CHANGELOG revision ea822d0d
12020-03-12
2    - 2.13.1
3    - [API] Use lsxpack_header structure to process incoming headers.
4    - [BUGFIX] Fix assertion when zero-padding Initial packet.
5    - [BUGFIX] Use idle timeout before we learn of peer's value.
6    - Use ls-hpack 2.0.1 -- has lsxpack_header changes.
7    - Use ls-qpack 0.14.0 -- new, common with ls-hpack, XXH seed (not
8      used yet).
9    - Code cleanup: prefix exported functions with "lsquic_".
10
112020-03-02
12    - 2.12.0
13    - [FEATURE] QUIC timestamps extension.
14    - [API] New: ea_alpn that is used when not in HTTP mode.
15    - [BUGFIX] SNI is mandatory only for HTTP/3 and gQUIC.
16    - [BUGFIX] Benign double-free -- issue #110.
17    - [BUGFIX] Printing of transport parameters.
18
192020-02-24
20    - 2.11.1
21    - [FEATURE] QUIC and HTTP/3 Internet Draft 27 support.
22    - [FEATURE] Add experimental delayed ACKs extension.
23    - Drop support for Internet Draft 24.
24    - Code cleanup.
25
262020-02-14
27    - 2.10.6
28    - [BUGFIX] HTTP/3 framing: don't misinterpret rare occurence as error.
29    - [BUGFIX] Send gap warning due to missing poisoned packet.
30    - Stream unit test for scenario in issue #106.
31
322020-02-13
33    - 2.10.5
34    - [BUGFIX] BBR: call cci_sent() with correct arguments and at correct
35      time.
36    - Refactor transport parameters module.
37    - Minor code cleanup.
38
392020-02-11
40    - 2.10.4
41    - [BUGFIX] Send HANDSHAKE_DONE only after Finished is received.
42    - [BUGFIX] Don't treat garbage UDP padding as library error; ignore
43      it instead.
44    - [BUGFIX] Fix compilation on FreeBSD (missing header).
45    - Code cleanup: remove unnecessary #includes.
46
472020-01-31
48    - 2.10.3
49    - [BUGFIX] Cancel path responses and challenges on old path when
50      switching to new path
51    - Logging network path information.
52
532020-01-30
54    - 2.10.2
55    - [BUGFIX] Do not delay ACKs for Initial and Handshake packets.
56    - [BUGFIX] Send PATH_CHALLENGE if path changed before mini conn
57      promotion.
58    - Logging improvements.
59    - http_client: discard data faster.
60
612020-01-29
62    - 2.10.1
63    - [BUGFIX] Coalesced packets could get longer than normal packet
64      size.
65    - Add spin bit configuration option es_spin (-o spin=[01]).
66    - Disable spin bit in 1/16 of connections.
67    - Improve logging a bit.
68
692020-01-28
70    - 2.10.0
71    - [FEATURE] QUIC and HTTP/3 Internet Draft 25 support.
72    - [API] Drop support for ID-23.
73    - [BUGFIX] Set key phase bit on outgoing packets correctly.
74    - Code cleanup.
75
762020-01-20
77    - 2.9.0
78    - [API] Drop support for Q039.
79    - Improve ACK-queuing logic.  Send an ACK once in a while if
80      peer keeps on sending non-ack-eliciting packets.
81    - Improve Alt-Svc string: Q050 and later are not included in
82      the old-style "quic" string.
83    - Send stateless resets if connection could not be promoted.
84    - Schedule MAX_DATA if needed when DATA_BLOCKED is received.
85    - Use ls-qpack 0.11.2 -- needed for server push optimization.
86    - Code cleanup: handle some error cases, improve logging.
87
882020-01-16
89    - 2.8.9
90    - [BUGFIX] Use ls-qpack 0.11.1
91    - [OPTIMIZATION] Generate random bytes in batches.
92    - Change loss_bits transport parameter ID to 0x1057 following
93      latest draft.
94    - Randomize period with which PINGs are sent to elicit ACKs.
95    - Some refactoring and code cleanup.
96
972020-01-14
98    - 2.8.8
99    - [BUGFIX] Invalid read when parsing IETF transport parameters
100      (this was benign).
101    - [OPTIMIZATION] Frame bundling when using buffered packets in
102      IETF QUIC: a) flush QPACK decoder stream and b) include ACKs
103      in opportunistic fashion.
104    - Fix HTTP/3 framing unit test.
105    - Code cleanup.
106
1072020-01-09
108    - 2.8.7
109    - [BUGFIX] Initial packet size check for IETF mini conn applies to
110      UDP payload, not QUIC packet.
111    - Support old and new school loss_bits transport parameter.
112    - Use Q run length of 64 as suggested in the loss bits Draft.
113    - Undo square wave count when packet is delayed.
114    - Code cleanup; minor fixes.
115
1162020-01-06
117    - 2.8.5
118    - [HTTP3] Verify number of bytes in incoming DATA frames against
119      content-length.
120    - [HTTP3] Stop issuing streams credits if peer stops opening QPACK
121      decoder window.  This addresses a potential attack whereby client
122      can cause the server to keep allocating memory.  See Security
123      Considerations in the QPACK draft.
124    - [BUGFIX] Mini conn: don't shorten max packet size for Q050 and later.
