The 0.98.5 windows installer is still overwriting existing installations of configobj. This package is not needed by matplotlib and should not be installed on windows, nor should traits.
I have removed configobj and enthought.traits from the trunk, but it is still present on the maintenance branches. On build systems, setup.cfg for the maintenance branches should be configured to not install configobj or traits under any circumstances. Once it has been so configured, the build directory needs to be deleted, because it probably contains a cached copy of configobj that will get incorporated into the installer the next time it is built, regardless of the settings in setup.cfg.
This bug has been assigned to me, but it is out of my hands at this point.
Darren
Since config obj is removed, it is mostly moot going forward, but it
appears that something still went wrong with the win32 build process.
Charlie, are you maintaining a setup.cfg file you use for the win32
builds that has configobj and traits off? You also need to make sure
that you delete the build dir after any changes to setup.cfg before
rebuilding since cruft in the builddir can get carried along. I will
add this to the devel release notes,
JDH
···
On Sat, Dec 13, 2008 at 4:52 AM, Darren Dale <dsdale24@...149...> wrote:
The 0.98.5 windows installer is still overwriting existing installations of
configobj. This package is not needed by matplotlib and should not be
installed on windows, nor should traits.
I have removed configobj and enthought.traits from the trunk, but it is
still present on the maintenance branches. On build systems, setup.cfg for
the maintenance branches should be configured to not install configobj or
traits under any circumstances. Once it has been so configured, the build
directory needs to be deleted, because it probably contains a cached copy of
configobj that will get incorporated into the installer the next time it is
built, regardless of the settings in setup.cfg.
This bug has been assigned to me, but it is out of my hands at this point.
I had all the build opts explicitly set, so I am not quite sure what
happened. I'm a little more worried about the error two people have
reported involving matplotlib.conf.template. This file is present in
the source.
- Charlie
···
On Sat, Dec 13, 2008 at 10:53 AM, John Hunter <jdh2358@...149...> wrote:
On Sat, Dec 13, 2008 at 4:52 AM, Darren Dale <dsdale24@...149...> wrote:
The 0.98.5 windows installer is still overwriting existing installations of
configobj. This package is not needed by matplotlib and should not be
installed on windows, nor should traits.
I have removed configobj and enthought.traits from the trunk, but it is
still present on the maintenance branches. On build systems, setup.cfg for
the maintenance branches should be configured to not install configobj or
traits under any circumstances. Once it has been so configured, the build
directory needs to be deleted, because it probably contains a cached copy of
configobj that will get incorporated into the installer the next time it is
built, regardless of the settings in setup.cfg.
This bug has been assigned to me, but it is out of my hands at this point.
Since config obj is removed, it is mostly moot going forward, but it
appears that something still went wrong with the win32 build process.
Charlie, are you maintaining a setup.cfg file you use for the win32
builds that has configobj and traits off? You also need to make sure
that you delete the build dir after any changes to setup.cfg before
rebuilding since cruft in the builddir can get carried along. I will
add this to the devel release notes,
JDH
I think it should not be an issue for new releases from the trunk, as long as the build directory has been deleted so there is no chance of a cached configobj finding its way into the installer.
···
On Sat, Dec 13, 2008 at 10:53 AM, John Hunter <jdh2358@…149…> wrote:
On Sat, Dec 13, 2008 at 4:52 AM, Darren Dale <dsdale24@…149…> wrote:
The 0.98.5 windows installer is still overwriting existing installations of
configobj. This package is not needed by matplotlib and should not be
installed on windows, nor should traits.
I have removed configobj and enthought.traits from the trunk, but it is
still present on the maintenance branches. On build systems, setup.cfg for
the maintenance branches should be configured to not install configobj or
traits under any circumstances. Once it has been so configured, the build
directory needs to be deleted, because it probably contains a cached copy of
configobj that will get incorporated into the installer the next time it is
built, regardless of the settings in setup.cfg.
This bug has been assigned to me, but it is out of my hands at this point.
Since config obj is removed, it is mostly moot going forward, but it
appears that something still went wrong with the win32 build process.
Charlie, are you maintaining a setup.cfg file you use for the win32
builds that has configobj and traits off? You also need to make sure
that you delete the build dir after any changes to setup.cfg before
rebuilding since cruft in the builddir can get carried along. I will
add this to the devel release notes,
Did you delete the build directory before you rebuilt for the installer?
···
On Sat, Dec 13, 2008 at 12:01 PM, Charlie Moad <cwmoad@…149…> wrote:
I had all the build opts explicitly set, so I am not quite sure what
happened. I’m a little more worried about the error two people have
reported involving matplotlib.conf.template. This file is present in
the source.