chore(protocol): import the shared protocol type family

[p0mvn]

Summary

• import zcash_protocol 0.10.5, zcash_address 0.13.0, zcash_transparent 0.10.0, and zip321 0.9.0 from commit 97aefdc3
• publish them as zakura-protocol, zakura-address, zakura-transparent, and zakura-zip321 while preserving their upstream library targets and APIs
• rewire zakura-keys, zakura-primitives, and ZIP 321 through the local packages so the workspace resolves one public-type family
• add workspace, CI, maintenance, changelog, and cargo-vet integration for the imported crates

Motivation

Zakura and its wallet libraries need a Zakura-owned protocol stack for NU7. Importing every crate that exposes address and protocol types closes the public-type dependency graph: values such as NetworkType, ZcashAddress, and Zatoshis no longer cross between distinct registry and workspace copies.

Tests

• cargo check --locked -p zakura-address -p zakura-transparent -p zakura-protocol -p zakura-primitives -p zakura-keys --all-targets --all-features
• cargo test --locked -p zakura-address -p zakura-transparent --all-features — 55 unit tests and 9 doctests passed
• cargo test -p zakura-zip321 --all-features — 16 unit tests passed
• cargo +1.91.0 check --locked -p zakura-address -p zakura-transparent -p zakura-protocol --all-targets --no-default-features
• cargo fmt --all -- --check
• ./scripts/changelog.py check
• cargo deny --locked --workspace check bans licenses sources
• cargo vet check --store-path ./supply-chain/ --locked --no-registry-suggestions
• git diff --check
• Cargo metadata contains only the Zakura address, protocol, transparent, and ZIP 321 packages, with no upstream copies

actionlint reports the existing SC2086 finding in the no-std build command; the same finding is present on main.

AI assistance

Codex assisted with the workspace adaptation and validation.

[v12-auditor]

Note

Complete: Audit complete. V12 found 12 issues worth reviewing.

Open the full results here.