125    - [BUGFIX] Init IETF connection flow controller using correct setting.
126    - [BUGFIX] Fix unintended sign extension when removing header protection.
127    - Code cleanup and minor fixes.
128
1292019-12-30
130    - 2.8.1
131    - [FEATURE] Use occasional packet number gaps to detect optimistic
132      ACK attacks.
133    - [BUGFIX] Q050 client: all packet numbers are in the App PNS.
134    - [OPTIMIZATION] Merge multi-range ACK frames, not just single-range
135      ACK frames.
136    - IETF QUIC: use RTT estimate in ack timeout calculation.
137    - IETF handshake: abort conn when unexpected errors occur.
138    - Use PING rather than MAX_DATA frames to elicit ACKs from peer.
139    - Server: enforce 1200 byte Initial minimum packet size.
140    - [CLEANUP] Remove code to disable gQUIC crypto.
141    - [CLEANUP] Remove n_timestamps from ACK info struct.
142    - Optimize driver: reuse previous ancillary message when possible.
143
1442019-12-23
145    - 2.8.0
146    - [FEATURE] Add support for Q050.
147    - [OPTIMIZATION] Reduce mallocs in gQUIC handshake.
148    - [BUGFIX] Disable redo of failed STREAM frame insertion with debug
149      logging.
150
1512019-12-18
152    - 2.7.3
153    - [DEBUG] Further dedup next advisory tick messages when reason is
154      the same.
155    - [BUGFIX] Update size of `a` array in TP struct.  Fixes (benign)
156      GitHub bug #94.
157    - Use Cubic by default again instead of BBR, as it delivers more
158      consistent performance.
159
1602019-12-11
161    - 2.7.2
162    - [BUGFIX] Send controller: update scheduled bytes when DCID length
163      changes (IETF client).
164    - [BUGFIX] Drop alarm check from sanity test.  It no longer works now
165      that we use loss chains.
166    - [PORTABILITY] Fix build on Alpine Linux.
167    - [PORTABILITY] Fix build using XCode.
168    - Client initial DCID length: use RAND_bytes() instead of rand(3).
169    - Add unit tests for connection min heap.
170    - [DEBUG] Log CID in gQUIC handshake module
171    - [DEBUG] Turn on extra checks for IETF client send controller.
172    - [DEBUG] Dedup next advisory tick messages when reason is IDLE timer.
173    - [DEBUG] QPACK decoder handler: log header error code.
174
1752019-12-05
176    - 2.7.1
177    - [BUGFIX] client: don't call ignore_init() in middle of batch send.
178      ignore_init() makes an assumption that the send controller has access
179      to all outgoing packets.  This change wraps a few IETF full connection
180      methods to delay calling ignore_init() until the engine returns all
181      outgoing packets that were batched.
182    - [BUGFIX] set errno to EAGAIN if sendmmsg() can't send all of them.
183      This needs to be done because the value of errno may be lost on
184      some platforms.
185    - [BUGFIX] Typo that set all bits in sm_qflags lead to crashes.
186    - [BUGFIX] Do not cancel header block processing after failure, as
187      QPACK releases the reference in that case.
188    - [CLEANUP] IETF encrypt: replace assert(0) with a warning.
189    - Several small improvements to the test server.
190
1912019-11-27
192    - 2.7.0
193    - [API, FEATURE] Close connection immediately when ea_packets_out()
194      fails with errno != EAGAIN.  The API change is that errno is now
195      examined.  Make sure to set it if using something other than
196      sendmsg() to send packets.
197    - [CLEANUP] Immediate close logic in IETF full conn.
198    - [CLEANUP] Fix bogus warning about uninitialized `pair' variable.
199
2002019-11-22
201    - 2.6.7
202    - [FEATURE] Implement the QL extension (offered by default).
203    - [BUGFIX] Abort when encountering unexpected HTTP/3 frames.
204    - [BUGFIX] Acknowledge (QPACK) HTTP/3 trailers correctly.
205    - [DEBUG] Turn on debug message for next advisory tick.
206
2072019-11-20
208    - 2.6.6
209    - [BUGFIX] Using HTTP/3 to HTTP/1.x converter.
210    - [BUGFIX] Truncate log messages instead of throwing them away.
211
2122019-11-15
213    - 2.6.5
214    - [BUGFIX] High priority buffered packet queue length.
215    - [BUGFIX] Rain time calculation: max_ack_delay is in milliseconds.
216
2172019-11-12
218    - 2.6.3
219    - [BUGFIX] Close DATA frames with empty payload correctly.
220
2212019-11-11
222    - 2.6.2
223    - [BUGFIX] SCID!=ODCID rule applies to Retry packets, not regular
224      packets.
225    - [BUGFIX] Zero-RTT: BoringSSL no longer flips read/write secrets.
226    - [BUGFIX] Truncate ACK frame rather instead of aborting IETF
227      connection.
228    - [BUGFIX] Client: don't send duplicate reset tokens.
229    - [BUGFIX] Remove invalid assertion in H3 framing code.
230    - Silence a warning in send ctl by restructuring switch() statement.
231
2322019-11-08
233    - 2.6.1
234    - [BUGFIX] set retry token on all resubmitted packets.
235    - Event log: log sent packet flags.  In particular, this allows one
236      to see whether token was sent.
