node -c rejects BOM before #!, but node accepts it

[bakkot]

Version: v12.1.0

$ node -e 'require("fs").writeFileSync("test.js", `\u{FEFF}#!`, "utf8")'
$ node -c test.js
(function (exports, require, module, __filename, __dirname) { #!
                                                              ^
SyntaxError: Invalid or unexpected token
$ echo $?
1
$ node test.js
$ echo $?
0

The bug is that check_syntax.js call stripShebang before stripBOM, but the actual loader calls stripBOM before stripShebag.

It's not totally clear which behavior is desired. The original bug which lead to the introduction of stripBOM actually had a script with a BOM preceding the #!, though the test included in that commit doesn't. But unix systems generally require the #! to be the first two bytes for it to be parsed as an interpreter directive, which means it cannot be preceded by a BOM.

[devsnek]

i'd expect a failure if both are present. whether we fail because BOM is invalid js or because a shebang is invalid js seems less relevant. to do this i'd combine them into a stripShebangOrBOM function which operates on the first few bytes of the file all at once so we can't accidentally strip one and then the other.

[bakkot]

(Strictly speaking the BOM is valid - it's parsed as whitespace.)

[devsnek]

that's interesting. i'd still say our logic is wrong here. BOM and shebang only have meaning as the first bytes of the file.

[devsnek]

I made this, if the logic seems alright I'll PR it. 429bf4a

[bakkot]

That allows two BOMs: here and here.

You should probably just make a PR and this sort of code review can go there, though.