bundle_context: project_name alphanum-ish only; output_dir uses project_name instead of bundle_dir

[ndushay]

Per instructions from Ben on Slack, captured in #78 comments, use the project name rather than the bundle dir in the output directory for a bundle context.

Also, per separate slack conversation, also captured in #78 comments, restrict the characters allowed for the project name.

Lastly, fixes broken travis build.

Connects to #78

[coveralls]

Pull Request Test Coverage Report for Build 988

• 2 of 2 (100.0%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.3%) to 59.812%

---

Totals
Change from base Build 969:	0.3%
Covered Lines:	573
Relevant Lines:	958

---

💛 - Coveralls

[atz]

I don't think this interaction is doing what you expect. attr_hash and bc are decoupled after the subject is established. It would read more cleanly (and work) if you just do assignment on the object like:

bc.project_name = 'foo'

Ideally, start from a valid object and

expect { bc.project_name = 'bad value' }.to change(bc, :valid?).to(false)

[jmartin-sul]

makes sense, i think...

for my own clarification, does that mean that the first time attr_hash is edited, it does affect bc (i.e. for the expectation this comment is on), and that for subsequent expectations, the changes to attr_hash don't affect bc, because bc got lazily loaded upon the first reference to it? or is bc already set up before this entire test is even entered, because it's declared using subject? or something else?

anyway, i'd agree the suggested usage is more readable.

[ndushay]

thanks, @atz. I have added a commit to address my test giving a false positive.