| Network Working Group | J.M. Snell |
| Internet-Draft | April 23, 2013 |
| Intended status: Informational | |
| Expires: October 25, 2013 |
HTTP/2.0 Discussion: Stored Header Encoding
draft-snell-httpbis-bohe-07
This memo describes a proposed alternative encoding for headers that combines the best concepts from the proposed Delta and HeaderDiff options with the typed value codecs introduced by previous versions of this draft.
This Internet-Draft is submitted to IETF in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on October 25, 2013.
Copyright (c) 2013 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document.
The Stored Header Encoding is an alternative "binary header encoding" for HTTP/2.0 that combines the best elements from three other proposed encodings, including:
The Stored Header Encoding seeks to find an elegant, efficient and simple marriage of the best concepts from each of these separate proposals.
The compressor and decompressor each maintain a cache of header value pairs. There is a static cache, prepopulated by the specification, and a dynamic cache, populated through the compression and decompression process. Each cache contains a maximum of 128 individual key+value pairs.
Each item in the index is referenced by an 8-bit identifier. The most significant bit identifies whether an item from the static or dynamic cache is being referenced. Note: the Nil byte (0x00) is a valid identifier for the dynamic cache.
0xxxxxxx -- Dynamic Cache
1xxxxxxx -- Static Cache
The dynamic cache is managed in a "least recently written" style, that is, as the cache fills to capacity in both number of entries and maximum stored byte size, the least recently written items are dropped and those index positions are reused.
Index positions from the dynamic cache are assigned in "encounter order", beginning from 0x00 and increasing monotonically to 0x7F. That is to say, the positions are assigned in precisely the same order that they are serialized, and thereby encountered by the decompressor upon reading and processing the block.
Each item in the store consists of a Header Name and a Value. The Name is a lower-case ISO-8859-1 character sequence. The Value is either a UTF-8 string, a number, a Timestamp or an arbitrary sequence of binary octets.
The available size of the stored compression state can be capped by the decompressor. Each stored value contributes to the accumulated size of the storage state. As new key+value pairs are assigned positions in the dynamic cache, the least-recently assigned items must be removed if necessary to free up the required space.
The size of string values is measured by the number of UTF-8 bytes required for the character sequence.
The size of number and timestamp values are measured by the number of unsigned variable length integer (uvarint) encoded bytes it takes to represent the value (see the section of value types below).
The size of raw binary values is measured by the number of octets.
Header names DO NOT contribute to the stored state size of the compressor; only the size of the value is considered. Duplicate values MUST be counted individually.
Headers are serialized into four typed header groups, each represented by a two-bit identifier. These groups are serialized sequentially. A serialized header block can contain, at most 256 header groups. The first byte of the serialized block is an unsigned, 0-based counter indicating the number of groups. A serialized block MUST contain at least one header group.
00 -- Index Header Group
01 -- Index Range Header Group
10 -- Cloned Index Header Group
11 -- Literal Header Group
The Cloned Index (10) and Literal (11) header group types have an additional "ephemeral" property that indicates whether or not the group affects the compression state.
Each header group contains a single 8-bit prefix and up to 32 distinct header instances.
Wire Format
header-block = OCTET *(index-header-group /
index-range-header-group /
cloned-index-header-group /
literal-header-group)
; Header Group Prefix = 8 bits ...
; First two bits = header-group-type
; Third bit = ephemeral flag
; Final five bits = instance counter
;
index-header-group-type = 00
index-range-group-type = 01
cloned-index-group-type = 10
literal-group-type = 11
count = 5bit
index-header-prefix = index-header-group-type
unset count ; 000xxxxx
index-range-header-prefix = index-header-group-type
unset count ; 010xxxxx
cloned-index-header-prefix = cloned-index-group-type
bit count ; 10?xxxxx
literal-header-prefix = literal-group-type
bit count ; 11?xxxxx
; Cache Index Identifier = 8 bits ...
; 0xxxxxxx = Dynamic Cache Identifier
; 1xxxxxxx = Static Cache Identifier
cache-index = %x00-FF
; Index Header Group
index-header-group = index-header-prefix
1*32cache-index
; Index-Range Header Group
; Contains a pair of cache-index values, second MUST
; me strictly higher in value than the first...
index-range-header-group = index-range-header-prefix
1*32(cache-index cache-index)
; Cloned-Index Header Group
cloned-index-header-group = cloned-index-header-prefix
1*32(cache-index value)
; Literal Header Group
literal-header-group = literal-header-prefix
1*32(name value)
value = text-value /
number-value /
timestamp-value /
binary-value
text-value-type = 00 ; two bits
number-value-type = 01
timestamp-value-type = 10
binary-value-type = 11
text-value-prefix = text-value-type
unset count ; 000xxxxx
number-value-prefix = number-value-type
unset count ; 010xxxxx
timestamp-value-prefix = timestamp-value-type
unset count ; 100xxxxx
binary-value-prefix = binary-value-type
unset count ; 110xxxxx
text-value = text-value-prefix *32string
number-value = number-value-type *32uvarint
timestamp-value = timestamp-value-prefix *32uvarint
binary-value = binary-value-prefix uvarint *OCTET
uvarint = *uvarint-continuation uvarint-final
uvarint-continuation = %x80-FF
uvarint-final = %x00-7F
name = 1*tchar
tchar = "!" / "#" / "$" / "%" / "&" /
"'" / "*" / "+" / "-" / "." /
"^" / "_" / "`" / "|" / "~"
/ DIGIT / ALPHA
string = uvarint *(HUFFMAN-ENCODED-CHAR)
HUFFMAN-EOF
padding-to-nearest-byte;
padding-to-nearest-byte = *7unset
bit = set / unset
unset = 0
set = 1
The Header Group Prefix is a single octet that provides three distinct pieces of information:
00 0 00000
The first two most significant bits of the header group prefix identify the group type.
The next bit is the "ephemeral flag" and is used only for Cloned and Literal group types. This bit indicates whether or not the group alters the stored compression state.
The remaining five bits specify the number of header instances in the group, with 00000 indicating that the group contains 1 instance and 11111 contains 32. A header group MUST contain at least one instance.
