No, this is just bad. The whole paradigm of having build setup depend on local variables in an interactive shell (which is pervasive, virtualenv is hardly the only bad actor here) is just broken inherently. It's lazy programming. It creates all sorts of failure modes -- builds can magically stop working when people update their .bashrc files (e.g. set PATH explicitly, but OOPS the build system expects to source a script and then spawn you an interactive shell), etc...
Implicit configuration is bad, OK? Just don't do it.
Edit: in fact something like /usr/bin/virt-python my-env is much nicer than /var/lib/my-app/env/bin/python. Basically like virtualenvwrapper but for Python scripts that are not run interactively.
I regularly write scripts that involve more than one virtualenv. At the moment this is trivial by just activating and deactivating envs from one script. Starting subshells is complicated because communication into the shell barely exists. Worse than that is that the shebang only supports two arguments and is heavily length limited.
Unless you are rapidly switching envs, you could launch the parts of your scripts that operate on different envs as subscripts. Then you can pipe data in and out. Can you provide an example where this does not work?
