Check that git, bazaar, and mercurial are installed before running tests that require them

[spenczar]

This is just a quality-of-life thing as a contributor.

Fixes #184.

[sdboyer]

Perfect :)

[codecov[bot]]

Codecov Report

Merging #186 into master will decrease coverage by 0.21%. The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #186      +/-   ##
==========================================
- Coverage   79.49%   79.27%   -0.22%     
==========================================
  Files          24       24              
  Lines        3804     3798       -6     
==========================================
- Hits         3024     3011      -13     
- Misses        586      589       +3     
- Partials      194      198       +4

Impacted Files	Coverage Δ
cmd.go	73.68% <0%> (-3.59%)	:x:
deduce.go	78.09% <0%> (-1.67%)	:x:
source_manager.go	91.38% <0%> (-0.75%)	:x:
solver.go	83.6% <0%> (+0.41%)	:white_check_mark:

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 9a583ef...249129b. Read the comment docs.

[sdboyer]

I'm kinda surprised and concerned that this would've resulted in a coverage decrease - that would seem to suggest more effects than this should really have. And this is based on the merge result of #185, which is more reasonable to have had those effects...hmm.

Well, not the end of the world.

One nit, though, before I merge - please limit the first line of git commit messages to 50 chars. I'm not as exacting as Tim Pope about his rules, but I do prefer following them.

[spenczar]

Yeah, the coverage decrease seems pretty weird. I've never used codecov, but I'm trying to figure out where the coverage decrease could have come from.

If you like, I'd be happy to amend the commit message and make a new PR to keep the history clean, I'd prefer to be a good citizen 😄

[sdboyer]

I appreciate it 😄 - just amending the commit and force-pushing to this PR would be fine. No need for a new one.

[spenczar]

This looks it could be a flaky test failure, maybe?

--- FAIL: TestSignalHandling (0.01s)
    manager_test.go:809: time to send signal: 13.449µs
    manager_test.go:812: time to return from Release(): 96.1µs
    manager_test.go:815: finished too fast (96.1µs); the necessary network request could not have completed yet
    manager_test.go:850: Expected error on statting what should be an absent lock file```

[sdboyer]

It is - it's a really hacky way of trying to approximate whether or not the signal handling is occurring correctly. Fails maybe 1/50 times. (I had to move on to other things, so went with the crappy heuristic)

Kicking it, should work on the next run...

[sdboyer]

OK, we're good.