The remaining serialization of the header group depends entirely on the group type.
The serialization of the Index Header Group consists of the Header Group Prefix and up to 32 additional octets, each referencing a single 8-bit storage index identifier for items in either the Static or Dynamic Cache.
For instance
00000000 00000000 = References item #0 from
the dynamic cache
00000001 00000000 10000000 = References item #0 from the
dynamic cache and item #0
from the static cache
Index Header Groups do not affect the stored compression state. If an Index Header Group references a header index that has not yet been allocated, the deserialization MUST terminate with an error. This likely means that the compression state has become out of sync and needs to be reestablished.
The serialization of the Index Range Header Group consists of the Header Group Prefix and up to 32 additional 2-octet (16 bits) pairs of 8-bit storage index identifiers. Each pair specifies a sequential range of adjacent ranges.
For instance:
01000000 00000000 00000100 = References items #0-#4 from
the dynamic cache.
(five distinct items total)
A range MAY span dynamic and static index values. Index values are treated as unsigned byte values, so indices from the static cache are numerically greater than dynamic cache values.. e.g.
01000000 01111111 10000001 = References item #127 from the
dynamic cache, and items #0
and #1 from the static cache.
Index Range Header Groups do not affect the stored compression state. If a range references a header index that has not yet been allocated, the deserialization MUST terminate with an error. This likely means that the compression state has become out of sync and needs to be reestablished.
The serialization of the Cloned Index Header Group consists of the Header Group Prefix and up to 32 Index+Value pairs. Each Index+Value pair consists of a leading 8-bit storage index of an existing stored header followed by a new serialized value. The serialization of the value depends on the value type (see discussion of Value serialization below).
The Cloned Header Group affects the stored compression state if, and only if, the "ephemeral" flag in the Header Group Prefix is NOT set. If the header group is not marked as being ephemeral, then the specified value is stored in the next available storage index using the key name from the referenced storage index.
For instance, assume the dynamic cache currently contains an item at index #1 with key name "foo" and value "bar", the following causes a new item to be added to the storage with key name "foo" and value "baz":
10000000 00000001 00000000 00000100
10111000 01001111 10110101 00100000
An explanation of the value syntax is given a bit later.
If a Cloned Header Group references a header index that has not yet been allocated, the deserialization MUST terminate with an error. This likely means that the compression state has become out of sync and needs to be reestablished.
The serialization of the Literal Header Group consists of the Header Group Prefix and up to 32 Name+Value pairs. Each Name+Value pair consists of a length-prefixed sequence of ASCII bytes specifying the Header Name followed by the serialized value. The serialization of the value depends on the value type. The length prefix is encoded as an unsigned variable length integer (uvarint). The length prefix SHOULD NOT be longer than five octets and SHOULD NOT specify a value larger than 0xFFFF.
The Literal Heaer Group affects the stored compression state if, and only if, the "ephemeral" flag in the Header Group Prefix is NOT set. If the header group is not marked as being ephemeral, then the specified key name and value is stored in the next available storage index.
For instance:
11000000 00000011 01100110 01101111
01101111 00000000 00000010 10111000
01000100 11010010
Stores a new header with name "foo" and value "baz" in the dynamic cache.
Each Header Group consists of up to 32 distinct Header Instances. If a particular serialization block contains more than 32 intances of a given type, then multiple instances of the Header Group Type can be included in the serialized block. For instance, if a given message contains 33 index references, the serialized block may contain two separate Index Header Groups. While this is allowed, it is expected to be rare.
Header Values can be one of four types, each identified by a two-bit identifier.
An individual value MAY consist of up to 32 distinct discreet "value instances". A value with multiple instances is considered, for all intensive purposes, to be a single value.
Each serialized value is preceded by an 8-bit Value Prefix.
00 0 00000
The first two most significant bits specifies the value type.
The third significant bit is a reserved flag. Future iterations of this specification might make use of this bit.
The final five least-significant bits specify the number of discreet instances in the value. 00000 indicates that one instance is included, 11111 indicates that 32 instances are included. The value MUST contain at least one instance.
The remaining serialization depends entirely on the type.
UTF-8 Text is encoded as a length-prefixed sequence of Huffman-encoded UTF-8 octets. The length prefix is encoded as an unsigned variable-length integer specifying the number of octets after applying the Huffman-encoding.
Numeric values are encoded as unsigned variable-length integers (uvarint) of up to a maximum of 10-octets. Unsigned values larger than 64-bits (0xFFFFFFFFFFFFFFFF) are supported by this format but SHOULD NOT be used. Negative values cannot be represented using this syntax. The uvarint syntax is described below.
Timestamp values are encoded as unsigned variable-length integers specifying the number of milliseconds that have passed since the standard Epoch (1970-01-01T00:00:00 GMT). The syntax is identical that used for Numeric Values. Dates prior to the epoch cannot be represented using this syntax.
Representing timestamps in this manner ensures that timestamps will always encode using six bytes up and until 2109-05-15T07:35:00 GMT, then as seven bytes up to and until 19809-03-05T11:03:41 GMT.
Binary values are encoded as a length prefixed sequence of arbitrary octets. The length prefix is encoded as an unsigned variable length integer.
The uvarint syntax is identical to that used by Google's protobufs. They are serialized with the least-significant bytes first in batches of 7-bits, with the most significant bit per byte reserved as a continuation bit. Values less than or equal to 127 are serialized using at most one byte; values less than or equal to 16383 are serialized using at most two bytes; values less than or equal to 2097151 are serialized using at most three bytes.
def uvarint(num):
return [] if num = 0
ret = []
while(num != 0):
m = num >>> 7 ; unsigned shift left 7 bits
ret.push (byte)((num & ~0x80) | ( m > 0 ? 0x80 : 0x00 ));
num = m;
return ret;
For example, the binary representation of the 32-bit integer 217 is:
00000000 00000000 00000000 11011001
The variable length encoding is:
11011001 00000001
The binary representation of the 32-bit integer 1386210052 is:
01010010 10011111 11100011 00000100
The variable length encoding is:
10000100 11000110 11111111 10010100 00000101
All UTF-8 text values are compressed using a modified static huffman code. "Modified" because the encoded version may contain compact-representations of raw, arbitrary UTF-8 bytes that are not covered by the static huffman code table.