237    - Don't migrate client if es_allow_migration is false.
238
2392019-11-07
240    - 2.6.0
241    - [FEATURE] QUIC and HTTP/3 Internet Draft 24 support
242
2432019-11-07
244    - 2.5.2
245    - [BUGFIX] argument order to gQUIC client constructor.  Regression
246      introduced in 2.5.0.
247    - [BUGFIX] split buffered packet error recovery: destroy the correct
248      packet.
249
2502019-11-04
251    - 2.5.1
252    - [BUGFIX] Fix double-free when emptying a packet number space.
253    - [BUGFIX] http_server: fix md5sum handler: handle EOF correctly.
254    - [BUGFIX] Use random values in bits 4 and 5 of the first byte of
255      verneg packets (regression introduced in 2.5.0).
256    - [OPTIMIZATION] Don't compile in expensive attq checks by default.
257    - [OPTIMIZATION] http_server: compile regexes only once.
258
2592019-10-31
260    - 2.5.0
261    - [API] lsquic_engine_connect() can now be passed QUIC version to use.
262    - [OPTIMIZATION] Queue opportunistic ACKs if there is data to be sent.
263    - [BUGFIX] Don't evict streams from priority iterator if there is
264      only one queue.
265    - [OPTIMIZATION, BUGFIX] Several other optimizations and bug fixes.
266    - Use ls-qpack v0.10.7.
267
2682019-10-24
269    - 2.4.10
270    - [BUGFIX] IETF QUIC server: fix uninitialized variable use.
271    - [BUGFIX] make sure TLSv1.3 is not disabled in SSL object.
272    - [BUGFIX] Use issuer name and serial number to cache certs (SKID
273      values are not unique).
274    - [BUGFIX] Always set the idle alarm in IETF connection so that it
275      can time out.
276    - Use ls-qpack v0.10.6, as it silences some warnings.
277
2782019-10-21
279    - 2.4.8
280    - [OPTIMIZATION, BUGFIX] Use ls-qpack v0.10.5.
281
2822019-10-15
283    - 2.4.7
284    - Add echo client and server to the distibution.
285    - Add MD5 client and server to the distibution.
286    - Fix http_client: check command-line arguments better, prevent crash.
287    - Fix IETF conn: can_write_ack() should only care about APP PNS.
288    - Client: delay stream creation until handshake succeds.
289    - Reset HTTP stream whose write end is closed prematurely.
290    - Fix tickable(): mirror behavior of tick() wrt buffered packets.
291    - Log reason why engine is tickable.
292
2932019-10-11
294    - 2.4.6
295    - Minor code cleanup and logging improvements.
296    - Server and client programs: include library version (e.g. 2.4.6)
297      into `server' and `user-agent' headers.
298
2992019-10-08
300    - 2.4.5
301    - [OPTIMIZATION]: flush encoder stream only when necessary.
302    - [BUGFIX] Use ls-qpack v0.10.2 for new API -- and for a bug fix.
303    - [BUGFIX] Typo in IETF conn SETTINGS writer.
304    - Use latest BoringSSL.
305
3062019-10-08
307    - 2.4.4
308    - [API] Add lsquic_alpn2ver() to aid parsing Alt-Svc header.
309    - [BUGFIX] NULL dereference when H3 frame header would be split.
310    - [BUGFIX] Do not close fixed-size H3 frame prematurely.
311    - [BUGFIX] Allow PING frames in IETF mini conn.
312    - [BUGFIX] Mini conns: don't send any packets after receiving
313      CONNECTION_CLOSE.
314    - [BUGFIX] Client migration: reserve slot for DCID from transport params.
315    - [BUGFIX] Allow max_early_data_size=0 -- early_data might not be there.
316    - [BUGFIX] Use an invalid stream number to reset BPT cache (zero is now a
317      valid stream number).
318    - [SPEC] Use FINAL_SIZE_ERROR when FIN mismatch is detected.
319    - [OPTIMIZATION] Closed connection only gets one chance to send packets.
320    - [OPTIMIZATION] Flush headers stream before packetizing stream data.
321    - [OPTIMIZATION] process QPACK encoder STREAM frames immediately.
322    - Update ls-qpack to v0.10.1.
323
3242019-09-30
325    - 2.4.3
326    - Add GQUIC versions to the list of h3 ALPNs for Alt-Svc header.
327      For example, h3-Q043.  Chrome will switch to using this format at
328      some point in the future.
