CHANGELOG revision 7fbe101b
12018-02-28 2 - Fix unit test regression: enable them correctly in cmake. 3 - Simplify connection has interface 4 52018-02-26 6 - [API Change] lsquic_engine_connect() returns pointer to the connection 7 object. 8 - [API Change] Add lsquic_conn_get_engine() to get engine object from 9 connection object. 10 - [API Change] Add lsquic_conn_status() to query connection status. 11 - [API Change] Add add lsquic_conn_set_ctx(). 12 - [API Change] Add new timestamp format, e.g. 2017-03-21 13:43:46.671345 13 - [OPTIMIZATION] Process handshake STREAM frames as soon as packet 14 arrives. 15 - [OPTIMIZATION] Do not compile expensive send controller sanity check 16 by default. 17 - [OPTIMIZATION] Add fast path to gquic_be_gen_reg_pkt_header. 18 - [OPTIMIZATION] Only make squeeze function call if necessary. 19 - [OPTIMIZATION] Speed up Q039 ACK frame parsing. 20 - [OPTIMIZATION] Fit most used elements of packet_out into first 64 bytes. 21 - [OPTIMIZATION] Keep track of scheduled bytes instead of calculating. 22 - [OPTIMIZATION] Prefetch next unacked packet when processing ACK. 23 - [OPTIMIZATION] Leverage fact that ACK ranges and unacked list are. 24 ordered. 25 - [OPTIMIZATION] Reduce function pointer use for STREAM frame generation 26 - Fix: reset incoming streams that arrive after we send GOAWAY. 27 - Fix: delay client on_new_conn() call until connection is fully set up. 28 - Fixes to buffered packets logic: splitting, STREAM frame elision. 29 - Fix: do not dispatch on_write callback if no packets are available. 30 - Fix WINDOW_UPDATE send and resend logic. 31 - Fix STREAM frame extension code. 32 - Fix: Drop unflushed data when stream is reset. 33 - Switch to tracking CWND using bytes rather than packets. 34 - Fix TCP friendly adjustment in cubic. 35 - Fix: do not generate invalid STOP_WAITING frames during high packet 36 loss. 37 - Pacer fixes. 38 392017-12-18 40 41 - Fix: better follow cubic curve after idle period 42 - Fix: add missing parts to outgoing packet splitting code 43 - Fix: compilation using gcc 4.8.4 44 452017-10-31 46 47 - Add APIs.txt -- describes LSQUIC APIs 48 492017-10-31 50 51 - [API Change] Sendfile-like functionality is gone. The stream no 52 longer opens files and deals with file descriptors. (Among other 53 things, this makes the code more portable.) Three writing functions 54 are provided: 55 56 lsquic_stream_write 57 lsquic_stream_writev 58 lsquic_stream_writef (NEW) 59 60 lsquic_stream_writef() is given an abstract reader that has function 61 pointers for size() and read() functions which the user can implement. 62 This is the most flexible way. lsquic_stream_write() and 63 lsquic_stream_writev() are now both implemented as wrappers around 64 lsquic_stream_writef(). 65 66 - [OPTIMIZATION] When writing to stream, be it within or without the 67 on_write() callback, place data directly into packet buffer, 68 bypassing auxiliary data structures. This reduces amount of memory 69 required, for the amount of data that can be written is limited 70 by the congestion window. 71 72 To support writes outside the on_write() callback, we keep N 73 outgoing packet buffers per connection which can be written to 74 by any stream. One half of these are reserved for the highest 75 priority stream(s), the other half for all other streams. This way, 76 low-priority streams cannot write instead of high-priority streams 77 and, on the other hand, low-priority streams get a chance to send 78 their packets out. 79 80 The algorithm is as follows: 81 82 - When user writes to stream outside of the callback: 83 - If this is the highest priority stream, place it onto the 84 reserved N/2 queue or fail. 85 (The actual size of this queue is dynamic -- MAX(N/2, CWND) -- 86 rather than N/2, allowing high-priority streams to write as 87 much as can be sent.) 88 - If the stream is not the highest priority, try to place the 89 data onto the reserved N/2 queue or fail. 90 - When tick occurs *and* more packets can be scheduled: 91 - Transfer packets from the high N/2 queue to the scheduled 92 queue. 93 - If more scheduling is allowed: 94 - Call on_write callbacks for highest-priority streams, 95 placing resulting packets directly onto the scheduled queue. 96 - If more scheduling is allowed: 97 - Transfer packets from the low N/2 queue to the scheduled 98 queue. 99 - If more scheduling is allowed: 100 - Call on_write callbacks for non-highest-priority streams, 101 placing resulting packets directly onto the scheduled queue 102 103 The number N is currently 20, but it could be varied based on 104 resource usage. 105 106 - If stream is created due to incoming headers, make headers readable 107 from on_new. 108 109 - Outgoing packets are no longer marked non-writeable to prevent placing 110 more than one STREAM frame from the same stream into a single packet. 111 This property is maintained via code flow and an explicit check. 112 Packets for stream data are allocated using a special function. 113 114 - STREAM frame elision is cheaper, as we only perform it if a reset 115 stream has outgoing packets referencing it. 116 117 - lsquic_packet_out_t is smaller, as stream_rec elements are now 118 inside a union. 119 1202017-10-12 121 122 - Do not send RST_STREAM when stream is closed for reading 123 - Raise maximum header size from 4K to 64K 124 - Check header name and value lengths against maximum imposed by HPACK 125 - Fix NULL dereference in stream flow controller 126 1272017-10-09 128 129 - Hide handshake implementation behind a set of function pointers 130 - Use monotonically increasing clock 131 - Make sure that retx delay is not larger than the max of 60 seconds 132 1332017-09-29 134 135 - A few fixes to code and README 136 1372017-09-28 138 139 - Add support for Q041; drop support for Q040 140 1412017-09-27 142 143 - Fix CMakeLists.txt: BoringSSL include and lib was mixed up 144 1452017-09-26 146 147 - Add support for Mac OS 148 - Add support for Raspberry Pi 149 - Fix BoringSSL compilation: include <openssl/hmac.h> explicitly 150 1512017-09-22 152 153 - Initial release 154