There are two huffman tables in use, one for HTTP Requests and another for HTTP Responses, each covers UTF-8 codepoints strictly less than 128 as well the fifty possible UTF-8 leading octets.
The encoded result MUST end with a specific terminal sequence of bits called the "HUFFMAN_EOF". Currently, the HUFFMAN_EOF is the same for both the Request and Response tables, but that could change if the tables are regenerated. Currently, the HUFFMAN_EOF sequence is 101001.
Codepoints >= 128 are handled by first taking the leading octet of the UTF-8 representation and serializing it's associated huffman code from the table to the output stream, then, depending on the octets value, serializing the six least significant bits from each of the remaining trailing octets.
For instance, the UTF-8 character U+00D4 (LATIN CAPITAL LETTER O WITH CIRCUMFLEX), with UTF-8 representation of C394 (hex) is encoded as:
11000100 01010010 10010000
The first 8-bits represents the huffman-table prefix, the first six most significant bytes of the second octet are taken directly from the six least significant bits of the second UTF-8 byte (0x94). Following those six bits are the six bits of the HUFFMAN_EOF 101001, followed by four unset padding bits.
The number of raw UTF-8 bits to write depends on the value of the leading octet. If the value is between 0xC2 and 0xDF (inclusive), six bits from the second continuation byte is encoded. If the value is between 0xE0 and 0xEF (inclusive), six bits from the second and third continuation bytes are encoded. If the value is between 0xF0 and 0xF4 (inclusive), six bits from the second, third and fourth continuation bytes are encoded. UTF-8 codepoints that require greater than four bytes to encode cannot be represented.
TBD
| [RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. |
| [RFC6265] | Barth, A., "HTTP State Management Mechanism", RFC 6265, April 2011. |
Request Table
( 0) |11111111|11111111|11111111|0 [25] 1fffffe [25]
( 1) |11111111|11111111|11111111|1 [25] 1ffffff [25]
( 2) |11111111|11111111|11100000 [24] ffffe0 [24]
( 3) |11111111|11111111|11100001 [24] ffffe1 [24]
( 4) |11111111|11111111|11100010 [24] ffffe2 [24]
( 5) |11111111|11111111|11100011 [24] ffffe3 [24]
( 6) |11111111|11111111|11100100 [24] ffffe4 [24]
( 7) |11111111|11111111|11100101 [24] ffffe5 [24]
( 8) |11111111|11111111|11100110 [24] ffffe6 [24]
( 9) |11111111|11111111|11100111 [24] ffffe7 [24]
( 10) |11111111|11111111|11101000 [24] ffffe8 [24]
( 11) |11111111|11111111|11101001 [24] ffffe9 [24]
( 12) |11111111|11111111|11101010 [24] ffffea [24]
( 13) |11111111|11111111|11101011 [24] ffffeb [24]
( 14) |11111111|11111111|11101100 [24] ffffec [24]
( 15) |11111111|11111111|11101101 [24] ffffed [24]
( 16) |11111111|11111111|11101110 [24] ffffee [24]
( 17) |11111111|11111111|11101111 [24] ffffef [24]
( 18) |11111111|11111111|11110000 [24] fffff0 [24]
( 19) |11111111|11111111|11110001 [24] fffff1 [24]
( 20) |11111111|11111111|11110010 [24] fffff2 [24]
( 21) |11111111|11111111|11110011 [24] fffff3 [24]
( 22) |11111111|11111111|11110100 [24] fffff4 [24]
( 23) |11111111|11111111|11110101 [24] fffff5 [24]
( 24) |11111111|11111111|11110110 [24] fffff6 [24]
( 25) |11111111|11111111|11110111 [24] fffff7 [24]
( 26) |11111111|11111111|11111000 [24] fffff8 [24]
( 27) |11111111|11111111|11111001 [24] fffff9 [24]
( 28) |11111111|11111111|11111010 [24] fffffa [24]
( 29) |11111111|11111111|11111011 [24] fffffb [24]
( 30) |11111111|11111111|11111100 [24] fffffc [24]
( 31) |11111111|11111111|11111101 [24] fffffd [24]
' ' ( 32) |11111111|0110 [12] ff6 [12]
'!' ( 33) |11111111|0111 [12] ff7 [12]
'"' ( 34) |11111111|111010 [14] 3ffa [14]
'#' ( 35) |11111111|1111100 [15] 7ffc [15]
'$' ( 36) |11111111|1111101 [15] 7ffd [15]
'%' ( 37) |011000 [6] 18 [6]
'&' ( 38) |1010100 [7] 54 [7]
''' ( 39) |11111111|1111110 [15] 7ffe [15]
'(' ( 40) |11111111|1000 [12] ff8 [12]
')' ( 41) |11111111|1001 [12] ff9 [12]
'*' ( 42) |11111111|1010 [12] ffa [12]
'+' ( 43) |11111111|1011 [12] ffb [12]
',' ( 44) |11111011|10 [10] 3ee [10]
'-' ( 45) |011001 [6] 19 [6]
'.' ( 46) |00010 [5] 2 [5]
'/' ( 47) |00011 [5] 3 [5]
'0' ( 48) |011010 [6] 1a [6]
'1' ( 49) |011011 [6] 1b [6]
