Allow underscored identifiers if within accented quotes

[Vindaar]

This PR slightly modifies the grammar for identifier names by allowing arbitrary underscores in identifiers iff they are written in accent quotes `, e.g.

let `__x__` = 5

is now allowed. While this example surely doesn't justify such a change to the spec, the ones shown below might.
If there's any chance this PR is considered to be merged, I'll happily modify the spec and provide a few test cases. To be fair I'm not sure if this change might have some unforeseen consequences. All tests seem to pass on my machine at least.

Identifier construction

Raised by @rayman22201 here https://gitter.im/nim-lang/Nim?at=5b85e53f58a3797aa3ce3c9d, identifier construction using accented quotes in templates allows for native snake_case:

template genProc(name: untyped): untyped =
  proc `get_ name`[T](x: T): T =
    discard

genProc(number)

Sure, underscores are ignored so in practice it's a non issue.

Python interoperability

If Nim wants to attract Python users with really good interoperability, this would be great to have. Python's "dunder" methods use double underscore identifiers, e.g. __dict__.
Discussed here yglukhov/nimpy#43 using nimpy calls to dunder methods need to be worked around using callMethod explicitly currently.
With this PR we would be able to just do the following:

import nimpy

let np = pyImport("numpy")
echo np.`__dict__`

cc: @erhlee-bird, @alehander42, @yglukhov

[Araq]

echo np.`__dict__` is ugly and I don't want more ugliness just because "Python interop" (which will always be subpar anyway, Python is too dynamic). -1 from me.

[alehander92]

I like it,
I guess "" is not ugly, because Nim uses it a lot, so __` is ugly: well that's very subjective. also if the user doesn't have that, he'll just use an uglier alternative, e.g. field("name")

[Araq]

It's objectively ugly because underscores are word separators and sep-sep-stuff-youre-actuallyinterestedin-sep-sep makes no sense at the first glance and and on the second glance you can guess "oh, ok, it means it's 'special' in some way".

[alehander92]

But my point is api-s with underscores exist and if those underscores would end up in the nim code anyway but wrapped in some kind of call

…

On Wed, 19 Sep 2018, 13:44 Andreas Rumpf, ***@***.***> wrote: It's objectively ugly because underscores are word separators and sep-sep-stuff-youre-actuallyinterestedin-sep-sep makes no sense at the first glance and and on the second glance you can guess "oh, ok, it means it's 'special' in some way". — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#9014 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAys-658dQjyoc8NoRttjCaCmWxEtHisks5uciABgaJpZM4WvxvW> .

[Vindaar]

echo np.`__dict__` is ugly

I totally agree, I don't like it either. But as @alehander42 already said, these things are out there. I can't change e.g. Python. Personally I'd only use it for such use cases. And given that one needs to use backticks, I suppose it's quite unlikely that anyone would willingly use such an API for Nim itself.

"Python interop" (which will always be subpar anyway, Python is too dynamic)

May very well be true, but some simple features helping in that regard don't hurt imo (although changing Nim's spec just for that is maybe too much).

But either way, if this is not merged, I won't be too sad either. :)

[yglukhov]

I was also surprised that such identifiers do not work in backticks, I thought backticks were specifically designed to support all sorts of ugly identifiers.

[dom96]

+1 from me.

Python interop is a good enough reason to have this.

[dom96]

Why is this an int and not a bool?

[Vindaar]

Uhm, because when I hacked it together originally a few weeks ago, I wanted to toggle it with xor to avoid
L.inAccent = if L.inAccent: false else: true.
I agree that it should be a bool though. Is there a builtin to toggle a bool or is this the most succinct?
edit: Oh, stupid me. I can do the same thing with a bool, haha.

[Stromberg90]

You could write L.inAccent = not L.inAccent

[Vindaar]

Fixed.

[Vindaar]

@Stromberg90 A brain is a funny thing. Me writing not L.inAccent a few lines below, but not thinking about it to flip on assignment. Sigh.
Thanks.

[Araq]

I'm pretty sure this wouldn't be good enough, identifier comparisons then also have to consider leading and trailing underscores. Too much complexity.