venvs
is a tool for configuring, in a single file, a set of
virtualenvs, which packages to install into each, and any binaries to
make globally available from within.
The usual:
$ pip install venvs
The best way to use venvs
is by creating a file named
~/.local/share/virtualenvs/virtualenvs.toml
. Here's an example of what goes
in it:
[virtualenv.development] install = [ "pudb", "twisted", ] link = ["trial"] [virtualenv.app] install = ["$DEVELOPMENT/myapp"]
After creating the above, running venvs converge
will create 2
virtualenvs, one called "development" with pudb
and twisted installed
into it and trial linked from within it onto your PATH
, and a second
called "app" installing the corresponding directory.
For a more intricate example, have a look at my own virtualenvs.toml.
That's about all you need to know. If you insist on reading further
though, venvs has an older, not-very-recommended mutable interface
which allows you to create virtualenvs in a central location without
tracking them in a config file (or converging them). For that, usage
is similar to mkvirtualenv
, although venvs
passes arguments
directly through to virtualenv
:
$ venvs create nameofvenv -- -p pypy
will create a virtual environment in an appropriate platform-specific
data directory, or in the directory specified by WORKON_HOME
for
compatibility with virtualenvwrapper
.
A common use case for virtualenvs is for single-purpose installations, e.g.:
"I want to install fabric and give it its own virtualenv so that its
dependencies can be independently upgraded, all while still being able to use
the fab
binary globally".
venvs
supports a --link
option for this use case:
$ venvs create -i fabric --link fab
will create a virtualenv for fabric (in the same normal location), but will
symlink the fab
binary from within the virtualenv into your
~/.local/bin
directory.
(You may have heard of pipsi which is a similar tool for this use case, but with less customization than I would have liked.)
I also find mktmpenv
useful for quick testing. To support its use case,
venvs
currently supports a different but similar style of temporary
virtualenv.
Invoking:
$ venv=$(venvs temporary)
in your shell will create (or re-create) a global temporary virtualenv,
and print its bin/
subdirectory (which in this case will be then
stored in the venv
variable). It can subsequently be used by, e.g.:
$ $venv/python
or:
$ $venv/pip ...
et cetera.
You may prefer using:
$ cd $(venvs temporary)
as your temporary venv workflow if you're into that sort of thing instead.
The global virtualenv is cleared each time you invoke venvs temporary
.
Unless you care, unlike virtualenvwrapper
's mktmpenv
, there's no
need to care about cleaning it up, whenever it matters for the next
time, it will be cleared and overwritten.
venvs
may support the more similar "traditional" one-use virtualenv in the
future, but given that it does not activate virtualenvs by default (see below),
the current recommendation for this use case would be to simply use the
virtualenv
binary directly.
Besides the venvs
for named-virtualenv creation and venvs
temporary
for temporary-virtualenv creation described above:
$ venvs find name foo
will output (to standard output) the path to a virtualenv with the given name
(see also --existing-only
), and:
$ venvs remove foo
will remove it.
There are a number of other slight variants, see the --help
information for
each of the three binaries.
Real documentation to come (I hope)
virtualenvwrapper
is great! I've used it for a few years. But I've
slowly settled on a much smaller subset of its functionality that I like
to use. Specifically:
I don't like activating virtualenvs.
virtualenvs are magical and hacky enough on their own, and piling activation on top just makes things even more messy for me, especially when moving around between different projects in a shell. Some people use
cd
tricks to solve this, but I just want simplicity.I don't need project support.
I've never attached a project to a virtualenv. I just use a naming convention, naming the virtualenv with the name of the repo (with simple coercion), and then using dynamic directory expansion in my shell to handle association.
Basically, I just want a thing that is managing a central repository of
virtualenvs for me. So that's what venvs
does.