'2' ( 50) |011100 [6] 1c [6]
'3' ( 51) |011101 [6] 1d [6]
'4' ( 52) |1010101 [7] 55 [7]
'5' ( 53) |1010110 [7] 56 [7]
'6' ( 54) |1010111 [7] 57 [7]
'7' ( 55) |1011000 [7] 58 [7]
'8' ( 56) |1011001 [7] 59 [7]
'9' ( 57) |1011010 [7] 5a [7]
':' ( 58) |011110 [6] 1e [6]
';' ( 59) |11111011|11 [10] 3ef [10]
'<' ( 60) |11111111|11111111|10 [18] 3fffe [18]
'=' ( 61) |011111 [6] 1f [6]
'>' ( 62) |11111111|11111110|0 [17] 1fffc [17]
'?' ( 63) |11110110|0 [9] 1ec [9]
'@' ( 64) |11111111|11100 [13] 1ffc [13]
'A' ( 65) |10111010 [8] ba [8]
'B' ( 66) |11110110|1 [9] 1ed [9]
'C' ( 67) |10111011 [8] bb [8]
'D' ( 68) |10111100 [8] bc [8]
'E' ( 69) |11110111|0 [9] 1ee [9]
'F' ( 70) |10111101 [8] bd [8]
'G' ( 71) |11111100|00 [10] 3f0 [10]
'H' ( 72) |11111100|01 [10] 3f1 [10]
'I' ( 73) |11110111|1 [9] 1ef [9]
'J' ( 74) |11111100|10 [10] 3f2 [10]
'K' ( 75) |11111111|010 [11] 7fa [11]
'L' ( 76) |11111100|11 [10] 3f3 [10]
'M' ( 77) |11111000|0 [9] 1f0 [9]
'N' ( 78) |11111101|00 [10] 3f4 [10]
'O' ( 79) |11111101|01 [10] 3f5 [10]
'P' ( 80) |11111000|1 [9] 1f1 [9]
'Q' ( 81) |11111101|10 [10] 3f6 [10]
'R' ( 82) |11111001|0 [9] 1f2 [9]
'S' ( 83) |11111001|1 [9] 1f3 [9]
'T' ( 84) |11111010|0 [9] 1f4 [9]
'U' ( 85) |11111101|11 [10] 3f7 [10]
'V' ( 86) |11111110|00 [10] 3f8 [10]
'W' ( 87) |11111110|01 [10] 3f9 [10]
'X' ( 88) |11111110|10 [10] 3fa [10]
'Y' ( 89) |11111110|11 [10] 3fb [10]
'Z' ( 90) |11111111|00 [10] 3fc [10]
'[' ( 91) |11111111|111011 [14] 3ffb [14]
'\' ( 92) |11111111|11111111|11111110 [24] fffffe [24]
']' ( 93) |11111111|111100 [14] 3ffc [14]
'^' ( 94) |11111111|111101 [14] 3ffd [14]
'_' ( 95) |1011011 [7] 5b [7]
'`' ( 96) |11111111|11111111|110 [19] 7fffe [19]
'a' ( 97) |00100 [5] 4 [5]
'b' ( 98) |1011100 [7] 5c [7]
'c' ( 99) |00101 [5] 5 [5]
'd' (100) |100000 [6] 20 [6]
'e' (101) |0000 [4] 0 [4]
'f' (102) |100001 [6] 21 [6]
'g' (103) |100010 [6] 22 [6]
'h' (104) |100011 [6] 23 [6]
'i' (105) |00110 [5] 6 [5]
'j' (106) |10111110 [8] be [8]
'k' (107) |10111111 [8] bf [8]
'l' (108) |100100 [6] 24 [6]
'm' (109) |100101 [6] 25 [6]
'n' (110) |100110 [6] 26 [6]
'o' (111) |00111 [5] 7 [5]
'p' (112) |01000 [5] 8 [5]
'q' (113) |11111010|1 [9] 1f5 [9]
'r' (114) |01001 [5] 9 [5]
's' (115) |01010 [5] a [5]
't' (116) |01011 [5] b [5]
'u' (117) |100111 [6] 27 [6]
'v' (118) |11000000 [8] c0 [8]
'w' (119) |101000 [6] 28 [6]
'x' (120) |11000001 [8] c1 [8]
'y' (121) |11000010 [8] c2 [8]
'z' (122) |11111011|0 [9] 1f6 [9]
'{' (123) |11111111|11111110|1 [17] 1fffd [17]
'|' (124) |11111111|1100 [12] ffc [12]
'}' (125) |11111111|11111111|0 [17] 1fffe [17]
'~' (126) |11111111|1101 [12] ffd [12]
(127) |101001 [6] 29 [6]
(0xC2) |11000011 [8] c3 [8]
(0xC3) |11000100 [8] c4 [8]
(0xC4) |11000101 [8] c5 [8]
(0xC5) |11000110 [8] c6 [8]
(0xC6) |11000111 [8] c7 [8]
(0xC7) |11001000 [8] c8 [8]
(0xC8) |11001001 [8] c9 [8]
(0xC9) |11001010 [8] ca [8]
(0xCA) |11001011 [8] cb [8]
(0xCB) |11001100 [8] cc [8]
(0xCC) |11001101 [8] cd [8]
(0xCD) |11001110 [8] ce [8]
(0xCE) |11001111 [8] cf [8]
(0xCF) |11010000 [8] d0 [8]
(0xD0) |11010001 [8] d1 [8]
(0xD1) |11010010 [8] d2 [8]
(0xD2) |11010011 [8] d3 [8]
(0xD3) |11010100 [8] d4 [8]
(0xD4) |11010101 [8] d5 [8]
(0xD5) |11010110 [8] d6 [8]
(0xD6) |11010111 [8] d7 [8]
(0xD7) |11011000 [8] d8 [8]
(0xD8) |11011001 [8] d9 [8]
(0xD9) |11011010 [8] da [8]
(0xDA) |11011011 [8] db [8]
(0xDB) |11011100 [8] dc [8]
(0xDC) |11011101 [8] dd [8]
(0xDD) |11011110 [8] de [8]
(0xDE) |11011111 [8] df [8]
(0xDF) |11100000 [8] e0 [8]
(0xE0) |11100001 [8] e1 [8]
(0xE1) |11100010 [8] e2 [8]
(0xE2) |11100011 [8] e3 [8]
(0xE3) |11100100 [8] e4 [8]
(0xE4) |11100101 [8] e5 [8]
(0xE5) |11100110 [8] e6 [8]
(0xE6) |11100111 [8] e7 [8]
(0xE7) |11101000 [8] e8 [8]
(0xE8) |11101001 [8] e9 [8]
(0xE9) |11101010 [8] ea [8]
(0xEA) |11101011 [8] eb [8]
(0xEB) |11101100 [8] ec [8]
(0xEC) |11101101 [8] ed [8]
(0xED) |11101110 [8] ee [8]
(0xEE) |11101111 [8] ef [8]
(0xEF) |11110000 [8] f0 [8]
(0xF0) |11110001 [8] f1 [8]
(0xF1) |11110010 [8] f2 [8]
(0xF2) |11110011 [8] f3 [8]
(0xF3) |11110100 [8] f4 [8]
(0xF4) |11110101 [8] f5 [8]
Response Table:
( 0) |11111111|11111111|11111111|0 [25] 1fffffe [25]
( 1) |11111111|11111111|11111111|1 [25] 1ffffff [25]
( 2) |11111111|11111111|11100000 [24] ffffe0 [24]
