MAINT: Do not install matplotlib backend without DISPLAY

[teddyrendahl]

Description

Check that we have a DISPLAY environment variable before configuring the matplotlib backend for IPython. This destroys the entire session which is strange. We should allow users without displays to access non-graphics functions. A warning is printed to the terminal that alerts the user they don't have a configured DISPLAY so they will be aware something is strange.

Just for fun, I tried seeing what would happen if I ran a bluesky scan with BestEffortCallback to see if it exploded without matplotlib. This seems to error nicely and print an error that is consistent with our warning.

RuntimeError: Invalid DISPLAY variable

We could be extra nice and not configure the BEC for non display users but this seems extreme to me.

Motivation and Context

As a fix for pcdshub/Bug-Reports-and-Requests#6

How Has This Been Tested?

Tested locally without X-Forwarding. Still confused how this function works in Travis

Where Has This Been Documented?

[codecov-io]

Codecov Report

Merging #132 into master will decrease coverage by 0.24%.
The diff coverage is 50%.

@@            Coverage Diff             @@
##           master     #132      +/-   ##
==========================================
- Coverage   97.18%   96.93%   -0.25%     
==========================================
  Files          15       15              
  Lines         781      783       +2     
==========================================
  Hits          759      759              
- Misses         22       24       +2

Impacted Files	Coverage Δ
hutch_python/cli.py	97.43% <50%> (-2.57%)	⬇️

---

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 0c0c897...0a0d3a7. Read the comment docs.

[ZLLentz]

I'm still confused over this bug existing. I'm happy with this solution. I don't think we should go through and make sure all the graphics options are disabled, but we may consider going back to revising the default RE callbacks if people are ssh-ing around without x-forwarding enabled and want to collect data... Seems unlikely.

[teddyrendahl]

Looks like test just calls this function and checks for OSError. Where does that get raised? Presumably before we actually go to embed the shell / it affects the non-existent display

[ZLLentz]

I would have expected travis to fail on

hutch-python/hutch_python/tests/test_tstpython.py

Lines 28 to 29 in 0c0c897

	ipy_text = check_output([tstpython], universal_newlines=True,
	input='unique_device\n')

with a CalledProcessError

[ZLLentz]

There are two bizarre things with the test that didn't fail:

• subprocess.Popen and affiliates behave subtly different than running scripts in a shell or with os.system
• ipython shells that get stdin passed into them before entering the shell never reach the xqcb error and exit early

I looked into adding a test that fails before this PR and succeeds after, but the test is awkward and seems to subtly ruin the session after running. Hopefully we won't hit similar issues...