refactor: have process_input delegate to process_multiple_input

[mxinden]

The Connection::process_input and Connection::process_multiple_input functions are identical, except that the latter handles multiple Datagrams.

To avoid any changes to one without updating the other, have process_input simply delegate to process_multiple_input.

Commit also does the equivalent change to neqo_http3::Http3Client.

---

I would be surprised if std::iter::once adds a performance hit. Let's see what the benchmarks say.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 93.12%. Comparing base (1dc8ea3) to head (a63c80d).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1792      +/-   ##
==========================================
+ Coverage   93.04%   93.12%   +0.07%     
==========================================
  Files         116      116              
  Lines       36098    36097       -1     
==========================================
+ Hits        33589    33614      +25     
+ Misses       2509     2483      -26     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[github-actions]

Benchmark results

Performance differences relative to a33fe60.

• coalesce_acked_from_zero 1+1 entries
  time: [197.94 ns 198.48 ns 199.04 ns]
  change: [+2.7004% +3.0993% +3.4984%] (p = 0.00 < 0.05)
  💔 Performance has regressed.

• coalesce_acked_from_zero 3+1 entries
  time: [238.54 ns 238.95 ns 239.43 ns]
  change: [+0.6338% +0.9377% +1.2685%] (p = 0.00 < 0.05)
  Change within noise threshold.

• coalesce_acked_from_zero 10+1 entries
  time: [238.59 ns 239.36 ns 240.27 ns]
  change: [+0.3417% +0.7690% +1.1552%] (p = 0.00 < 0.05)
  Change within noise threshold.

• coalesce_acked_from_zero 1000+1 entries
  time: [219.68 ns 219.85 ns 220.04 ns]
  change: [+0.2062% +1.1037% +1.9807%] (p = 0.01 < 0.05)
  Change within noise threshold.

• RxStreamOrderer::inbound_frame()
  time: [118.30 ms 118.38 ms 118.48 ms]
  change: [-1.6867% -1.5909% -1.4887%] (p = 0.00 < 0.05)
  💚 Performance has improved.

• transfer/Run multiple transfers with varying seeds
  time: [121.07 ms 121.33 ms 121.59 ms]
  thrpt: [32.899 MiB/s 32.968 MiB/s 33.039 MiB/s]
  change:
  time: [+1.3341% +1.6258% +1.9459%] (p = 0.00 < 0.05)
  thrpt: [-1.9087% -1.5998% -1.3165%]
  Change within noise threshold.

• transfer/Run multiple transfers with the same seed
  time: [122.09 ms 122.26 ms 122.43 ms]
  thrpt: [32.673 MiB/s 32.718 MiB/s 32.763 MiB/s]
  change:
  time: [+1.8113% +1.9985% +2.1829%] (p = 0.00 < 0.05)
  thrpt: [-2.1363% -1.9593% -1.7791%]
  Change within noise threshold.

• 1-conn/1-100mb-resp (aka. Download)/client
  time: [1.1008 s 1.1173 s 1.1391 s]
  thrpt: [87.787 MiB/s 89.504 MiB/s 90.846 MiB/s]
  change:
  time: [-3.3272% -1.7385% +0.1247%] (p = 0.09 > 0.05)
  thrpt: [-0.1246% +1.7692% +3.4417%]
  No change in performance detected.

• 1-conn/10_000-1b-seq-resp (aka. RPS)/client
  time: [388.16 ms 390.82 ms 393.51 ms]
  thrpt: [25.412 Kelem/s 25.587 Kelem/s 25.763 Kelem/s]
  change:
  time: [-0.9081% +0.1116% +1.0817%] (p = 0.82 > 0.05)
  thrpt: [-1.0702% -0.1114% +0.9164%]
  No change in performance detected.

• 100-seq-conn/1-1b-resp (aka. HPS)/client
  time: [3.3804 s 3.3833 s 3.3862 s]
  thrpt: [29.532 elem/s 29.557 elem/s 29.582 elem/s]
  change:
  time: [+0.4209% +0.5462% +0.6668%] (p = 0.00 < 0.05)
  thrpt: [-0.6624% -0.5432% -0.4191%]
  Change within noise threshold.

Client/server transfer results

Transfer of 134217728 bytes over loopback.

Client	Server	CC	Pacing	Mean [ms]	Min [ms]	Max [ms]	Relative
msquic	msquic			824.0 ± 243.1	519.4	1387.3	1.00
neqo	msquic	reno	on	2171.7 ± 249.5	1908.4	2545.3	1.00
neqo	msquic	reno		2020.4 ± 214.2	1868.1	2433.5	1.00
neqo	msquic	cubic	on	1913.8 ± 48.4	1856.8	2005.3	1.00
neqo	msquic	cubic		2107.8 ± 290.2	1895.2	2877.7	1.00
msquic	neqo	reno	on	3341.0 ± 171.6	3247.4	3781.4	1.00
msquic	neqo	reno		3436.0 ± 240.2	3176.5	3783.9	1.00
msquic	neqo	cubic	on	3264.3 ± 74.4	3204.0	3428.2	1.00
msquic	neqo	cubic		3241.7 ± 115.5	3114.9	3552.4	1.00
neqo	neqo	reno	on	2998.2 ± 212.5	2819.5	3567.0	1.00
neqo	neqo	reno		2853.3 ± 57.9	2766.1	2928.6	1.00
neqo	neqo	cubic	on	3175.7 ± 192.2	2956.3	3647.7	1.00
neqo	neqo	cubic		3166.0 ± 169.0	3074.9	3641.4	1.00

⬇️ Download logs

[martinthomson]

I'm fairly sure that std::iter::once optimizes out well enough that this won't matter. We have far too much noise in benchmarks to notice, either way.

At this layer of the code, simply avoid duplication is worth the extra overhead.