( 3) |11111111|11111111|11100001 [24] ffffe1 [24]
( 4) |11111111|11111111|11100010 [24] ffffe2 [24]
( 5) |11111111|11111111|11100011 [24] ffffe3 [24]
( 6) |11111111|11111111|11100100 [24] ffffe4 [24]
( 7) |11111111|11111111|11100101 [24] ffffe5 [24]
( 8) |11111111|11111111|11100110 [24] ffffe6 [24]
( 9) |11111111|11111111|11100111 [24] ffffe7 [24]
( 10) |11111111|11111111|11101000 [24] ffffe8 [24]
( 11) |11111111|11111111|11101001 [24] ffffe9 [24]
( 12) |11111111|11111111|11101010 [24] ffffea [24]
( 13) |11111111|11111111|11101011 [24] ffffeb [24]
( 14) |11111111|11111111|11101100 [24] ffffec [24]
( 15) |11111111|11111111|11101101 [24] ffffed [24]
( 16) |11111111|11111111|11101110 [24] ffffee [24]
( 17) |11111111|11111111|11101111 [24] ffffef [24]
( 18) |11111111|11111111|11110000 [24] fffff0 [24]
( 19) |11111111|11111111|11110001 [24] fffff1 [24]
( 20) |11111111|11111111|11110010 [24] fffff2 [24]
( 21) |11111111|11111111|11110011 [24] fffff3 [24]
( 22) |11111111|11111111|11110100 [24] fffff4 [24]
( 23) |11111111|11111111|11110101 [24] fffff5 [24]
( 24) |11111111|11111111|11110110 [24] fffff6 [24]
( 25) |11111111|11111111|11110111 [24] fffff7 [24]
( 26) |11111111|11111111|11111000 [24] fffff8 [24]
( 27) |11111111|11111111|11111001 [24] fffff9 [24]
( 28) |11111111|11111111|11111010 [24] fffffa [24]
( 29) |11111111|11111111|11111011 [24] fffffb [24]
( 30) |11111111|11111111|11111100 [24] fffffc [24]
( 31) |11111111|11111111|11111101 [24] fffffd [24]
' ' ( 32) |11111111|0110 [12] ff6 [12]
'!' ( 33) |11111111|0111 [12] ff7 [12]
'"' ( 34) |11111111|111010 [14] 3ffa [14]
'#' ( 35) |11111111|1111100 [15] 7ffc [15]
'$' ( 36) |11111111|1111101 [15] 7ffd [15]
'%' ( 37) |011000 [6] 18 [6]
'&' ( 38) |1010100 [7] 54 [7]
''' ( 39) |11111111|1111110 [15] 7ffe [15]
'(' ( 40) |11111111|1000 [12] ff8 [12]
')' ( 41) |11111111|1001 [12] ff9 [12]
'*' ( 42) |11111111|1010 [12] ffa [12]
'+' ( 43) |11111111|1011 [12] ffb [12]
',' ( 44) |11111011|10 [10] 3ee [10]
'-' ( 45) |011001 [6] 19 [6]
'.' ( 46) |00010 [5] 2 [5]
'/' ( 47) |00011 [5] 3 [5]
'0' ( 48) |011010 [6] 1a [6]
'1' ( 49) |011011 [6] 1b [6]
'2' ( 50) |011100 [6] 1c [6]
'3' ( 51) |011101 [6] 1d [6]
'4' ( 52) |1010101 [7] 55 [7]
'5' ( 53) |1010110 [7] 56 [7]
'6' ( 54) |1010111 [7] 57 [7]
'7' ( 55) |1011000 [7] 58 [7]
'8' ( 56) |1011001 [7] 59 [7]
'9' ( 57) |1011010 [7] 5a [7]
':' ( 58) |011110 [6] 1e [6]
';' ( 59) |11111011|11 [10] 3ef [10]
'<' ( 60) |11111111|11111111|10 [18] 3fffe [18]
'=' ( 61) |011111 [6] 1f [6]
'>' ( 62) |11111111|11111110|0 [17] 1fffc [17]
'?' ( 63) |11110110|0 [9] 1ec [9]
'@' ( 64) |11111111|11100 [13] 1ffc [13]
'A' ( 65) |10111010 [8] ba [8]
'B' ( 66) |11110110|1 [9] 1ed [9]
'C' ( 67) |10111011 [8] bb [8]
'D' ( 68) |10111100 [8] bc [8]
'E' ( 69) |11110111|0 [9] 1ee [9]
'F' ( 70) |10111101 [8] bd [8]
'G' ( 71) |11111100|00 [10] 3f0 [10]
'H' ( 72) |11111100|01 [10] 3f1 [10]
'I' ( 73) |11110111|1 [9] 1ef [9]
'J' ( 74) |11111100|10 [10] 3f2 [10]
'K' ( 75) |11111111|010 [11] 7fa [11]
'L' ( 76) |11111100|11 [10] 3f3 [10]
'M' ( 77) |11111000|0 [9] 1f0 [9]
'N' ( 78) |11111101|00 [10] 3f4 [10]
'O' ( 79) |11111101|01 [10] 3f5 [10]
'P' ( 80) |11111000|1 [9] 1f1 [9]
'Q' ( 81) |11111101|10 [10] 3f6 [10]
'R' ( 82) |11111001|0 [9] 1f2 [9]
'S' ( 83) |11111001|1 [9] 1f3 [9]
'T' ( 84) |11111010|0 [9] 1f4 [9]
'U' ( 85) |11111101|11 [10] 3f7 [10]
'V' ( 86) |11111110|00 [10] 3f8 [10]
'W' ( 87) |11111110|01 [10] 3f9 [10]
'X' ( 88) |11111110|10 [10] 3fa [10]
'Y' ( 89) |11111110|11 [10] 3fb [10]
'Z' ( 90) |11111111|00 [10] 3fc [10]
'[' ( 91) |11111111|111011 [14] 3ffb [14]
'\' ( 92) |11111111|11111111|11111110 [24] fffffe [24]
']' ( 93) |11111111|111100 [14] 3ffc [14]
'^' ( 94) |11111111|111101 [14] 3ffd [14]
'_' ( 95) |1011011 [7] 5b [7]
'`' ( 96) |11111111|11111111|110 [19] 7fffe [19]
'a' ( 97) |00100 [5] 4 [5]
'b' ( 98) |1011100 [7] 5c [7]
'c' ( 99) |00101 [5] 5 [5]
'd' (100) |100000 [6] 20 [6]
'e' (101) |0000 [4] 0 [4]
'f' (102) |100001 [6] 21 [6]
'g' (103) |100010 [6] 22 [6]
'h' (104) |100011 [6] 23 [6]
'i' (105) |00110 [5] 6 [5]
'j' (106) |10111110 [8] be [8]
'k' (107) |10111111 [8] bf [8]
'l' (108) |100100 [6] 24 [6]
'm' (109) |100101 [6] 25 [6]
'n' (110) |100110 [6] 26 [6]
