[regression] empty shebang parsing

[Mark-Simulacrum]

We fail to parse this (admittedly questionable) code as of beta:

https://play.rust-lang.org/?version=nightly&mode=debug&edition=2018&gist=4fd6b726ceaddf0c06e67761d2276277

#!

use std::any;

fn main() {}

This broke one repository (not crate) in the beta crater run: https://crater-reports.s3.amazonaws.com/beta-1.45-1/beta-2020-06-03/gh/grng3r.rust_todo/log.txt

[Mark-Simulacrum]

cc @petrochenkov

I personally suspect that we may want to ask T-lang if we're good accepting this as expected breakage. Alternatively I guess we can adjust the shebang detection to not require a non-empty shebang.

[lcnr]

Assigning P-medium as discussed as part of the Prioritization WG procedure.

[petrochenkov]

For reference, here's the code responsible for parsing shebang in Linux kernel - https://github.com/torvalds/linux/blob/master/fs/binfmt_script.c

(Fully whitespace (including empty) #! line isn't considered a shebang.)

[Mark-Simulacrum]

I am inclined to close as won't fix; let's cc @grng3r as the author of that repository as well.

I'm going to go ahead and nominate this so that we can get a 👍 (or not) at the T-compiler meeting confirming won't fix status.

[petrochenkov]

I made a fix in #73596, which also has a benefit of simplifying the rules slightly.