329    - [BUGFIX] Send correct value in max_streams_uni transport param
330    - [SPEC] Abort IETF QUIC connection if max_early_data_size is not set
331      to 0xFFFFFFFF
332
3332019-09-23
334    - 2.4.2
335    - [BUGFIX] H3 framing: fix zero-byte write when space is available
336    - [BUGFIX] Don't send STREAM frame when incoming unidirectgional stream
337      is closed
338    - [BUGFIX] Cancel all pending writes by stream reset by a GOAWAY
339    - [BUGFIX] Fix use-after-free in IETF full conn
340    - [OPTIMIZATION] Wait for session tickets for two seconds and then drop
341      SSL object and crypto streams.
342
3432019-09-18
344    - 2.4.0
345    - [FEATURE] QUIC and HTTP/3 Internet Draft 23 support
346
3472019-09-13
348    - 2.3.1
349    - [BUGFIX] Fix memory leaks
350    - [BUGFIX] Fix unit tests
351
3522019-09-12
353    - 2.3.0
354    - [FEATURE] BBR congestion control is on by default
355    - [BUGFIX] BBR app-limited logic
356    - [BUGFIX] Fix uninitialized warnings in IETF
357    - [BUGFIX] Update ls-qpack to v0.9.14
358    - [CLEANUP] Code cleanup
359
3602019-09-11
361    - 2.2.0
362    - [FEATURE] Server code is included in the library
363    - [FEATURE] IETF QUIC and HTTP/3 Support (ID-22)
364
3652019-05-13
366    - 1.21.2
367    - [OPTIMIZATION] HPACK: use history to improve compression performance
368
3692019-05-06
370    - 1.21.1
371    - [BUGFIX] If FIN or RST not received, don't delay stream destruction.
372    - [OPTIMIZATION] Speed up HPACK encoder by using same hash value to
373      search static and dynamic tables.
374
3752019-04-12
376    - 1.21.0
377    - [FEATURE] Add qlog log module.
378
3792019-04-01
380    - 1.20.0
381    - [FEATURE] Add support for Q046.
382
3832019-03-19
384    - 1.19.6
385    - [BUGFIX] Ensure that Largest Observed does not decrease in ACKs we
386      send.  ACK frames placed in packets in buffered queues (optimization
387      introduced in 1.17.15) can be preceded by an ACK frame generated later.
388      In this case, the older ACK frame should not be sent out, as Chromium-
389      based servers flags decrease in the ACK frame's Largest Observed value
390      as an error.
391
3922019-03-05
393    - 1.19.5
394    - [BUGFIX] Use correct public key from PUBS based on KEXS index.
395    - [BUGFIX] Check flags before dispatching writes, avoiding assert.
396    - [BUGFIX] Set :scheme to "https" (instead of "HTTP").
397
3982019-02-25
399    - 1.19.4
400    - [BUGFIX] Check buffer bounds when looking up version in 0-RTT blob.
401    - [BUGFIX] http_client: don't fetch 0-rtt info if handshake failed.
402    - Log number of pacer calls at DEBUG, rather than NOTICE, level.
403
4042019-02-18
405    - 1.19.3
406    - [BUGFIX] Q044: don't encode packet number in 6 bytes.  Six-byte
407      packet number encoding does not exist in Q044.  This fixes a
408      regression introduced in '[BUGFIX] Buffered packets can contain
409      ACK frames' -- we need to keep QUIC version in mind when selecting
410      the longest possible packet number encoding used for the buffered
411      packet that carries the ACK.
412    - [BUGFIX] Do not increase CWND when timeout occurs.
413    - http_client: support setting handshake timeout on command line.
414      Use -o handshake_to=timeout.
415    - http_client: use -k to connect UDP socket to pick up ICMP errors.
416    - http_client: allow pathless mode, when only handshake is performed
417      without issuing any requests.  This can be done by simply not
418      specifying a -p flag on the command line.
419
4202019-02-11
421    - 1.19.2
422    - [BUGFIX] Begin negotiation with version provided in 0-RTT info.
423    - [BUGFIX] Version checking in zero_rtt deserialize function.
424
4252019-02-04
426    - 1.19.1
427    - [BUGFIX] Fix Windows build.
428
4292019-02-04
430    - 1.19.0
431    - [FEATURE, API Change] 0-RTT support.  Add function to export 0-RTT
432      information; it can be supplied to a subsequent connect() call.
433    - [FEATURE] Add -0 flag to http_client to exercise 0-RTT support.
434    - [BUGFIX] Resuscitate the Windows build.
435    - [BUGFIX] Send HTTP settings (max header list size) if necessary.
436    - [BUGFIX] Buffered packets can contain ACK frames.
437    - [BUGFIX] Make packet writeable once all STREAM frames are elided.
438    - [BUGFIX] Fix potential null dereference when realloc fails.
439    - cmake: simplify build configuration.
440
4412019-01-28
442    - 1.18.0
443    - [API Change] Can specify clock granularity in engine settings.
444    - [BUGFIX] Pacer uses fixed clock granularity.  Since the change on
445      2018-04-09, it is not necessary to try to approximate the next tick
446      time in the pacer: it can use fix clock granularity specified by
447      the user.
448    - [BUGFIX] Do not tick constantly before handshake is done.
449    - [BUGFIX] Do not exit tick prematurely: reuse packet with ACK.  Even
450      if we cannot allocate *more* packets, we could still be able to write
451      to one already allocated.
452    - [BUGFIX] Do not schedule pacer if there are no lost packets.
453
4542019-01-17
455    - 1.17.15
456    - [BUGFIX] http_client: make sure only one read per on_read() callback
457      is performed in the header conversion bypass (-B) mode.
458    - http_client: with -E, assign random priority when stream is created.
459    - [OPTIMIZATION] On immediate write, place an ACK frame into the first
460      buffered packet if an ACK is queued.  This reduces the number of
461      standalone ACK packets.
462    - [OPTIMIZATION] Allow placing more than one STREAM frame from the same
463      stream into an outgoing packet.  This change minimizes the number of
464      buffered packets required to store several small HTTP messages by
465      virtue of allowing more than one STREAM frame from HEADERS stream in
466      the same packet.