'o' (111) |00111 [5] 7 [5]
'p' (112) |01000 [5] 8 [5]
'q' (113) |11111010|1 [9] 1f5 [9]
'r' (114) |01001 [5] 9 [5]
's' (115) |01010 [5] a [5]
't' (116) |01011 [5] b [5]
'u' (117) |100111 [6] 27 [6]
'v' (118) |11000000 [8] c0 [8]
'w' (119) |101000 [6] 28 [6]
'x' (120) |11000001 [8] c1 [8]
'y' (121) |11000010 [8] c2 [8]
'z' (122) |11111011|0 [9] 1f6 [9]
'{' (123) |11111111|11111110|1 [17] 1fffd [17]
'|' (124) |11111111|1100 [12] ffc [12]
'}' (125) |11111111|11111111|0 [17] 1fffe [17]
'~' (126) |11111111|1101 [12] ffd [12]
(127) |101001 [6] 29 [6]
(0xC2) |11000011 [8] c3 [8]
(0xC3) |11000100 [8] c4 [8]
(0xC4) |11000101 [8] c5 [8]
(0xC5) |11000110 [8] c6 [8]
(0xC6) |11000111 [8] c7 [8]
(0xC7) |11001000 [8] c8 [8]
(0xC8) |11001001 [8] c9 [8]
(0xC9) |11001010 [8] ca [8]
(0xCA) |11001011 [8] cb [8]
(0xCB) |11001100 [8] cc [8]
(0xCC) |11001101 [8] cd [8]
(0xCD) |11001110 [8] ce [8]
(0xCE) |11001111 [8] cf [8]
(0xCF) |11010000 [8] d0 [8]
(0xD0) |11010001 [8] d1 [8]
(0xD1) |11010010 [8] d2 [8]
(0xD2) |11010011 [8] d3 [8]
(0xD3) |11010100 [8] d4 [8]
(0xD4) |11010101 [8] d5 [8]
(0xD5) |11010110 [8] d6 [8]
(0xD6) |11010111 [8] d7 [8]
(0xD7) |11011000 [8] d8 [8]
(0xD8) |11011001 [8] d9 [8]
(0xD9) |11011010 [8] da [8]
(0xDA) |11011011 [8] db [8]
(0xDB) |11011100 [8] dc [8]
(0xDC) |11011101 [8] dd [8]
(0xDD) |11011110 [8] de [8]
(0xDE) |11011111 [8] df [8]
(0xDF) |11100000 [8] e0 [8]
(0xE0) |11100001 [8] e1 [8]
(0xE1) |11100010 [8] e2 [8]
(0xE2) |11100011 [8] e3 [8]
(0xE3) |11100100 [8] e4 [8]
(0xE4) |11100101 [8] e5 [8]
(0xE5) |11100110 [8] e6 [8]
(0xE6) |11100111 [8] e7 [8]
(0xE7) |11101000 [8] e8 [8]
(0xE8) |11101001 [8] e9 [8]
(0xE9) |11101010 [8] ea [8]
(0xEA) |11101011 [8] eb [8]
(0xEB) |11101100 [8] ec [8]
(0xEC) |11101101 [8] ed [8]
(0xED) |11101110 [8] ee [8]
(0xEE) |11101111 [8] ef [8]
(0xEF) |11110000 [8] f0 [8]
(0xF0) |11110001 [8] f1 [8]
(0xF1) |11110010 [8] f2 [8]
(0xF2) |11110011 [8] f3 [8]
(0xF3) |11110100 [8] f4 [8]
(0xF4) |11110101 [8] f5 [8]
0x80 "date" = NIL
0x81 ":scheme" = "https"
0x82 ":scheme" = "http"
0x83 ":scheme" = "ftp"
0x84 ":method" = "get"
0x85 ":method" = "post"
0x86 ":method" = "put"
0x87 ":method" = "delete"
0x88 ":method" = "options"
0x89 ":method" = "patch"
0x8A ":method" = "connect"
0x8B ":path" = "/"
0x8C ":host" = NIL
0x8D "cookie" = NIL
0x8E ":status" = 100
0x8F ":status" = 101
0x90 ":status" = 102
0x91 ":status" = 200
0x92 ":status" = 201
0x93 ":status" = 202
0x94 ":status" = 203
0x95 ":status" = 204
0x96 ":status" = 205
0x97 ":status" = 206
0x98 ":status" = 207
0x99 ":status" = 208
0x9A ":status" = 300
0x9B ":status" = 301
0x9C ":status" = 302
0x9D ":status" = 303
0x9E ":status" = 304
0x9F ":status" = 305
0xA0 ":status" = 307
0xA1 ":status" = 308
0xA2 ":status" = 400
0xA3 ":status" = 401
0xA4 ":status" = 402
0xA5 ":status" = 403
0xA6 ":status" = 404
0xA7 ":status" = 405
0xA8 ":status" = 406
0xA9 ":status" = 407
0xAA ":status" = 408
0xAB ":status" = 409
0xAC ":status" = 410
0xAD ":status" = 411
0xAE ":status" = 412
0xAF ":status" = 413
0xB0 ":status" = 414
0xB1 ":status" = 415
0xB2 ":status" = 416
0xB3 ":status" = 417
0xB4 ":status" = 500
0xB5 ":status" = 501
0xB6 ":status" = 502
0xB7 ":status" = 503
0xB8 ":status" = 504
0xB9 ":status" = 505
0xBA ":status-text" = "OK"
0xBB ":version" = "1.1"
0xBC "accept" = NIL
0xBD "accept-charset" = NIL
0xBE "accept-encoding" = NIL
0xBF "accept-language" = NIL
0xC0 "accept-ranges" = NIL
0xC1 "allow" = NIL
0xC2 "authorization" = NIL
0xC3 "cache-control" = NIL
0xC4 "content-base" = NIL
0xC5 "content-encoding" = NIL
0xC6 "content-length" = NIL
0xC7 "content-location" = NIL
0xC8 "content-md5" = NIL
0xC9 "content-range" = NIL
0xCA "content-type" = NIL
0xCB "content-disposition" = NIL
0xCC "content-language" = NIL
0xCD "etag" = NIL
0xCE "expect" = NIL
0xCF "expires" = NIL
0xD0 "from" = NIL
0xD1 "if-match" = NIL
0xD2 "if-modified-since" = NIL
0xD3 "if-none-match" = NIL
0xD4 "if-range" = NIL
0xD5 "if-unmodified-since" = NIL
0xD6 "last-modified" = NIL
0xD7 "location" = NIL
0xD8 "max-forwards" = NIL
0xD9 "origin" = NIL
0xDA "pragma" = NIL
0xDB "proxy-authenticate" = NIL
0xDC "proxy-authorization" = NIL
0xDD "range" = NIL
0xDE "referer" = NIL
0xDF "retry-after" = NIL
