Sent packets vec

[martinthomson]

This is NOT faster than main and I don't quite understand why.

I'm putting this up so that I can get benchmarks (my machine doesn't produce anything useful).

If this line of investigation doesn't work out, I might try to salvage some of the cleanup that I did here.

[larseggert]

I don't understand why CI isn't running here. Force push to unstick it?

[codecov-commenter]

Codecov Report

Attention: 14 lines in your changes are missing coverage. Please review.

Comparison is base (5074583) 92.93% compared to head (e4e45be) 92.93%.

Files	Patch %	Lines
neqo-transport/src/cc/classic_cc.rs	78.57%	6 Missing ⚠️
neqo-transport/src/recovery/mod.rs	92.15%	4 Missing ⚠️
neqo-transport/src/recovery/sent.rs	98.84%	3 Missing ⚠️
neqo-transport/src/path.rs	50.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1662      +/-   ##
==========================================
- Coverage   92.93%   92.93%   -0.01%     
==========================================
  Files         119      121       +2     
  Lines       37260    37456     +196     
==========================================
+ Hits        34629    34811     +182     
- Misses       2631     2645      +14     

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

[larseggert]

If I had to venture a guess why it's slower, it'd be because a BTreeMap holds a bunch of items contiguous in memory, whereas a VecDeque has each item in a different location in memory. So the BTreeMap should traverse faster.

From https://doc.rust-lang.org/std/collections/struct.VecDeque.html

Since VecDeque is a ring buffer, its elements are not necessarily contiguous in memory. If you want to access the elements as a single slice, such as for efficient sorting

Maybe a call to make_contiguous in a strategic place could help?

[larseggert]

Argh! I followed the steps GitHub showed to merge main into this branch. Not sure what it did.

[larseggert]

Reverted and re-extracted the rebased PR into #1761. Sorry for the mess.