Re: [Json] draft-williams-json-text-sequence-00

[Nico Williams <[e-mail]>]

On Sat, Mar 15, 2014 at 2:41 PM, John Cowan <[e-mail]> wrote:
>
> 1) Your terminology of "online parsers" and "incremental parsers"
> is very confusing.  What you call an "online parser", I would call a
> "streaming parser", because it returns portions of arrays, objects, and
> possibly even strings as a stream of events.  Your "incremental parser"
> is simply a parser that stops reading when it reaches the end of a JSON
> text without insisting that there be no additional characters.

Oops, in an earlier version (not submitted) I defined these terms, but
after someone made a convincing argument to not require that texts not
have embedded newlines... I removed it.  I didn't remove the remaining
references to online/incremental parsers.

> 2) The sentence "Encoders MUST emit" should refer explicitly to
> the grammar.  Currently the grammar is completely unmoored from the
> requirements.  "Encoders MUST emit JSON text sequences that conform to
> the ABNF" is sufficient.

True.  Also a remnant of an earlier version.  I'll fix this too.

> 3) The sentence "Two contiguous separators do not" is unnecessary,
> because there are no empty JSON texts.

I know this.  I just figure someone will wonder if I don't say it.

> 4) In "Parsers MUST be able", for "to parser" read "to parse".  But as
> in #2, this should simply say "Parsers MUST be able to parse any text
> conforming to the ABNF" for the same reasons.

Sure.

> 5) The recommendation seems entirely unnecessary to me.

Well, it's superfluous if encoders must adhere to the ABNF, that's true.

> 6) I still don't see that there is any ambiguity in "truefalse" considered
> as two JSON-texts.  So I would strip "Parsers MAY parse" to just say
> "sequences where the separators are missing."  Note that such sequences
> don't conform anyway.

It's true that 'trusfalse' does not conform to the ABNF, therefore
it's not a legal sequence.  Removing the options (and recommendation)
would suffice.

> 7) If you want newline-free JSON-texts to be in serious use, the second
> MAY should be promoted to a SHOULD.

I no longer do.  I wasn't ever sure that was a good idea either.

Nico
--