[SH] use RWS in inner-list

[phluid61]

• update ABNF for inner-list to use RWS
• update parsing algorithm to accept HTAB
• update imports to include RWS and HTAB

For #961

[reschke]

RWS implies "required". Is that really the intent?

[phluid61]

Yes. We require spaces between the items in an inner-list, otherwise they can abut.

[ioggstream]

imho RWS and TABs make things error-prone and difficult to debug.

[phluid61]

Just so we're aware, this PR was to address an issue identified in #961 where the ABNF says OWS (including htab) but the reference algorithm requires SP before OWS after an sh-item.

So changing the ABNF to match the algorithm gives:

inner-list = "(" OWS [ sh-item *( SP OWS sh-item ) [ SP OWS ] ] ")" *parameter

Alternatively, by extending the algorithm to also accept a htab immediately after an item (this PR), the ABNF becomes:

inner-list = "(" OWS [ sh-item *( RWS sh-item ) OWS ")" *parameter

An alternative is to remove the htab altogether, which removes OWS, so we're left with:

inner-list = "(" *SP [ sh-item *( 1*SP sh-item ) *SP ")" *parameter

Which is fine by me, but removing all OWS is a bigger change.

[kazuho]

I think we should retain the use of OWS, rather than removing them or changing (some of) them to RWS.

Not all SH header fields would be generated by a serializer that understands SH. Rather, I'd assume that most of SH header fields would be written by hand, or generated by concatenating strings (in various programming languages).

I do not think we can easily enforce everyone to insert whitespaces is a particular way (or drop them).

[mnot]

@kazuho I'm not as worried about that; tabs in existing headers are extremely uncommon, IME (looking at things like the HTTP Archive dump). If SH implementations are consistent (and they should be if we test well), this will get caught early.

[kazuho]

@mnot I'm not worried about tabs. I think it would be fine if the change is going to be either or both of the following and nothing more.

• OWS -> *SP
• RWS -> 1*SP

[mnot]

@kazuho yep +1

[phluid61]

I made a new branch (with PR) to see what sort of change is involved replacing OWS with *SP. It's not as invasive as I'd thought it might be.

[mnot]

OBE