467    - [OPTIMIZATION] Flush headers when writing to buffered packets.  This
468      causes the headers to be written to the same buffered packet queue,
469      thereby improving packet utilization, especially for small HTTP
470      messages.
471
4722019-01-16
473    - 1.17.14
474    - [FEATURE] http_client can now collect stats: time to connect, TTFB,
475      time to download, requests per seconds, and bandwidth.  See -t flag.
476    - [BUGFIX] http_client: -B, -K, and -r can be used at the same time.
477    - [BUGFIX] http_client: do not display cert chain before each request.
478      Now this is only done once per connection, if the handshake is
479      successful and -a option is given.
480    - [BUGFIX] Do not wait to dispatch on_close() after reading.  If a
481      stream is scheduled to be closed after on_read() callback returns,
482      close it immediately instead of waiting until the end of the tick.
483      If client creates new request from on_close() event, they will be
484      processed in the same tick instead of waiting for the next one.
485    - [BUGFIX] HEADERS stream is critical: always treat it with highest
486      priority.
487
4882019-01-10
489    - 1.17.12
490    - [FEATURE] http_client can now issue parallel requests in the context
491      of a single connection.  See -w option.
492
4932019-01-03
494    - 1.17.11
495    - Fix strict aliasing warning in optimized compilation.
496
4972018-12-27
498    - 1.17.10
499    - Fix the example program to be able to use parallel connections
500      again.  (See the -n argument.)
501
5022018-12-18
503    - 1.17.9
504    - [BUGFIX] Engine: reduce minimum batch size from 256 to 4
505
5062018-12-10
507    - 1.17.8
508    - [BUGFIX] Fix compilation on FreeBSD and 32-bit Linux
509
5102018-12-03
511    - 1.17.7
512    - [BUGFIX] Do not unset PING alarm before ringing expired alarms.
513      This prevented PING from ever being sent.
514
5152018-11-29
516    - 1.17.6
517    - Add failsafe: resume sending packets after some time
518
519      The change puts a 1-second limit on the amount of time the engine
520      will not send packets after some packets are delayed.  This makes
521      the library robust in case the user does not unblock the engine
522      explicitly using lsquic_engine_send_unsent_packets() call.
523
524    - [BUGFIX] Handle corner cases in send controller when packets are
525      a) delayed or b) dropped during repackaging.
526    - [BUGFIX] Memory leak: destroy buffered packets during controller
527      cleanup.
528
5292018-11-16
530    - 1.17.3
531    - [BUGFIX] Do not send STOP_WAITING frames when using Q044
532
5332018-10-19
534    - 1.17.2
535    - [BUGFIX] Memory leak in test_frame_rw unit test.
536    - [BUGFIX] Parsing packets with short IETF header: do not overwrite
537      flags.  (Only a problem in unit tests -- benign otherwise.)
538
5392018-10-16
540    - 1.17.0
541    - [API Change] Packet out Memory Interface (PMI) update:
542      - Split PMI pool return into pmi_release and pmi_return
543      - PMI callbacks take peer_ctx and is_ipv6 arguments
544    - [BUGFIX] Fix use-after-free when certificate is updated
545    - Silence gcc warning in optimized mode by performing useless
546      initialization
547    - cmake: use the standard variable CMAKE_BUILD_TYPE instead of
548      DEVEL_MODE
549
5502018-10-03
551    - 1.16.0
552    - [API Change] Add lsquic_conn_n_avail_streams()
553    - [BUGFIX] only dispatch crypto stream read events if WANT_READ is on
554
5552018-09-27
556    - 1.15.0
557    - [API Change] Add LSCONN_ST_PEER_GOING_AWAY to the list of conn statuses
558    - [BUGFIX] free uncompressed headers correctly when error occurs
559
5602018-09-12
561    - 1.14.3
562    - [BUGFIX] Do not abort conn on STREAM frame for a reset stream
563    - [BUGFIX] Drop packets that would become empty due to repackaging.
564      Packets on the scheduled queue may be marked for repackaging.
565      Frames such as ACK frame that are never resent are removed from
566      repackaged packets.  We must check that the newly repackaged packet
567      would not be empty.  If it would be, it is destroyed instead and
568      the next packet on the scheduled queue is used.  Note that this
569      change only affects the logic to return the next packet to be sent.
570      Lost packets that are being rescheduled are already processed in
571      this fashion.
572    - Byteswap CID before logging it - this makes it match Chrome CIDs.
573      (Except Q035, which is the last little-endian GQUIC version we
574      support.)
575
5762018-09-06
577    - 1.14.0
578    - [API Change] Disable packet sending if full batch cannot be sent
579      If lsquic_packets_out_f() cannot send the whole batch, disable
580      packet sending until lsquic_engine_send_unsent_packets() is called.
581    - [BUGFIX] Handle case when STREAM frame does not fit.
582    - [BUGFIX] Always allow incoming STREAM frames to overlap.  Peers
583      may send overlapping STREAM frames even if using versions older
584      than Q043.
585    - Custom header set fixes:
586      - set "FIN reached" flag when custom header with FIN flag is
587        claimed;
588      - do not return custom header set for a reset stream.
589
5902018-08-27
591
592    - 1.13.0
593    - [FEATURE, API Change] Add ability to create custom header set
594      objects via callbacks.  This avoids reading and re-parsing
595      headers from the stream.
