[AryanRG]

Summary

Preserve the two fields of HeaderTuple and NeverIndexedHeaderTuple when they are copied or serialized. The inherited tuple reconstruction hook passes the entire pair as one argument to these classes' variadic constructor; copying or pickling therefore nests the pair, and re-encoding a copied decoded header can raise IndexError.

__getnewargs__() now supplies the two fields as separate constructor arguments. The change also adds copy, deepcopy, pickle-protocol, and decode–copy–encode regressions, plus a changelog entry.

Validation

• The new regression selection has 14 failures and 4 passing controls on the base; all 18 pass with this patch.
• Full local suite: 520 passed each on Python 3.11, 3.13, and 3.14, with 100% reported statement/branch coverage.
• Ruff, strict mypy, and whitespace checks pass. Other platforms, hosted CI, packaging, and documentation builds remain unverified.
• Existing malformed pickles are not repaired.

AI assistance was used to prepare the implementation, tests, and this description.

[agammann]

__getnewargs__ supplies the tuple’s two fields as separate arguments to HeaderTuple.__new__, so copy and pickle reconstruct both values instead of nesting them as one item. The tests cover both concrete header classes, shallow and deep copy, every pickle protocol, and re-encoding a copied decoder result. I reviewed the patch and tests; I did not run them locally.