3-branch for this work in Python svn.
msg63280 - (view) Author: Thomas Heller (theller) Date: 2008-03-05 14:58
libffi3-branch has been merged to trunk; the Modules/_ctypes/libffi
files are from the libffi 3.0.4 release now.
History
Date User Action Args
2008-03-05 14:58:26thellersetstatus: open -> closed
resolution: fixed
messages: + msg63280
2008-02-21 18:20:11thellersetstatus: closed -> open
messages: + msg62638
2008-02-07 09:46:39dokosetstatus: open -> closed
messages: + msg62139
2008-01-16 20:09:06thellersetmessages: + msg60005
2007-11-02 19:51:11thellersetmessages: + msg57063
2007-10-31 21:26:26dokosetmessages: + msg57005
2007-10-31 20:00:13thellersetmessages: + msg57001
2007-10-31 19:59:14thellersetmessages: + msg57000
2007-10-25 19:58:01thellersetmessages: + msg56753
2007-10-24 20:00:29dokosetfiles: + arm.diff
messages: + msg56726
2007-10-18 09:23:06dokocreate