Re: [hybi] Payload only compression extension, again

[John Tamplin <[e-mail]>]

On Wed, Apr 27, 2011 at 8:18 PM, Brodie Thiesfield <[e-mail]>wrote:

> Since you have spent the effort to create this spec, which is much
> better than the current deflate-stream method that we have at the
> moment, wouldn't it be worthwhile to also take over one of the
> reserved bits and use it to specify if the frame is compressed or not?
>
> That way we get the data only compression with an uncompressed data
> option. Pretty much ticks all of the boxes.
>

This was exactly what was proposed a year ago when we were talking about
framing, and we couldn't come to consensus on exactly how it would work
then.  I am not sure there is any reason to believe it is more likely to
achieve consensus now.  Among other considerations, you have to define how
an implementation can choose when to send a compressed frame versus an
uncompressed frame.  Do you allow/require the sender to try and compress it
first and then send it uncompressed if that is smaller?  Doing so might
impose unwelcome headaches for some implementations if they can't easily
save the compression state and revert to a previous checkpoint.

My gut feeling is at this point, it is better to get the base spec released,
experiment with compression options in private-use extensions, and then have
a formal proposal for a real compression extension once we get that
experience.

-- 
John A. Tamplin
Software Engineer (GWT), Google