Re: CFC: Getting rid of 'extras'

I absolutely agree that we should push extras down into a structured
payload object. I have one concern with the proposal. The type in your
examples are the MIME-type of the value embedded in the structured payload,
rather than a type describing the structured payload  itself. Don't we want
to document the literal type of the structured payload object?

Some alternatives:

*MIME-type in payload:*

  type: fancy.org/types/mimedata
  date: {  // mimedata instance
    mimetype: image/png,
    blob: blob image contents,
    url: url of image contents,
    filename: representative filename,
  }

*MIME-type as parameter to structured payload type:*

One problem with including the MIME-type in the payload is that a service
is presumed to be registered against a single type value AND we provide
special handling of MIME-types (wildcards). We can address this issue with
type parameters (similar to those supported by the MIME standard itself).
Basically embed the MIME-type in the type field as a parameter to the real
type.

  type: fancy.org/types/mimedata[image/png]
  data: {  // mimedata instance
    blob: blob image contents,
    url: url of image contents,
    filename: representative filename,
  }

With the MIME-type embedded in the type field service registrations could
still be made using a single type value.

Any way we slice this up, I think we want to keep the mapping between type
and value direct and literal.
Less black magic == easier to document, use, and understand == happier
developers and healthier ecosystem.

Steve



On Wed, Aug 22, 2012 at 4:07 PM, Greg Billock <gbillock@google.com> wrote:

>
> Definitely. The big driver for extras was MIME types, so here's a MIME
> type before:
>
> type: image/png
> data: blob = image contents
> extras: { "url" : image url,
>             "filename" : representative filename,
>           }
>
> and a proposal for a format with no extras:
>
> type: image/png
> data: { "blob" : blob image contents,
>           "url" : url of image contents,
>           "filename" : representative filename,
>         }
>
> Using a map as a recommended base type for data provides a lot of
> flexibility. It provides an open namespace into which new optional
> features can be placed to satisfy new use cases as usage evolves. A
> goal will be to create a consistent vocabulary across types, so for
> example "url" means "a url pointing to the content" for all MIME
> types.
>
>
> On Wed, Aug 22, 2012 at 3:44 PM, James Hawkins <jhawkins@google.com>
> wrote:
> > Can you provide an example to show before and after?  A few disparate
> cases
> > would be helpful to see the impact of the change.
> >
> > Thanks,
> > James
> >
> >
> > On Wed, Aug 22, 2012 at 2:09 PM, Greg Billock <gbillock@google.com>
> wrote:
> >>
> >> In June, we discussed getting rid of getExtra() in favor of an extras
> >> attribute (i.e. [1]). I'm increasingly persuaded that we should just
> >> get rid of extras altogether. When this was proposed [2], it was
> >> because we knew we needed to be able to attach metadata to the payload
> >> data for pre-existing types (such as MIME types). But a better scheme
> >> to deal with that is to create extensible payload formats to begin
> >> with. Such formats have to be documented anyway (see e.g. [3]), and so
> >> it is better to just use the existing structured-clone payload to
> >> describe the type. Any extra header information can just be included
> >> there.
> >>
> >> So the proposal is to strike |getExtra()| from the API, and add a best
> >> practices recommendation that any documented payload type passed
> >> through |data| be a Javascript Object type so that fields may be added
> >> to it as necessity arises.
> >>
> >> This would mean a revision to documents such as [3], but going forward
> >> it puts data passing on a firmer footing where types get designed with
> >> extensibility in mind from the beginning. I don't think we even really
> >> lose any generality -- web developers need to consult documentation to
> >> figure out which extras keys mean what, anyway, and as we've learned
> >> from trying to pass MIME types, there end up being reasons to want to
> >> pass, say, both URLs and Blobs through the type anyway, not to mention
> >> even better future constructs, such as Stream [4], so there isn't much
> >> intuitive power in the type anyway. If you have to consult the
> >> documentation, it's easier to document the type as one object, rather
> >> than as the object and then extras data.
> >>
> >>
> >> [1]
> >>
> http://lists.w3.org/Archives/Public/public-web-intents/2012Jun/0118.html
> >> [2]
> >>
> http://lists.w3.org/Archives/Public/public-web-intents/2012Mar/0003.html
> >> [3] http://www.w3.org/wiki/WebIntents/MIME_Types
> >> [4]
> http://dvcs.w3.org/hg/streams-api/raw-file/a3858f3ef0ae/Overview.htm
> >>
> >
>
>