Don't go wantonly creating reference cycles

[alex]

No description provided.

[glyph]

Build failures are occurring on non-pypy builders and succeeding on pypy because cryptography distributes wheels for the non-pypy pythons being tested here, and the binary wheels include OpenSSL 1.1.1. By contrast the host OS on travis uses OpenSSL 1.0.2 (and therefore pypy's cryptography gets built against that version). These failures are almost certainly unrelated to this change, and are instead related to new behaviors in 1.1.1.

[markrwilliams]

The failures are in fact related to OpenSSL 1.1.1.

Wireshark shows that the client uses NPN, not ALPN, and OpenSSL's TLS 1.3 implementation doesn't support NPN. As a result no next protocol is negotiated.

[codecov-io]

Codecov Report

Merging #23 into master will decrease coverage by 0.5%.
The diff coverage is 0%.

@@          Coverage Diff           @@
##           master   #23     +/-   ##
======================================
- Coverage    95.5%   95%   -0.5%     
======================================
  Files           6     6             
  Lines         400   400             
  Branches       28    28             
======================================
- Hits          382   380      -2     
- Misses         10    12      +2     
  Partials        8     8

Impacted Files	Coverage Δ
txsni/snimap.py	89% <0%> (-2%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 5feb4a9...c86b4a1. Read the comment docs.