Re: Jesse's feedback on HTTP/2 header compression

On 30 August 2013 08:00, Tatsuhiro Tsujikawa <tatsuhiro.t@gmail.com> wrote:
> And the length is the number of bytes name/value contains, not the number of
> unicode characters. In other words, name/value are actually byte arrays.
> Java String object does not work well in this case.

The way that I would implement this is to create a wrapper for entries
in the table.  That wrapper can track (or calculate) octet length of
the header field.  It increases the memory commitment, but not by
much.

> In practice, I think it will be capped to 4K or 8K.

I suspect that some implementations will want to support more than
this.  We've heard stories of 10Mbyte cookie headers (a little
extreme), but there are also several more common cases that hit
64Kbyte.

Received on Friday, 30 August 2013 17:10:35 UTC