W3C home > Mailing lists > Public > ietf-http-wg@w3.org > July to September 2013

Re: JavaScript header compressor/decompressor updated to HPACK-03

From: Fred Akalin <akalin@google.com>
Date: Wed, 18 Sep 2013 14:28:01 -0700
Message-ID: <CANUYc_T-ZRb51eGjyQ9W1AOugN7nD2QQpczca_Pa2GbB+bBC0w@mail.gmail.com>
To: Roberto Peon <grmocg@gmail.com>
Cc: HTTP Working Group <ietf-http-wg@w3.org>
Just to clarify, is the oldest item the one with index 0, or some other
definition of oldest?


On Fri, Sep 13, 2013 at 3:16 PM, Roberto Peon <grmocg@gmail.com> wrote:

> Awesome!
>
> The expiry mechanism should always be oldest item first.
>
> -=R
> On Sep 13, 2013 2:53 PM, "Fred Akalin" <akalin@google.com> wrote:
>
>> Hey all,
>>
>> I updated
>> http://akalin-chromium.github.io/httpbis-header-compression/compressor_test.html to
>> implement the HPACK-03 draft. In particular, I tried to make it a complete
>> an implementation as possible, and I added copious comments and references
>> to the spec to make it easy to validate and understand.
>>
>> The only thing I didn't implement is UTF-8 validation for header values.
>> Hopefully, the need for that will go away.
>>
>> Some thoughts:
>>
>> - There aren't any tests. I wanted to see how correct I can make the
>> implementation without them (which will be measured when the compliance
>> suite comes out). I'm sure there are bugs.
>>
>> - I didn't try very hard to make the encoder smart, but I did try to make
>> it exercise all the opcodes.
>>
>> - I found it quite helpful that the encoding context was precisely
>> defined (as a header table plus the reference set). However, I ultimately
>> found it better to encode the reference set as part of the header table (by
>> having a bit per entry) instead of having a separate data structure, since
>> it eliminates a bunch of logic to keep the indices in the two in sync. This
>> may have been obvious to some people, but not to me. I wonder if it's in
>> the scope of the spec to suggest this.
>>
>> - I also found it helpful to have a 'touch' flag per entry since
>> encoding/decoding requires processing of the untouched subset of the
>> reference set.
>>
>> - For encoding I also needed to keep track of the number of touches
>> (representing the number of times the entry would be explicitly emitted),
>> and I needed to make a distinction between no touches and 0 touches
>> (representing an implicit emission). This is to support duplicate headers,
>> which was tricky to get right.
>>
>> - It would be nice to have explicit bounds for encoded integers, string
>> lengths, header lengths, etc. I didn't try to make the encoder/decoder
>> streaming, since that would complicate the implementation, but it seems
>> difficult to guarantee memory bounds without the above explicit bounds.
>>
>> - It would be nice to clarify the behavior when the max header table size
>> is reduced. I just implemented popping from the front until the new bound
>> is satisfied.
>>
>> - I didn't find the need to encode index vs. index + 1 too confusing this
>> time around. I feel like making the header table start at 1 would simply
>> move the off-by-one bugs someplace else. I don't feel too strongly about
>> this, though.
>>
>> Comments, pull requests, etc. welcome!
>>
>> -- Fred
>>
>
Received on Wednesday, 18 September 2013 21:29:24 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 17:14:15 UTC