Don't include default pattern when user specifies their own in Rake task.

[JonRowe]

When configuring rspec via Rake task and a developer specifies a pattern via rspec_opts, don't include the default pattern. Should fix #2303.

[myronmarston]

s/wont/does not/

(At least, we tend to use present-tense "does not" everywhere else in doc strings like these instead of future-tense "will not").

[myronmarston]

Also, this doc string is confusing. It looks like it does show that it adds --pattern if rspec_opts includes --pattern. How about:

it 'does not add the default --pattern if `rspec_opts` includes --pattern'

[myronmarston]

Weird, this PR didn't build on travis. Not sure why :(.

Anyhow, I definitely think that the change here is appropriate (if the user has configured --pattern in rspec_opts there's no reason to include the default pattern, too) but I think there's still a bug in the logic in #2002 that this does not address. My understanding of #2002 is that it ORs together multiple patterns -- meaning that any spec files matching any of the patterns should get included in the loaded spec files. But we are seeing in the case reported in #2303 that it does not work properly. It seems like if someone runs rspec --pattern pattern1 --pattern default_patern (where pattern1 is the pattern from example repo #2303) nothing would run and it would be a bug.

Do you have any plans to look more into that, @JonRowe?

[JonRowe]

My understanding of #2002 is that it ORs together multiple patterns -- meaning that any spec files matching any of the patterns should get included in the loaded spec files. But we are seeing in the case reported in #2303 that it does not work properly. It seems like if someone runs rspec --pattern pattern1 --pattern default_pattern (where pattern1 is the pattern from example repo #2303) nothing would run and it would be a bug.

It actually did work for me on 3.5.0, and is working when pattern is used, because it's escaped, it's the fact the pattern is unescaped causes the misbehaviour, I'm not sure what to do about that...

[myronmarston]

actually did work for me on 3.5.0, and is working when pattern is used, because it's escaped, it's the fact the pattern is unescaped causes the misbehaviour, I'm not sure what to do about that...

Strange, when I played around with the repo from #2303 I shellescaped the pattern (since I was getting a different error w/o it), and saw the exact issue reported by @xmik in #2303. Did you try applying the patch I posted in that issue that allowed me to repro the issue?

[myronmarston]

Changes here LGTM, merge when green.

[JonRowe]

It's the one in rspec_opts that needs shell escaping, I was getting the reported behaviour identically on 3.3.x and 3.4.x whilst using it in my own project was working on 3.5.x

[myronmarston]

It's the one in rspec_opts that needs shell escaping, I was getting the reported behaviour identically on 3.3.x and 3.4.x whilst using it in my own project was working on 3.5.x

I know, my patch escaped the alternate pattern everywhere it appeared, including in rspec_opts. Did you try applying my patch?