On 01/07/2017 08:42 PM, Cameron Simpson wrote:
On 07Jan2017 19:45, jim <jf_byr...@comcast.net> wrote:
On 01/07/2017 05:58 PM, Clint Moyer wrote:
So to minimize your issues with installing Python packages, take the
path of least resistance and install through your system repo. And use
Python2 or Python3 explicitly to avoid conflicts.
As I mentioned in another post, most of the more popular modules I had
installed on my old system using pip are available in the repository
and I will use the repository to install them on the new system. I now
understand that using sudo is a bad idea.
Cool.
One question from the earlier post that did not get answered concerned
upgrading a repository installed module with pip.
I would recommend not. As soon as you get there:
- if the vendor updates it, use apt-get (I know this doesn't fit your
situation with Ubuntu 16.04 LTS)
- if you want a more modern version, now is the time to use virtualenv
The thing about LTS is that the vendor guarentees its stability. If you
upgrade the vendor installed package using /usr/bin/pip (or pip3) as
root, you're replacing the vendor supplied module with a newer one,
which may break vendor supplied packages which use that module expected
the _old_ behaviour.
So endeavour to leave the vendor suppplied stuff entirely alone except
for apt-get style updates.
Instead, make a virtualenv an upgrade it with a newer package.
To get started on the new system I installed pip3 from the repository.
The first time I used it to install a module it said a newer version
was available and gave the command to update it. What are the
consequences of using pip to upgrade repository installed modules?
In theory, if the newer version of the module installs a breaking change
it can break things in the system which might use that module and expect
its old behaviour. Also, _sometimes_, vendors supply patches versions of
packages, possibly including python modules. If they're
modified/patched, there is probably a reason. You'd be undoing that patch.
I ask because 16.04 is LTS and won't be getting version upgrades
unless they are security related. Also pandas is in the repositories
but the module pandas-datareader, which I may need to use, is not.
Again: make a virtualenv as yourself.
Its pip can be run as yourself, therefore cannot accidentally modify
system module. It's pip can freely upgrade a older module. If you
install a module _not_ supplied by the vendor, the virtualenv pip can
freely do that, and _also_ upgrade a required module should that be
required (PyPI package definitions let a module specify required
revisions of other modules it may use).
Cheers,
Cameron Simpson <c...@zip.com.au>
You've convinced me. I will have to buckle down and figure out to use
virtualenv/venv. I see Chris recommends venv. I read through PEP405 and
this link https://docs.python.org/3.5/library/venv.html. I don't pretend
to fully understand it all but it does seem to make a good case for
using venv over virtualenv.
thanks, Jim
--
https://mail.python.org/mailman/listinfo/python-list
