`syntax::parse::attr` implementation disagrees with BNF in manual

[dcrewi]

According to the manual, multiple comma-separated attributes may be specified within a single pair of square brackets. The parser does not do this at present (02ba8e2). Which one is correct?

Use case:

#[inline, cfg(target_word_size = "32")]
fn assert_fits_in_machine_word() {/*...*/}
#[inline, cfg(target_word_size = "64")]
fn assert_fits_in_machine_word() {/*...*/}

[alexcrichton]

I believe the manual is correct in this case, that looks like it should be a valid attribute.

[huonw]

But, which is the behaviour we want?

[dcrewi]

Yes, I could imagine either scheme becoming blessed as the preferred style, so I feel this needs more discussion if it hasn't already been decided. Either way, there will need to be a test to preserve the preferred style in the future.

[alexcrichton]

Nominating.

[pnkfelix]

Not a 1.0 milestone issue, Asssigning P-low.

[brson]

IMO, we should not accept a comma in attributes. We never used that feature, I've long wanted to remove it, and didn't realize that it was broken.