Fix ExtrasError.

[lambdafu]

I noticed that ExtrasError wasn't showing the extra argument in a subcommand. So I dug deeper, and I could identify a bug in remaining(true), which was overwriting the miss_list instead of extending it. So, I added some tests and fixed it.
But then I noticed some bargled logic in triggering ExtrasError: Firstly, there is a mismatch between remaining(true) and remaining_size() (without true). Secondly, it didn't handle allow_extras_ and prefix_command_ on a per-subcommand basis. I tried a couple of things, but the easiest fix was to just let each subcommand fail on its own, by not using recurse=true at all.
As recurse=true is not used anymore (and because it does not handle allow_extras_ and prefix_command_), the recurse=true case could be removed from the code base. My patch doesn't do that, though, to keep the change minimal, in case there are some other plans or cases I missed. My recommendation would be to remove it if it doesn't serve any other purpose. I am not firm on the logic behind the fallthrough_ cases, so it may be that I am missing something.

[henryiii]

I'll look into this in more detail when I get a chance in the next day or two, but not surprised - that code was added very recently. I really need a series of more complex subcommsnd tests that mix features...

[codecov]

Codecov Report

Merging #49 into master will decrease coverage by 0.32%.
The diff coverage is 78.94%.

@@            Coverage Diff             @@
##           master      #49      +/-   ##
==========================================
- Coverage     100%   99.67%   -0.33%     
==========================================
  Files           9        9              
  Lines        1230     1235       +5     
  Branches      248      248              
==========================================
+ Hits         1230     1231       +1     
- Misses          0        4       +4

Impacted Files	Coverage Δ
include/CLI/App.hpp	99.41% <78.94%> (-0.59%)	⬇️

---

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 675e753...bcb6528. Read the comment docs.

[henryiii]

I've added a fix for the subcommand ordering, as well.

[henryiii]

I think recurse=true is still useful if someone just wants to collect all extras, possibly for debugging. I've added a test that makes sure it can rebuild the command line in order - which also fixes the callback order. I'm not sure if it has a practical use other than for diagnostics, since I would think users would always want to have the subcommands, as well. I agree with you that it should not be used if allow_extras is false - that should fail. Let me know if you have any thoughts!

By the way, the main purpose for fallthrough is to allow something like this:

./run_model model1 --common-opt --model1-opt

to work, where common options can be set on run_model but be given to subcommands. I'm not sure if the interaction with other models like prefix commands is perfect yet.

[henryiii]

Thanks!

[henryiii]

FYI, Windows requires #include <iterator> to use back_inserter. Fixed in master. :)