596
5972018-08-27
598
599    - 1.12.4
600    - Fix memory leak when engine is destroyed
601    - Fix memory leak in http_client
602    - Fix gcc warning in unit tests
603
6042018-08-22
605
606    - 1.12.3
607    - [BUGFIX] Fix duplicate STREAM frame detection
608
6092018-08-20
610
611    - 1.12.2
612    - [BUGFIX] Update count of scheduled bytes when adjusting size of
613      an already-scheduled packet.
614    - Emit info instead of warning messages when stream is used in
615      unexpected ways.
616
6172018-08-17
618
619    - 1.12.0
620    - [FEATURE, API Change] Add support for certificate verification
621
6222018-08-16
623
624    - 1.11.1
625    - [BUGFIX] Fix STOP_WAITING frame processing in the new Q044 code
626
6272018-08-15
628
629    - 1.11.0
630    - [FEATURE] Add support for Q044.
631
6322018-08-09
633
634    - 1.10.2
635    - [BUGFIX] Don't go over limit when creating delayed streams
636
6372018-07-10
638
639    - 1.10.1
640    - [BUGFIX]  process connections after each batch of packets is read
641      This avoids a problem of accumulating a very large list of packets
642      (possible when speeds are high and socket's receive buffer is large)
643      and processing it all at once.
644    - If glibc is older than 2.17, link with rt.  This is necessary for
645      clock_getres(2).
646    - Add version macros to lsquic.h; remove unnecessary includes.
647
6482018-06-13
649
650    - [BUGFIX] allow multiple parallel connections by default
651
652      Use the original method of tracking connections by CIDs by default.
653      If zero-sized CID support is turned on, connections are tracked by
654      the address.  A new connection is not created if another connection
655      is using the same network address
656
6572018-05-30
658
659    - [FEATURE, API CHANGE] Support zero-sized CIDs in received packets
660
6612018-05-24
662
663    - Close connection properly when packet encryption fails
664
6652018-05-23
666
667    - [BUGFIX] Do not produce packet sequence gaps due to delayed packets
668
6692018-05-21
670
671    - [API Change] Add optional callback to call when handshake is done
672    - [API Change, BUGFIX] After send failure, wait until transport available
673
6742018-05-18
675
676    - [API] Expose useful lsquic_ver2str[] in lsquic.h
677    - [BUGFIX] Do not produce packet sequence gaps due to STREAM frame elision
678    - Improve checks of number of incoming streams limit and associated
679      error reporting.
680    - [BUGFIX] Command-line option `-6` now works correctly.
681
6822018-05-16
683
684    - [FEATURE] DNS resolution
685    - [BUGFIX] Frame insertion mis-ID as overlap instead of dup
686    - http_client: fix priority range generated by -E flag
687
6882018-05-09
689
690    - [FEATURE] Add support for Q043.
691    - Support for versions Q037, Q038, Q041, and Q042 has been removed.
692    - Fix typo in debug message.
693    - Fix code indentation.
694    - Add /* fallthru */ comment to stop newer gcc from complaining.
695    - Logger: fix compilation of optimized Windows build.
696
6972018-05-04
698
699    - [FEATURE] Add support for Q042.
700    - Remove comment: MSPC is obsolete (no code changes)
701    - Prog: use lsquic_str2ver() when processing -o version flag
702    - Remove unused CTIM and SRBF transport parameters
703    - Disable QUIC versions Q037 and Q038 by default
704    - Fix Windows compilation by including compat header file in
705      lshpack.c
706    - Address warnings produced by newer versions of gcc
707    - Future-proof: turn off -Werror
708
7092018-05-02
710
711    - [BUGFIX] Make lsquic_conn_quic_version() available
712    - Switch to using ls-hpack 1.1
713    - [BUGFIX] Do not ignore stream resets after receiving FIN
714
7152018-04-27
716
717    - HPACK: do not allow header block to end with table size update.
718
7192018-04-25
720
721    - [BUGFIX] Do not create gap in sent packnos when squeezing delayed
722      packets.
723    - [BUGFIX] sendctl checks for all unacked bytes, not just retx bytes.
724    - [BUGFIX] connections with blocked scheduled packets are not tickable
725      for sending.
726    - [BUGFIX] Conn is tickable if it wants to send a connection-level
727      frame.
728
7292018-04-23
730
731    - Fix busy loop: tickable must make progress.  When connection is
732      self-reporting as tickable, it must make progress when ticked.  There
733      are two issues:
734        1. If there are buffered packets, the connection is only tickable if
735           they can be sent out.
736        2. A connection is tickable if there are streams on the servicing
737           queue.  When the tick occurs, we must service the stream
738           independent of whether any packets are sent.
739    - Fix assertion in pacer which can be incorrect under some
740      conditions.
741    - cmake: do not turn on address sanitizer if in Travis.
742
7432018-04-20
744
745    - [BUGFIX] Fix bug in lsquic_engine_connect() exposed by yesterday's
746      changes.
