In certain situations, file references (.py[co]) for Python files that
fail to compile with compileall() are still added to distutils --record
output.
This output is used for pkg-plist generation and must only contain
references to files that will be installed.
One example of a failure condition is when a Python 2/3 compatible
package containing a file containing Python 3.x only code is built with
Python 2.x, such as Gunicorn's _gaiohttp.py [1]
This change backports patches submitted against upstream issue 20397 [2]
that has not yet been committed.
- For Python 2.7 and 3.5, backport both install_lib and test
- For Python 3.2, 3.3 and 3.4, only backport install_lib
[1] https://svnweb.freebsd.org/changeset/ports/404558
[2] https://bugs.python.org/issue20397
Thank you to Brendan Molloy for producing and submitting the patches
against upstream sources.
Reviewed by: sbz (python)
MFH: 2016Q1
Differential Revision: D4832
There are some inefficiencies in python.mk that significantly slow down
full tree scanning. The use of bmake to obtain the current version of
a specific python is responsible for the majority of the slow done.
This commit splits out the PYTHON_PORTVERSION definition (which is the
same as the lang/python* PORTVERSION) into separate files. With this
change, python.mk can simple include the makefile fragment instead of
spawning a new instance of make.
Different Revision: https://reviews.freebsd.org/D4660
Approved by: antoine (python), mva (python)
non-default Python versions:
- Add pyXY-{sqlite3,gdbm,tkinter} ports for generating binary packages
- Improve/add pkg-message to point users to install respective packages of
separated Python standard modules
- Add COMMENT to explicitly show the Python version that package should be
used with
- Simplify version-related PYTHON_* for lang/python35
Reviewed by: koobs
Differential Revision: https://reviews.freebsd.org/D4170
In Python 3.4+, upstream added and switched to using a shell
implementation of the python-config script [1]. The Python
implementation (python-config.py) remained used by all versions < 3.4.
While the shell implementation returns the path to the Python
shared library when using the --ldflags script argument, the Python
implementation of the script does not. The bug has been reported, but
has not yet been merged [2].
The Python ports currently default to including ${LOCALBASE}/lib
in LIBS when the NLS option is enabled (which it is by default).
When built *with* NLS (gettext) support, the flags added to LIBS
are returned in `pythonX.Y-config --ldflags` output, which happens
to match the path to the Python shared library.
If the NLS option is disabled, ${LOCALBASE}/lib is not added to LIBS,
and are therefore not returned in --ldflags output.
This results in potential linking errors for software that uses
python-config to obtain the correct library path, when the NLS option is
disabled:
$ make WITH=PYTHON -C audio/alsa-lib
[...]
--- smixer-python.la ---
CCLD smixer-python.la
/usr/bin/ld: cannot find -lpython2.7
This change modifies the python-config.in script to match the shell
implementation, outputting the library path in --ldflags output.
While I'm here:
for Python 3.2 and Python 3.3 ports, backport a library order
change [3]. This could affect linking with static libraries.
Use standard length lines and reduce diffs in pkg-message
[1] https://bugs.python.org/issue16235
[2] https://bugs.python.org/issue7352
[2] https://bugs.python.org/issue18096
PR: 197757
Submitted by: jbeich
MFH: 2015Q4
Parser/pgen code intermittently and non-deterministically fails
at build time causing errors including, among others:
* Parser/pgen.o: file not recognized: File truncated
* pgenmain.c:(.text+0x244): undefined reference to `_Py_pgen'
This is apparently due to incorrect uses of recursive make [1] which
was fixed in the upstream Python 'default' (3.5) branch [2].
This change marks all Python port versions as MAKE_JOBS_UNSANFE until
the the original changeset [1] and the resulting regression in
cross-builds [3], whos fix is still pending, can be backported.
[1] https://bugs.python.org/issue22359
[2] https://hg.python.org/cpython/rev/c2a53aa27cad
[3] https://bugs.python.org/issue22625
PR: 200622
Reported by: marino
MFH: 2015Q3
- Replace ${MASTER_SITE_FOO} with FOO.
- Merge MASTER_SITE_SUBDIR into MASTER_SITES when possible. (This means 99.9%
of the time.)
- Remove occurrences of MASTER_SITE_LOCAL when no subdirectory was present and
no hint of what it should be was present.
- Fix some logic.
- And generally, make things more simple and easy to understand.
While there, add magic values to the FESTIVAL, GENTOO, GIMP, GNUPG, QT and
SAMBA macros.
Also, replace some EXTRACT_SUFX occurences with USES=tar:*.
Checked by: make fetch-urlall-list
With hat: portmgr
Sponsored by: Absolight
As per lang/python27 (r377581):
- Add BROKEN for i386 without LIBFFI option, and add upstream
issue references.
While I'm here, clean up after the LIBFFI option addition:
- Sort options variables: OPTIONS_* and *_DESC
- Use OPTIONS helpers
- Reduce diffs between lang/python* ports
Remove patches and hacks that were used to work around the previous
situation
This allows to stage more ports as a regular user
Differential Revision: https://reviews.freebsd.org/D703
Reviewed by and discussed with: bapt
With hat: portmgr
Bump PORTREVISION, as prior to the previous change that fixed upstream Issue
#21166, Python builds could be potentially polluted by the host environment
which could caused runtime issues because of ABI differences.
Backport fix for upstream Issue #21166:
Prevent possible segfaults and other random failures of python
--generate-posix-vars in pybuilddir.txt build target by ensuring
that pybuilddir.txt is always regenerated when configure is run and
that the newly built skeleton python does not inadvertently import
modules from previously installed instances. [1]
This changeset has been committed for release in 2.7.9, 3.4.2, and 3.5.0.