0xE0 "server" = NIL
0xE1 "set-cookie" = NIL
0xE2 "status" = NIL
0xE3 "te" = NIL
0xE4 "trailer" = NIL
0xE5 "transfer-encoding" = NIL
0xE6 "upgrade" = NIL
0xE7 "user-agent" = NIL
0xE8 "vary" = NIL
0xE9 "via" = NIL
0xEA "warning" = NIL
0xEB "www-authenticate" = NIL
0xEC "access-control-allow-origin" = NIL
0xED "get-dictionary" = NIL
0xEE "p3p" = NIL
0xEF "link" = NIL
0xF0 "prefer" = NIL
0xF1 "preference-applied" = NIL
0xF2 "accept-patch" = NIL
0xF3 NIL
0xF4 NIL
0xF5 NIL
0xF6 NIL
0xF7 NIL
0xF8 NIL
0xF9 NIL
0xFA NIL
0xFB NIL
0xFC NIL
0xFD NIL
0xFE NIL
0xFF NIL
In order to properly deal with the backwards compatibility concerns for HTTP/1, there are several important rules for use of Typed Codecs in HTTP headers:
A Note of warning: Individual header fields MAY be defined such that they can be represented using multiple types. Numeric fields, for instance, can be represented using either the uvarint encoding or using the equivalent sequence of ASCII numbers. Implementers will need to be capable of supporting each of the possible variations. Designers of header field definitions need to be aware of the additional complexity and possible issues that allowing for such alternatives can introduce for implementers.
Based on an initial survey of header fields currently defined by the HTTPbis specification documents, the following header field definitions can be updated to make use of the new types
| Field | Type | Description |
|---|---|---|
| content-length | Numeric or Text | Can be represented as either an unsigned, variable-length integer or a sequence of ASCII numbers. |
| date | Timestamp or Text | Can be represented as either a uvarint encoded timestamp or as text (HTTP-date). |
| max-forwards | Numeric or Text | Can be represented as either an unsigned, variable-length integer or a sequence of ASCII numbers. |
| retry-after | Timestamp, Numeric or Text | Can be represented as either a uvarint encoded timestamp, an unsigned, variable-length integer, or the text equivalents of either (HTTP-date or sequence of ASCII numbers) |
| if-modified-since | Timestamp or Text | Can be represented as either a uvarint encoded timestamp or as text (HTTP-date). |
| if-unmodified-since | Timestamp or Text | Can be represented as either a uvarint encoded timestamp or as text (HTTP-date). |
| last-modified | Timestamp or Text | Can be represented as either a uvarint encoded timestamp or as text (HTTP-date). |
| age | Numeric or Text | Can be represented as either an unsigned, variable-length integer or a sequence of ASCII numbers. |
| expires | Timestamp or Text | Can be represented as either a uvarint encoded timestamp or as text (HTTP-date). |
In the current design, dynamic cache storage slots are assigned by the decompressor in "encounter order". While this is effective, it requires that all compressor and decompressor implementations utilize the exact same algorithm for assigning storage slots, precluding any implementation from experimenting with more efficient algorithms. An alternative approach is to allow the compressor to assign the slots and communicate the assigned slots with each header. This would require a single additional byte per header instance in the Literal and Cloned Header Groups.
The serialization syntax for Cloned and Literal Header Groups would change to:
; Cloned-Index Header Group
cloned-index-header-group = cloned-index-header-prefix
1*32(cache-index cache-index value)
; Literal Header Group
literal-header-group = literal-header-prefix
1*32(cache-index name value)
The decompressor would used the specified storage index locations to store header values rather than relying on the encounter order, increasing the general efficiency of the algorithm with a minimal impact on transmission size.
On the downside, this puts the compressor in greater control over the decompressor-maintained state because the decompressor would not know how long it needs to hold on to stored items. This opens the decompressor up to possible abuse from malicious compressors. The current least-recently-written cache approach allows the decompressor to reliably drop items as space fills without fear of falling out of sync with the compressor.