747
7482018-04-19
749
750    - [BUGFIX] Add connection to Tickable Queue on stream write
751    - cmake: use MSVC variable instead of trying to detect
752    - engine: improve connection incref/decref logging
753    - stream: don't ignore errors that may occur on triggered flush
754    - connection: remove obsolete method
755    - engine: indicate connection as tickable if previous call went
756      over threshold
757
7582018-04-09
759
760    [API Change, OPTIMIZATION] Only process conns that need to be processed
761
762    The API is simplified: do not expose the user code to several
763    queues.  A "connection queue" is now an internal concept.
764    The user processes connections using the single function
765    lsquic_engine_process_conns().  When this function is called,
766    only those connections are processed that need to be processed.
767    A connection needs to be processed when:
768
769        1. New incoming packets have been fed to the connection.
770        2. User wants to read from a stream that is readable.
771        3. User wants to write to a stream that is writeable.
772        4. There are buffered packets that can be sent out.  (This
773           means that the user wrote to a stream outside of the
774           lsquic library callback.)
775        5. A control frame (such as BLOCKED) needs to be sent out.
776        6. A stream needs to be serviced or delayed stream needs to
777           be created.
778        7. An alarm rings.
779        8. Pacer timer expires.
780
781    To achieve this, the library places the connections into two
782    priority queues (min heaps):
783
784        1. Tickable Queue; and
785        2. Advisory Tick Time queue (ATTQ).
786
787    Each time lsquic_engine_process_conns() is called, the Tickable
788    Queue is emptied.  After the connections have been ticked, they are
789    queried again: if a connection is not being closed, it is placed
790    either in the Tickable Queue if it is ready to be ticked again or
791    it is placed in the Advisory Tick Time Queue.  It is assumed that
792    a connection always has at least one timer set (the idle alarm).
793
794    The connections in the Tickable Queue are arranged in the least
795    recently ticked order.  This lets connections that have been quiet
796    longer to get their packets scheduled first.
797
798    This change means that the library no longer needs to be ticked
799    periodically.  The user code can query the library when is the
800    next tick event and schedule it exactly.  When connections are
801    processed, only the tickable connections are processed, not *all*
802    the connections.  When there are no tick events, it means that no
803    timer event is necessary -- only the file descriptor READ event
804    is active.
805
806    The following are improvements and simplifications that have
807    been triggered:
808
809        - Queue of connections with incoming packets is gone.
810        - "Pending Read/Write Events" Queue is gone (along with its
811          history and progress checks).  This queue has become the
812          Tickable Queue.
813        - The connection hash no longer needs to track the connection
814          insertion order.
815
8162018-04-02
817
818    - [FEATURE] Windows support
819
820    - Reduce stack use -- outgoing packet batch is now allocated on the heap.
821
8222018-03-09
823
824    - [OPTIMIZATION] Merge series of ACKs if possible
825
826      Parsed single-range ACK frames (that is the majority of frames) are
827      saved in the connection and their processing is deferred until the
828      connection is ticked.  If several ACKs come in a series between
829      adjacent ticks, we check whether the latest ACK is a strict superset
830      of the saved ACK.  If it is, the older ACK is not processed.
831
832      If ACK frames can be merged, they are merged and only one of them is
833      either processed or saved.
834
835    - [OPTIMIZATION] Speed up ACK verification by simplifying send history.
836
837      Never generate a gap in the sent packet number sequence.  This reduces
838      the send history to a single number instead of potentially a series of
839      packet ranges and thereby speeds up ACK verification.
840
841      By default, detecting a gap in the send history is not fatal: only a
842      single warning is generated per connection.  The connection can continue
843      to operate even if the ACK verification code is not able to detect some
844      inconsistencies.
845
846    - [OPTIMIZATION] Rearrange the lsquic_send_ctl struct
847
848      The first part of struct lsquic_send_ctl now consists of members that
849      are used in lsquic_send_ctl_got_ack() (in the absense of packet loss,
850      which is the normal case).  To speed up reads and writes, we no longer
851      try to save space by using 8- and 16-bit integers.  Use regular integer
852      width for everything.
853
854    - [OPTIMIZATION] Cache size of sent packet.
855
856    - [OPTIMIZATION] Keep track of the largest ACKed in packet_out
857
858      Instead of parsing our own ACK frames when packet has been acked,
859      use the value saved in the packet_out structure when the ACK frame
860      was generated.
861
862    - [OPTIMIZATION] Take RTT sampling conditional out of ACK loop
863
864    - [OPTIMIZATION] ACK processing: only call clock_gettime() if needed
865
866    - [OPTIMIZATION] Several code-level optimizations to ACK processing.
867
868    - Fix: http_client: fix -I flag; switch assert() to abort()
869
8702018-02-26
871    - [API Change] lsquic_engine_connect() returns pointer to the connection
872      object.
873    - [API Change] Add lsquic_conn_get_engine() to get engine object from
874      connection object.
875    - [API Change] Add lsquic_conn_status() to query connection status.
876    - [API Change] Add add lsquic_conn_set_ctx().
877    - [API Change] Add new timestamp format, e.g. 2017-03-21 13:43:46.671345
878    - [OPTIMIZATION] Process handshake STREAM frames as soon as packet
879      arrives.
880    - [OPTIMIZATION] Do not compile expensive send controller sanity check
881      by default.
882    - [OPTIMIZATION] Add fast path to gquic_be_gen_reg_pkt_header.
883    - [OPTIMIZATION] Only make squeeze function call if necessary.
