Allow runpy to consider non-standard Hy source extensions

[brandonwillard]

Adds Hy parsing of files with unrecognized source extensions to cmdline Hy.

The fallback is tried only after first attempting standard Python compilation. If the attempt raises a SyntaxError (and the file is not clearly identified as Python source via its extension), then it tries Hy.
The reason for choosing the order Python-then-Hy is somewhat arbitrary, except for the impression that Python's parsing might be more efficient than Hy's and, thus, better as a first attempt.

[brandonwillard]

@Kodiologist, the one failure on Travis — for pypy3 — is really weird; do you mind retrying that check manually?

I've seen spurious errors for other projects with pypy3 on Travis recently, and I can never reproduce them locally. Perhaps it's because I have pypy at Python 3.5.3[pypy-6.0.0-final] on my machine, but, still, the errors make no sense whatsoever (relative to the changes, or otherwise).

[Kodiologist]

I don't think this is the right way to address #1677. For example, if there's a syntax error for both Python and Hy interpretations of the file, which error do you show?

@Kodiologist, the one failure on Travis — for pypy3 — is really weird; do you mind retrying that check manually?

I don't think that's an option on Travis, but you can trivially change your top commit (e.g. git commit --amend --reset-author to change the timestamp) and then force-push.

[brandonwillard]

In this case, the Hy error, just as if we only assumed the file was Hy. Otherwise, do you think Python should never even be tried for unknown filetypes?

For Travis, I've been able to retry with this before.

[Kodiologist]

For Travis, I've been able to retry with this before.

Oh, I didn't notice that button. I've pushed it.

Otherwise, do you think Python should never even be tried for unknown filetypes?

In the case of invoking hy FILENAME, indeed it shouldn't be tried. runpy.run_path, on the other hand, should probably assume Python. Otherwise you'll see Hy syntax errors when you try to run a Python script that you made a typo in, right?

[brandonwillard]

@Kodiologist, this new approach should do just that. I chose to copy the runpy module to produce a runhy that works exactly like runpy except that it prefers Hy for unknown file types. It uses the same patching as before, but with a different, more inclusive condition for Hy files.

With this approach, we preserve the functionality of runpy for the Hy-centric cmdline case and the standard runpy (still patched for limited Hy parsing) without wastefully reproducing its logic.

[Kodiologist]

Neat! Try adding a test where the shebang line is used by running the file as an executable, too. Assuming it's possible to arrange for that in pytest.

[brandonwillard]

It is possible, but wouldn't that only be testing the shell's ability to correctly run a script?

[Kodiologist]

Not quite, since this failed even though my shell didn't have a relevant bug.

[Kodiologist]

Nah, forget it, it's probably overkill. I'll review this soon.

[Kodiologist]

Excellent work. Thanks so much. No NEWS update is needed since the bug being fixed is a regression since the previous release.

[Kodiologist]

Sorry I forgot about Hy for a few days there. I'll merge this now.