A HUGE thank you to Ned Deily from the Python Core Development Team
for helping to identify the underlying cause, produce a fix and
wonderfully document the explanation.
[1] http://bugs.python.org/issue21166
preparatory step to convert bsd.python.mk into a USES file.
- Remove the shared/static build separation, which is the source of many
problems and even more hacks. Instead build only the shared version, which
greatly simplifies the build.
- Use NLS_LIBS instead of NLS_LDFLAGS as done for lang/python27 (r357486)
- Remove the FPECTL option to align the build with the clean "template" from
lang/python34.
- Remove PORTDATA and EXAMPLES. Those will be made available via separate
ports.
- Add a new DEBUG option to enable debug builds as for lang/python34.
- Add a new TSC option for precise timestamp counter support as for
lang/python34.
- Reactivate curses/ncurses support.
- Use buildbottest in the regression-test: target.
Phabric: D410
Exp-run: 192242, 192244
Reviewed by: koobs, bapt
With hat: python@
Copy the second part of a change previously made to python27 [1], to
python31, python32 and python33.
This fixes staging and packaging of these ports by a non-root user by
running ranlib on the archive prior to it being installed read-only.
While I'm here:
- python27: Add breadcrumbs and references to the patch header
- python34: Update breadcrumbs and references to the patch header
[1] https://svnweb.freebsd.org/ports?view=revision&revision=350207
Submitted by: antoine
Reviewed by: kwm, sbz
Copy change committed to python27 [1] to python31, python32 and
python33 to fix builds of some extensions with Clang 3.4.
Also add breadcrumbs to the patch header in lang/python27 referencing
the upstream issue. [2]
The Python 3.4 port (lang/python34) already carries the patch.
[1] https://svnweb.freebsd.org/ports?view=revision&revision=346428
[2] http://bugs.python.org/issue20767
A vulnerability was reported [1] in Python's socket module, due to a
boundary error within the sock_recvfrom_into() function, which could be
exploited to cause a buffer overflow.
This could be used to crash a Python application that uses the
socket.recvfrom_info() function or, possibly, execute arbitrary code
with the permissions of the user running vulnerable Python code.
This vulnerable function, socket.recvfrom_into(), was introduced in
Python 2.5. Earlier versions are not affected by this flaw. This is
fixed in upstream branches for version 2.7, 3.1, 3.2 and 3.3.
[1] http://bugs.python.org/issue20246
MFH: 2014Q1
Security: 8e5e6d42-a0fa-11e3-b09a-080027f2d077
The current FreeBSD/ARM __clear_cache() implementation does nothing #if
__i386__ || __x86_64__ #else abort();
cognet@ advises this is an issue for anything !Apple that is using the
libcompiler_rt provided by Clang on ARM, and requires upstreaming.
This is the root cause of abort() on import for the ctypes module in
Python, as they bundle libffi. [1]
This change patches the bundled libffi library in all Python ports, even
though it is a NOOP for the ports that use devel/libffi. These ports,
currently python31, will get the fix via ports/184517
A huge shout out to cognet@ who helped diagnose the issue and created
the patch to address it. Thank you!
PR: ports/149167 [1]
PR: ports/184517
Submitted by: cognet [3]
Reviewed by: cognet, eadler, milki, ak
and lang/python2 and lang/python3. This change brings us closer to the goal
of making Python ports usable with different Python versions at the same
time.
- Add a new lang/python2 port to handle the symlinks for bin/python2,
bin/idle2, bin/pydoc2 and so on.
- Add a new lang/python3 port to handle the symlinks for bin/python3,
bin/idle3, bin/pydoc3 and so on.
- Bump the PORTREVISION on all lang/python* ports.
build problems with pmake, that's cannot be reproduced reliable. There is
also upstream bug-report [1]. The solution is suggested by Marcus von Appen.
PORTREVISION isn't bumped, because there is no need in gmake for people
that already built python33 successfully.
Reported by: many
Submitted by: mva (by mail)
Tested by: demon
. lang/python27: 2.7.3 -> 2.7.5
. lang/python32: 3.2.3 -> 3.2.4
. lang/python33: 3.3.0 -> 3.3.1
- update Mk/bsd.python.mk with new versions
- mark lang/python26 and lang/python31 as deprecated (set them to
upstream EoL dates)
- update docs (lang/python-doc-html)
- align databases/py-bsddb patch for python27 - most of it was applied
upstream. Raise BDB version to 4.3 atleast, according to
upstream requirements.
Many thanks to Martin (miwi) for his time on this update.
PR: 178506
Submitted by: rm (myself)
Exp-run by: portmgr (miwi)
- revert erroneous threads patch in lang/python26 and lang/python27,
that was added after ports/131080. It was rejected upstream, because it's
not actually a bug, but misuse.
Gabor Pali (pgj) in collaboration with Kubilay Kocak (koobs) did an
independent investigation regard the issue. See here for details:
http://lists.freebsd.org/pipermail/freebsd-python/2013-April/005376.html
PR: 153167
Submitted by: Duncan Findlay <duncan@duncf.ca>
Reported by: pgj/koobs (at python@ ML)
Exp-run by: portmgr (miwi)
for an exp-run of updated python versions.
- trim Makefile headers
- remove leading indefinite article from COMMENT
- use PYTHON shortcut in MASTER_SITES
- whitespace fixes
- remove checks for unsupported versions of FreeBSD
- use static value ``33'' instead of PYTHON_SUFFIX in lang/python33/pkg-plist,
because this value is not supposed to be changed across the branch and for
consistency with other python3 ports
- remove conflicts in lang/python-mode.el with not more existing python-2.4