884    - [OPTIMIZATION] Speed up Q039 ACK frame parsing.
885    - [OPTIMIZATION] Fit most used elements of packet_out into first 64 bytes.
886    - [OPTIMIZATION] Keep track of scheduled bytes instead of calculating.
887    - [OPTIMIZATION] Prefetch next unacked packet when processing ACK.
888    - [OPTIMIZATION] Leverage fact that ACK ranges and unacked list are.
889      ordered.
890    - [OPTIMIZATION] Reduce function pointer use for STREAM frame generation
891    - Fix: reset incoming streams that arrive after we send GOAWAY.
892    - Fix: delay client on_new_conn() call until connection is fully set up.
893    - Fixes to buffered packets logic: splitting, STREAM frame elision.
894    - Fix: do not dispatch on_write callback if no packets are available.
895    - Fix WINDOW_UPDATE send and resend logic.
896    - Fix STREAM frame extension code.
897    - Fix: Drop unflushed data when stream is reset.
898    - Switch to tracking CWND using bytes rather than packets.
899    - Fix TCP friendly adjustment in cubic.
900    - Fix: do not generate invalid STOP_WAITING frames during high packet
901      loss.
902    - Pacer fixes.
903
9042017-12-18
905
906    - Fix: better follow cubic curve after idle period
907    - Fix: add missing parts to outgoing packet splitting code
908    - Fix: compilation using gcc 4.8.4
909
9102017-10-31
911
912    - Add APIs.txt -- describes LSQUIC APIs
913
9142017-10-31
915
916    - [API Change] Sendfile-like functionality is gone.  The stream no
917      longer opens files and deals with file descriptors.  (Among other
918      things, this makes the code more portable.)  Three writing functions
919      are provided:
920
921        lsquic_stream_write
922        lsquic_stream_writev
923        lsquic_stream_writef    (NEW)
924
925      lsquic_stream_writef() is given an abstract reader that has function
926      pointers for size() and read() functions which the user can implement.
927      This is the most flexible way.  lsquic_stream_write() and
928      lsquic_stream_writev() are now both implemented as wrappers around
929      lsquic_stream_writef().
930
931    - [OPTIMIZATION] When writing to stream, be it within or without the
932      on_write() callback, place data directly into packet buffer,
933      bypassing auxiliary data structures.  This reduces amount of memory
934      required, for the amount of data that can be written is limited
935      by the congestion window.
936
937      To support writes outside the on_write() callback, we keep N
938      outgoing packet buffers per connection which can be written to
939      by any stream.  One half of these are reserved for the highest
940      priority stream(s), the other half for all other streams.  This way,
941      low-priority streams cannot write instead of high-priority streams
942      and, on the other hand, low-priority streams get a chance to send
943      their packets out.
944
945      The algorithm is as follows:
946
947      - When user writes to stream outside of the callback:
948        - If this is the highest priority stream, place it onto the
949          reserved N/2 queue or fail.
950            (The actual size of this queue is dynamic -- MAX(N/2, CWND) --
951             rather than N/2, allowing high-priority streams to write as
952             much as can be sent.)
953        - If the stream is not the highest priority, try to place the
954          data onto the reserved N/2 queue or fail.
955      - When tick occurs *and* more packets can be scheduled:
956        - Transfer packets from the high N/2 queue to the scheduled
957          queue.
958        - If more scheduling is allowed:
959          - Call on_write callbacks for highest-priority streams,
960            placing resulting packets directly onto the scheduled queue.
961        - If more scheduling is allowed:
962          - Transfer packets from the low N/2 queue to the scheduled
963            queue.
964        - If more scheduling is allowed:
965          - Call on_write callbacks for non-highest-priority streams,
966            placing resulting packets directly onto the scheduled queue
967
968      The number N is currently 20, but it could be varied based on
969      resource usage.
970
971    - If stream is created due to incoming headers, make headers readable
972      from on_new.
973
974    - Outgoing packets are no longer marked non-writeable to prevent placing
975      more than one STREAM frame from the same stream into a single packet.
976      This property is maintained via code flow and an explicit check.
977      Packets for stream data are allocated using a special function.
978
979    - STREAM frame elision is cheaper, as we only perform it if a reset
980      stream has outgoing packets referencing it.
981
982    - lsquic_packet_out_t is smaller, as stream_rec elements are now
983      inside a union.
984
9852017-10-12
986
987    - Do not send RST_STREAM when stream is closed for reading
988    - Raise maximum header size from 4K to 64K
989    - Check header name and value lengths against maximum imposed by HPACK
990    - Fix NULL dereference in stream flow controller
991
9922017-10-09
993
994    - Hide handshake implementation behind a set of function pointers
995    - Use monotonically increasing clock
996    - Make sure that retx delay is not larger than the max of 60 seconds
997
9982017-09-29
999
1000    - A few fixes to code and README
1001
10022017-09-28
1003
1004    - Add support for Q041; drop support for Q040
1005
10062017-09-27
1007
1008    - Fix CMakeLists.txt: BoringSSL include and lib was mixed up
1009
10102017-09-26
1011
1012    - Add support for Mac OS
1013    - Add support for Raspberry Pi
1014    - Fix BoringSSL compilation: include <openssl/hmac.h> explicitly
1015
10162017-09-22
1017
1018    - Initial release
1019