Perf: eliminate hot-loop overhead in seed stretching, base58 validation, and XOR operations

[Copilot]

• Merge latest origin/master (f25290f) into branch — includes 3 new master commits:
  • Fix OpenCL kernels for Apple Silicon GPUs (Fix OpenCL kernels for Apple Silicon GPUs #692)
  • Remove hashes and trim requirements-full.txt (Remove hashes and trim requirements-full.txt to directly-imported packages only #693)
  • Remove Python 3.9 from CI, update 3.14, fix failing CI jobs (Remove Python 3.9 from CI, update 3.14 to official release, fix failing CI jobs #696)
• No merge conflicts — clean merge
• All 670 tests pass after merge

---

💬 Send tasks to Copilot coding agent from Slack and Teams to turn conversations into code. Copilot posts an update in your thread when it's finished.

[Copilot]

Pull request overview

This PR focuses on micro-optimizations in btcrecover’s hottest paths (seed stretching/derivation loops, base58 “looks valid” checks, and script parsing), aiming to reduce Python-level overhead while keeping behavior identical.

Changes:

• Refactors several tight loops in btcrseed.py (seed stretching, address generation inner loops, HMAC construction) to cache locals and avoid repeated conversions.
• Speeds up base58-character validation in btcrpass.py using a precomputed base58 byte set and all(...).
• Reworks XOR and authentication checks in aezeed.py and caches common script-prefix constants in addressset.py for faster parsing.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

File	Description
btcrecover/btcrseed.py	Reduces overhead in mnemonic/seed handling and inner derivation loops; refactors byte/word conversion helpers.
btcrecover/btcrpass.py	Replaces base58 range checks with set-membership validation.
btcrecover/aezeed.py	Replaces byte-wise XOR loops with integer-based XOR; changes AEZ authentication verification logic.
btcrecover/addressset.py	Caches common script-prefix/suffix byte patterns used in block parsing.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

bytes_to_words() now assumes byte_array[i:i+4] is a bytes-like object accepted by int.from_bytes(), but there are call sites that pass a plain list of ints (e.g., decode_v3() builds [b1[2], b1[3], 0, 0]). That will raise TypeError at runtime. Consider normalizing the input inside bytes_to_words() (e.g., wrapping with bytes(...)/bytearray(...) when needed) or updating call sites to pass a bytes-like object.

Suggested change

	return [int.from_bytes(byte_array[i:i+4], byteorder='big', signed=False)
	return [int.from_bytes(bytes(byte_array[i:i+4]), byteorder='big', signed=False)

[Copilot]

The authentication check in _aez_decrypt() was changed from a full-length XOR accumulation to x[:tau] != ciphertext[:tau]. Plain bytes comparison is not constant-time and can reintroduce a timing side-channel for tag verification. Use a constant-time comparison (e.g., hmac.compare_digest) to preserve the original security properties while still avoiding Python-level loops where possible.

[Copilot]

if any(x[-tau:]) short-circuits on the first non-zero byte, which can leak information via timing and differs from the prior constant-time style check. If this is part of the AEZ authentication/verification step, prefer a constant-time zero check (e.g., hmac.compare_digest(x[-tau:], b"\x00" * tau) or an XOR-accumulator over the slice).