Fix another unbound variable error

[ackalker]

This fixes the case of bash: PS1: unbound variable when PYENV_VIRTUALENV_DISABLE_PROMPT isn't set.
This should have been part of #423, sorry for that. I also fixed the tests related to virtualenv activation/deactivation which were failing due to my changes.

[native-api]

IIRC Bats runs its own instance of Bash so your main shell's settings shouldn't affect it.

So I doubt that this part and similar ones below and in other files are necessary.

[ackalker]

I'm sorry for the misunderstanding. I discovered the problem by running my own tests, outside of Bats. You can reproduce the problem easily by running this test script on a tree without my latest commits applied, i.e. current master:

test-pyenv-set-u.bash

#!/usr/bin/bash

set -u

eval "$(pyenv init -)"

# Uncomment either of these blocks for tracing

# export PYENV_DEBUG=1

# -or-

export PS4='+(${BASH_SOURCE:-<none>}:${LINENO}): ${FUNCNAME[0]:+${FUNCNAME[0]}(): }'
set -x

pyenv virtualenv system testenv
echo created
pyenv activate testenv
echo activated
pyenv deactivate
echo deactivated
pyenv virtualenv-delete testenv
echo deleted

set +x

Output of running this on my system, for both methods of tracing is in this gist.
Note the problem is in these two lines: https://gist.github.com/ackalker/4de61234e8579fe96cd3a4c7da90da3b#file-test-pyenv-set-u-ps4-log-L46-L47 , which are in a snippet which is evaled, which doesn't show up in the trace with PYENV_DEBUG=1. Note that PS1 is unbound while running scripts,

[native-api]

I'm not putting the validity of the problem into question. I'm only questioning if this particular part of the changes is really necessary to fix it.

The other changes change the output of sh-activate and tests that check for that output. But these changes to the tests, I do not see why they are necessary.

[ackalker]

I think that I understand the confusion now. The changes to sh-deactivate and sh-activate should have been in one commit, the changes to the tests address changes to both files.

[ackalker]

Gist of running bats test, failing (tests-failing.log) at commit 8f721ac , and passing (tests-passing.log) after applying bc4c73a .
The remaining failing tests are unrelated to my fix.

[native-api]

Thanks for clarifying!