* NLTK support: Update test to use multiple corpora
So that the incorrect handling of multiple IDs seen in #444 would
have been caught.
Also switches to some of the smaller corpora, to reduce time spent
downloading during tests (see sizes on http://www.nltk.org/nltk_data/).
* NLTK support: Fix passing of multiple corpora identifiers
As part of fixing the shellcheck warnigns in #438, double quotes had
been placed around `$nltk_packages` passed to the `nltk.downloader`,
which causes multiple identifiers to be treated as though it were just
one identifier that contains spaces.
The docs for the shellcheck warning in question recommend using arrays
if the intended behaviour really is to split on spaces:
https://github.com/koalaman/shellcheck/wiki/SC2086#exceptions
As such, `readarray` has been used, which is present in bash >=4.
The `[*]` array form is used in the log message, to prevent shellcheck
warning SC2145, whereas `[@]` is used when passed to `nltk.downloader`
to ensure the array elements are unpacked as required.
Note: Both before and after this fix, using anything but unix line
endings in `nltk.txt` will also cause breakage.
* Test that the cache is invalidated when the stack changes
* Improve UX when clearing cache due to the stack changing
Now outputs a message informing that the cache was cleared, and
clears the cache first to avoid a redundant message about removing
an old Python version.
The pip-uninstall step stopped working when it was moved to after
the pip-install step in f27a84e.
This regression was temporarily fixed by part of #397, however that
PR was reverted in #404.
Adds a test to hopefully catch any future regressions :-)
Fixes#393.
Ed Morley
kennethreitz