This specification currently uses the number of milliseconds from the UNIX Epoch to represent timestamps. This 64-bit number is encoded using the same uvarint encoding as Numeric fields. This means that the timestamp is encoded using a variable width that, right now (for about the next 100 years or so), will encode in six bytes, then seven bytes for the reasonable future.
One possible alternative approach we can take is similar to NTP's handling of Era's. We can take the current timestamp and generate an Era value, with a maximum of 255 (0xFF). This is used as a multiplier for the timestamp value. The two values are calculated using the following formula:
m = 4294967296000
now = milliseconds since UNIX Epoch
era = now / m
ts = now % m
The allowable values for "era" would be capped at 255. This value is encoded as a single byte, followed by a uvarint encoding of ts. This ensures that the timestamp will never be encoded using more than 7-bytes total, though it may be encoded in as few as two bytes on extremely rare occasions (specifically, immediately following each era rollover).
Wire Syntax:
era = %x00-FF
ts = uvarint
date-time = era ts
To convert back to the Epoch, the formula is equally simple:
now = era * m + ts
The largest date we can encode using this format is "36812-02-20T00:36:15.999Z". Dates prior to the epoch cannot be represented.
The significant drawback with this approach is that current dates would encode in 7-bytes until the next era rollover, which will occur at approximately 2106-02-07T06:28:15.999Z (give or take a few leap seconds thrown in here and there).
Note: The byte lengths assume we want millisecond precision. If we opted to keep the second precision currently in HTTP/1, then this alternative encoding ensures that our timestamps never exceed six-bytes in length.
The uvarint encoding currently specified by this specification is certainly not the only possible option we can use. I chose it simply because it is drop dead simple to implement. There are quite a few other approaches we can take, each of which can be used as drop-in replacements for the current approach. Below are just a couple alternatives. There are plenty others. We just need to pick the one that we feel will work the best.
It ought to be noted that while each of these schemes vary in details such as endianess, specific wire-format, etc, each will typically encode the same numbers using the same number of bytes with variations of only a single byte only in the edge cases. Processing time for each is also equivalent when dealing with any number less or equal to 64-bits in length. This means that the choice is largely a matter of style than substance.
With this option, leading bits are used to indicate the total number of bytes used to encode the number value, with no fixed upper limit. Values strictly less than 128 are encoded using a single byte.
0 xxxxxxx
10 xxxxxx OCTET
110 xxxxx 2OCTET
1110 xxxx 3OCTET
11110 xxx 4OCTET
111110 xx 5OCTET
1111110 x 6OCTET
11111110 7OCTET
11111111 0xxxxxxx 7OCTET
11111111 10xxxxxx 8OCTET
...
The number of leading 1 bits specify the number of additional bytes used to serialize the value. A single 0 bit is used to mark the end of this prefix, the remaining bits are used to encode the minimum bits necessary to encode the value (with appropriate leading 0 bits to ensure proper byte-alignment).
For instance, the integer value 500, which is represented in binary as 00000001 11110100, can be encoded using two bytes, 10000001 11110100
The integer value 9770098, which is represented in binary as 10010101 00010100 01110010, can be encoded using four bytes: 11100000 10010101 00010100 01110010
This option is generally identical to the previous with the exception of being capped at a maximum of nine encoded octets total. Rather than growing indefinitely, the encoded value must never require more then eight continuation bytes to encode. Because of this restriction, there is no need for a trailing 0-bit spilling over past the first leading byte.
0 xxxxxxx
10 xxxxxx OCTET
110 xxxxx 2OCTET
1110 xxxx 3OCTET
11110 xxx 4OCTET
111110 xx 5OCTET
1111110 x 6OCTET
11111110 7OCTET
11111111 8OCTET
...
The number of leading 1 bits specify the number of additional bytes used to serialize the value. If the number of bytes required is less than 8, a single 0 bit is used to mark the end of this prefix, the remaining bits are used to encode the minimum bits necessary to encode the value (with appropriate leading 0 bits to ensure proper byte-alignment).
For instance, the integer value 500, which is represented in binary as 00000001 11110100, can be encoded using two bytes, 10000001 11110100
The integer value 9770098, which is represented in binary as 10010101 00010100 01110010, can be encoded using four bytes: 11100000 10010101 00010100 01110010
This format is not capable of encoding any number requiring more than 64-bits.
The Cookie and Set-Cookie header fields are interesting in that they establish a subordinate layer of key+value pairs exchanged within an HTTP dialog. Up to this point, Cookies have been handled largely as blobs of text, which is generally inefficient and inflexible. Even though it works, the design is not optimum.
The introduction of a new header serialization mechanism and HTTP/2's new framing provides us with an interesting opportunity to make significant improvements to the way cookies work.
Consider, for example, rather than defining Set-Cookie as a header field, we could create a modified Headers Frame that contains header fields that ought to be persisted and returned by the client.
The new Headers Frame flags would be:
These flags would pertain only to the header fields contained by the one Headers frame.
Within the header block, similar to how special ':' prefixed HTTP header fields are handled, there would be a handful of special ':' fields:
When the Server wishes to send Cookies to the client, rather than sending a separate Set-Cookie header for every individual name+value pair it wishes the client to persist, it would send a Stateful Header Frame. The client would receive this, persist the contained set of name+value pairs just as it would persist cookies today, then return those to the server in a separate Headers Frame in all appropriate subsequent streams.
This approach has a number of distinct advantages of the current Set-Cookies mechanism:
For example, suppose the server wishes the client to store two cookies for 10 seconds with the HttpOnly and Secure flags set. The server wishes the client to also clear any previous cookies it had stored. To accomplish this, the server would send a Stateful Headers Frame to the client with bit flags P, C, H and S set to 1 and the following key-value pairs encoded in the serialized header block:
:host = "example.com"
:path = "/"
:max-age = 10
foo = "bar"
baz = 123
Upon creating a new streams for the domain "example.com" and path "/", the client would include the "foo" and "baz" header fields in the set of headers encoded in the request:
:method = GET
:path = "/foo"
:host = "example.com"
foo = "bar"
baz = 123
Note that this approach is backwards compatible with Set-Cookie and Cookie in that those header fields could continue to be used by existing implementations.