Add test script to compare rg --help output to zsh completion function

[okdana]

Per our discussion in #536: Here's something to hopefully prevent the zsh completion function from getting out-of-synch with the actual app.

I used a slightly different strategy from the one i mentioned before, but it's the same basic idea — pull all of the options out of the --help output, pull all of the options out of the completion function, compare the two. I tried to address your concerns about false positives/negatives as best i could; i've documented the assumptions i make in the script, and i think it should be OK as long as nobody does anything bizarre.

I've tested this in dash, bash, and zsh. It also works in ksh93 if you make local an alias for typeset... but you use local in your other scripts so i assume ksh compatibility isn't a big concern.

Anyway, it turns out actually that the -f option is missing from the completion function, which is fortuitous i guess because it seems to demonstrate that the script behaves as expected! Here's the output:

heartswap:ripgrep % ci/test_complete.sh
zsh completion options differ from `--help` options:
--- `rg --help`
+++ complete/_rg
@@ -77,5 +77,4 @@
 -c
 -e
--f
 -g
 -h

Unless you have concerns with it i think i'm happy with the script itself. I'm not sure how to go about integrating it with your other tests though. Should i just add it to appveyor.yml or does it belong somewhere else?

[BurntSushi]

This looks pretty good! And yeah, I don't think I care too much about korn shell compatibility. :-) (Note that I am not the originator of these shell scripts.) If you wanted to make this #!/bin/bash to be more precise, then that's fine!

To add this to CI, I think it needs to go in ci/script.sh, which is run by Travis. Once you do that, it should be testable by pushing to this PR.

[BurntSushi]

We probably can't do this exact thing. This would at, the very least, depend on ripgrep being built before these tests are run. I think we could probably do it by using a debug build, but right now, a release build is not built in CI. It might be easier to just use grep. :P (I think you could set rg=grep and it will just work.)

[BurntSushi]

Oh hmm I see you're using the --replace flag. If you just change rg to target/${TARGET}/debug/rg, then I think that should be OK.

[okdana]

Yeah, i could easily have replicated most of the find/replace stuff with grep and/or sed, but i figured that since i needed rg for the --help output anyway i might as well just use it for everything

[okdana]

Updated the rg path and added the script to script.sh — seems to have worked!

Would you like me to add completion for -f as part of this PR, or should i make another one?

[BurntSushi]

Would you like me to add completion for -f as part of this PR, or should i make another one?

Whichever is most convenient for you! I'm find smushing it into this one via a separate commit.

[okdana]

Done, seems like everything's passing now!

[BurntSushi]

@okdana You're awesome. :-) Thanks!