Finding	Severity	Details
F-283635	🟡 Medium	Zero denomination floor loops forever largest_one_two_five is public and accepts an unconstrained u64 floor. Calling it with floor == 0 bypasses the hi < floor return, initializes pow to zero, and repeatedly observes that 0 * 10 is still at most any unsigned hi. The loop body leaves pow unchanged at zero, so the function never returns. Any downstream migration-sizing path that forwards an unvalidated floor can therefore hang indefinitely.
F-283641	🔵 Low	Saturated expiries fail canonical validation canonical_expiry adds the window using saturating BlockHeight arithmetic, but is_canonical_expiry_value recognizes only exact modulus multiples. Even with the default ZIP-318 parameters, references near u32::MAX generate a saturated expiry of u32::MAX, which the reference-based method accepts. The value-only method rejects that same generated value because u32::MAX is not divisible by 34,560. The two public canonicality forms therefore disagree in the saturation region despite internally consistent parameters.
F-283642	🔵 Low	Incomplete evidence becomes cacheable conformance classify rejects anchor_on_grid and fee_is_canonical only when they are explicitly Some(false); missing values are dropped from the predicate. Otherwise canonical evidence with those checks omitted therefore returns Conforms, and to_code persists the same code used when both checks were positively verified. The evidence contract simultaneously permits None to become a value as wallet data arrives, so resolving an omitted field to Some(false) changes the result from Conforms to Nonconforming. This contradicts the documented monotonicity and cacheability guarantee for classification decisions.
F-283643	🔵 Low	Contradictory evidence passes classification The public evidence builder permits mutually inconsistent combinations, and each shape-specific classifier ignores fields it does not consume. Preparation evidence can declare zero destination actions while also declaring a destination value and still return Conforms(Preparation), even though zero actions is documented as no destination bundle. Conversely, a crossing can declare source_is_send_to_self = Some(false) and still return Conforms(Transfer) because that branch ignores the explicit negative fact. Affirmative classifications can therefore be derived from evidence that cannot consistently describe the claimed transaction shape.
F-283644	🔵 Low	Preparation transactions wrongly require grid anchors The module states that preparation transactions need temporal spacing but no anchor bucketing, while only pool-crossing transfers anchor to retained boundaries. classify nevertheless applies anchor_on_grid == Some(false) before selecting the preparation or transfer branch. An otherwise canonical preparation with a truthful off-grid anchor is therefore returned as Nonconforming. Even anchor evidence alone can definitively rule out every ZIP-318 shape despite being irrelevant to preparations.
F-283646	🔵 Low	Unknown codes are lost on re-encoding from_code maps every unrecognized persisted integer to Unknown, while Unknown::to_code() always emits zero. An older consumer reading a valid classification code introduced by a newer append-only producer therefore loses the original integer. A normal decode, update, and re-encode sequence transforms code 4 or any later code into the default unclassified value 0. Invalid or corrupted integers are also indistinguishable from rows that were genuinely never classified.
F-283650	🔵 Low	Partial schedules yield contradictory branch ranges LocalNetwork accepts partial schedules and defines None as an upgrade that is not activated. With Overwinter at height 1, Sapling absent, and Blossom at height 10, BranchId::for_height(..., 10) selects Blossom because it scans all configured upgrades in reverse order. BranchId::Overwinter.height_bounds(), however, uses only Sapling as its upper bound and returns an unbounded range starting at 1, incorrectly including height 10. An all-None schedule similarly selects Sprout at every height while Sprout.height_bounds() returns None, documented as meaning Sprout is never in effect.
F-283651	🔵 Low	Rounded-up height is not a boundary boundary_at_or_above promises the least interval boundary not below its input, but rounds with saturating addition. For AnchorBucketInterval::ZIP_318 and height = u32::MAX, rounding up mathematically requires a height 33 blocks beyond the representable range. The method saturates to u32::MAX, even though u32::MAX % 144 == 111 and is_boundary therefore returns false for the result. The two APIs give contradictory answers about the same returned checkpoint.
F-283652	🔵 Low	Truncated memos are silently canonicalized Memo::from_bytes documents rejection of inputs that are not 512 bytes, but it delegates to MemoBytes::from_bytes, which accepts every length up to 512 and fills the missing suffix with zeroes. A one-byte [0xFF] input is expanded into an arbitrary memo with 511 zero payload bytes; [0xF6] becomes an empty memo; and [0xF7] becomes a zero-filled future-format memo. Calling encode then emits a complete 512-byte value that was never present in the input. The parser consequently canonicalizes truncated wire data instead of rejecting it.
F-283653	🔵 Low	Nonzero text padding is accepted Canonical text memos contain UTF-8 followed by NUL padding, so every byte after the first padding NUL must be zero. MemoBytes::as_slice instead searches backward for the last nonzero byte and removes only the final run of zeroes. An exact 512-byte input containing A\0B followed by zeroes is therefore passed to str::from_utf8 as A\0B, which is valid UTF-8, and becomes Memo::Text. Encoding the text copies the embedded NUL and later nonzero byte back verbatim, preserving a noncanonical memo.
F-283654	🔵 Low	Malformed empty memos become future formats The canonical 0xF6 memo marker is valid only when followed by 511 zero bytes. If any suffix byte is nonzero, the guarded empty branch fails and the wildcard branch accepts the buffer as Memo::Future. Future serialization preserves the raw bytes unchanged. Known malformed instances of the empty format are therefore conflated with reserved future formats, even though valid future tags are separate values.
F-283655	🔵 Low	Raw memo accessor truncates payloads Memo::Arbitrary defines all 511 bytes after its 0xFF format byte as payload, including zero-valued suffix bytes. MemoBytes::as_slice nevertheless removes every trailing zero from the entire representation without checking the format. Encoding Memo::Arbitrary([0; 511]) and then calling as_slice() returns only [0xFF], not the complete arbitrary memo representation. Reserved future formats with meaningful trailing zero bytes are affected for the same reason.

And 10 more auto-invalidated findings.

Analyzed 11 files, diff c6cb513...4e1376a.

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	cargo/​base64@​0.22.1
	cargo/​nom@​7.1.3
	cargo/​percent-encoding@​2.3.2

View full report

[ebfull]

Suggested change

	next: Some(self.0.start),
	next: (self.0.start < self.0.end).then_some(self.0.start)),