Remove `input` and `output` arguments

[jyn514]

This is a mess of a merge, sorry about that.

Highlights:

• --input and --output have been removed completely as arguments. This simplifies logic considerably. Shell redirection (< > |) can be used as an alternative
• when stdin is not a terminal and run without arguments, default to encoding. Solves https://github.com/charlesdaniels/bitshuffle/issues/55
• main is no longer a function; since __name__ will always be __main__, it would have always run
• many tests have been modified for compatibility. For the most part, replacing input/output with </> was sufficient. The exceptions:
  • "When run with no args, prints help... " can no longer be tested (since this should only occur when interactive)
  • "Test infer decode from --output and non-tty stdin... " no longer makes sense to test

Note also that Powershell tests are very broken and I don't know how to fix them. From the errors, it should be fairly simple, I just don't know powershell. This request should not be merged until the tests have been fixed.

Minor changes:

• --decode and --encode are now mutually exclusive. Previously, if both were specified the message would be encoded
• --version now uses argparse's built-in version action
• wrapper.py no longer uses execvp; it was incompatible with shell redirection. This is not due to the change, it was just made very evident
• fixed a couple typos
• some tests have been styled a little to make them easier to read

[charlesdaniels]

I will attempt to fix the PowerShell issues myself, since I have some experience with PS.

This looks good, but I would like justification for why we got rid of main(). I also have some reservations about the implementation of the wrapper, but I don't really have a better idea yet.

[charlesdaniels]

Hmm, it looks like even with a4b3ae2, the wrapper methodology still does not work on Windows, unless I am reading the AppVeyor logs incorrectly.

[jyn514]

Interestingly, tests besides --version succeed. No idea why --version would give an error.

[jyn514]

Logic behind removing main - this requires a bit of background, apologies if you know most of this.

In Python, it's customary to put 'main' routines in their own block, called at the end by if __name__ == "__main__". This allows imports to access the script programmatically, without having to worry about arguments; i.e. something like from bitshuffle import decode would work fine.

Since we've already separated functions that can be imported into their own files (library and __init__), there's no reason to keep a guard for __name__. Even disregarding style, __name__ is set to the name of the file if not run directly; since the file it's in is named __main__.py, if __name__ == "__main__" will always be true.

[jyn514]

I agree this was probably not the place to make such a change, I'll revert it and propose it later.

[jyn514]

@charlesdaniels may I ask what you dislike about the wrapper? It's important it be in python so that it keeps Windows compatibility; if you have a suggestion it would be welcome. Otherwise I thought I did it fairly reasonably.

[charlesdaniels]

The logic behind removing main() is sound and I agree.

It would be better if the wrapper could run without aubprocess it exec, by importing bishuffle and calling main. This could be accomplished outside of a pip using sys.path.append.

I will investigate the PS issue further later today when I’m not on mobile.

[jyn514]

bitshuffle is dead, long live bitshuffle 🐍