Why is "pip install -U --no-deps PyBERT" failing, in the setup.py file of one of PyBERT's dependencies?

Question

One of my users is reporting the following failure, when attempting to update the PyBERT package, using:

pip install -U --no-deps PyBERT

> Requirement already up-to-date: pybert in
> c:\anaconda2\lib\site-packages Collecting traitsui (from pybert)  
> Using cached traitsui-5.1.0.zip Collecting chaco (from pybert)   Using
> cached chaco-4.6.0.tar.gz
>     Complete output from command python setup.py egg_info:
>     Traceback (most recent call last):
>       File "<string>", line 1, in <module>
>       File "c:\users\twesterh\appdata\local\temp\pip-build-yxql2k\chaco\setup.py",
> line 131, in <module>
>         write_version_py()
>       File "c:\users\twesterh\appdata\local\temp\pip-build-yxql2k\chaco\setup.py",
> line 111, in write_version_py
>         match = re.match(r'.*?\.dev(?P<dev_num>\d+)', full_v)
>     NameError: global name 'full_v' is not defined

Does anyone know:

1. Why is the "pip install..." process calling the chaco setup.py file, when pip has already decided to use the installed/cached version of chaco?
2. Where the global name, full_v, is normally defined?

My user reports that, if he uninstalls/reinstalls PyBERT, then everything works fine. Can anyone explain what subtle differences occur, for "pip install -U {package}", as compared to "pip uninstall {package}" / "pip install {package}"?

Please, note that chaco was installed with "conda install chaco", as opposed to "pip install chaco". I'm using mixed installation commands, because:

1. I find that I get better results, in general, when I give conda the first crack at new package installation.
2. PyBERT has not, yet, been registered in conda-forge.

Thanks!
-db